I've written quite a bit on the AI bubble over the past six or seven months. The situation for the likely consolidation of assets (and power) is continuing to play out as I detail below. Check out my previous article on the structure of the AI bubble:

How to prepare for a recession if you have a lot of money

Jeff Bezos announced yesterday that he is in early discussions to raise $100 billion to acquire legacy manufacturers and automate them with AI. Three weeks ago, Blackstone announced plans to launch a publicly traded company to acquire operating data centers, targeting tens of billions from sovereign wealth funds. These are not coincidental announcements. They are the most visible markers of a broader trend in which the world’s most capitalized investment firms are building acquisition vehicles now, before stress in the AI sector forces smaller and more leveraged players to sell. Whether or not a recession materializes, the architecture for a sweeping consolidation of AI infrastructure is being assembled now.

The Blackstone vehicle, as reported by Bloomberg in late February 2026, is designed to acquire already-built and fully leased data centers, explicitly bypassing development-stage projects or raw land. It is soliciting sovereign wealth funds and other institutions for initial investments, seeking to raise tens of billions of dollars, with the eventual aim of broadening to retail investors through public markets. While details are thin, the structure—a publicly traded acquisition company competing directly with REITs like Digital Realty and Equinix—tells us something at least. Blackstone isn’t simply trying to own more data centers for the heck of it. It’s trying to set the price discovery framework for the sector. By creating a benchmark valuation entity backed by sovereign capital, Blackstone positions itself as the de facto market-maker for distressed and opportunistic data center transactions whenever they materialize. The firm already reports a $55 billion portfolio and a $70 billion prospective pipeline in data centers, giving it unmatched informational asymmetry in any transaction it chooses to pursue.

The Bezos announcement, which broke today, is somewhat analogous but aimed at a different segment of the economy (and potentially more worrying). Bezos is reportedly in early discussions to raise $100 billion for a fund described in investor documents as a “manufacturing transformation vehicle,” targeting companies in semiconductors, defense, and aerospace. The fund is designed to acquire legacy manufacturers and inject them with AI systems developed through Project Prometheus, Bezos’s AI startup focused on industrial automation. One backer quoted by the Wall Street Journal called the current moment “a huge buying opportunity” as legacy manufacturers struggle to keep pace with technological shifts. Bezos himself has called AI infrastructure a “good” kind of bubble, comparing it to past infrastructure bubbles in which excess capital produced failures and investor losses that nonetheless left behind infrastructure reshaping entire industries—a framing that doubles as an implicit prediction of the distress to come and an announcement of his intention to harvest it.

Neither of these vehicles is operating in isolation. KKR completed its largest Asia Pacific infrastructure deal ever in early 2026, acquiring 82 percent of data center operator ST Telemedia Global Data Centres for $5.1 billion, valuing the company at $10.6 billion. Brookfield closed its second Global Transition Fund at $20 billion in late 2025, with anchor LPs including Norway’s sovereign wealth fund, CalPERS, Singapore’s GIC, and Temasek. BlackRock led a consortium in a $33.4 billion acquisition of AES Corp specifically to control the electricity supply for AI data centers in the Mid-Atlantic. The full scope of 2025 data center mergers and acquisitions reached $69 billion across 113 transactions, a record. What is being assembled, across these deals and vehicles, is an infrastructure ownership class—a layer of capitalized incumbents whose holdings span compute, power, and physical production capacity, organized precisely to absorb the assets of firms that cannot sustain their current debt loads through a market correction.

Bezos sees the same opportunity as past infrastructure barons

The closest historical precedent for this configuration may very well be the fiber glut of the dot-com bubble, but it has just as much in common with the late nineteenth-century cycle of infrastructural overexpansion, financial distress, and consolidation under the direction of large banking houses. By the 1880s, American rail lines had been built far in excess of demand, financed through layered debt structures and optimistic projections of traffic and settlement. When the Panic of 1893 struck, those assumptions collapsed. What followed was not simply a wave of bankruptcies but a systematic reorganization of ownership. The failed railroads were reorganized under terms that wiped out existing equity and transferred control to creditors and financiers.¹ The overbuilt network didn’t disappear. It was consolidated, rationalized, and brought under the control of actors with sufficient capital to impose a new order on what had previously been a fragmented competitive field.

This process became associated with J.P. Morgan and the practice contemporaries called “Morganization”—the reorganization and merger of distressed firms under centralized financial control, often through interlocking directorates and coordinated lending. The Pujo Committee’s investigation into the “Money Trust” captured how a few men and their associates had come to control practically all of the most important industries through their command over credit and capital flows.² Samuel Untermyer, counsel to the investigation, described the mechanism as the “welding together” of major banks, railroads, and industrial firms into a system in which nominally separate enterprises were effectively governed as a single coordinated structure.³ Consolidation wasn’t only a matter of buying cheap during downturns. It was a reorganization of entire sectors around financial command—what Hilferding and Lenin would later theorize as the structural merger of bank and industrial capital into a dominant organizational form,⁴ defined less by the scale of individual firms than by the integration of ownership and control across them.

The analogy to the present is not exact, but it is informative. Periods of rapid infrastructural expansion financed through fragile capital structures tend to produce not only growth but the conditions for subsequent consolidation. When stress emerges—through recession, credit tightening, or sector-specific overcapacity—actors with deep balance sheets are positioned to reorganize ownership, absorb fragmented assets, and establish new norms of valuation and control. Infrastructure doesn’t dissipate like the speculative visions once behind them—it is absorbed into a more concentrated and financially mediated system. Bezos explicitly invoked this logic in describing the current AI investment wave as analogous to prior infrastructure booms in which excess capital produced failures that nonetheless left behind durable assets. Bezos is right about the pattern, even if his framing obscures who benefits from it.

The stars align with sovereign capital as less capitalized firms grow vulnerable

The preconditions—and signs—for that consolidation are visible and accumulating. The AI infrastructure buildout has been financed in significant part through high-yield debt taken on by firms without the balance sheet depth to weather a demand shortfall. Oracle now holds over $100 billion in debt after issuing $18 billion in bonds for AI infrastructure; CoreWeave tapped the high-yield bond market for $3.75 billion across two transactions, borrowing at around 9 percent each time. These are not anomalies. AI-related companies and projects tapped debt markets for at least $200 billion in 2025 alone, with projections in the hundreds of billions for 2026. As JPMorgan projects data center securitization reaching $30 to $40 billion annually in commercial mortgage-backed securities and asset-backed securities markets through 2027, the AI sector is acquiring the debt structure of a real estate cycle—highly sensitive to any deceleration in lease rates, revenue projections, or broader credit conditions.

Even before the U.S.-Israel war on Iran has sent oil prices skyrocketing, the macroeconomic environment makes all of this increasingly plausible. In May 2025, J.P. Morgan assigned a 40 percent probability to a U.S. recession in 2026—and that’s after moderating its estimate in response to the recent U.S.-China tariff détente. In November 2025, Morgan Stanley’s outlook identified lagged effects from monetary policy, tariffs, and immigration restrictions as the primary recession triggers, with real GDP growth potentially turning negative in the first half of 2026. In January 2026, Stanford’s economic policy researchers flagged the stagflation risk as genuine—a weakening job market argues for rate cuts while tariff-induced price pressures argue for restraint, limiting the Fed’s ability to provide cushion. Deloitte put the vulnerability most directly—with so much consumer spending and business investment reliant on AI-related stock prices and anticipated AI returns, a drop in AI-related spending could be sufficient to push the economy into recession on its own. In the neocloud segment—the GPU-as-a-service operators and specialized compute providers central to the AI buildout—S&P has flagged consolidation as likely, with smaller and less-specialized players at greatest risk of absorption. The firms building the acquisition vehicles described in the opening of this piece will not be among those absorbed. They will be the buyers.

The geopolitical economy of this moment has a specific geography that connects the domestic consolidation story to a broader circuit of capital flows. Both Blackstone and Bezos have made sovereign wealth funds the first call in their respective fundraising efforts. Bezos recently traveled to the Middle East to discuss the manufacturing fund with sovereign wealth representatives, and to Singapore as part of the same effort. Blackstone is approaching sovereign wealth funds as the anchor investors for its data center acquisition vehicle before broadening to retail markets. This is the contemporary form of petrodollar intermediation which I’ve examined in a previous article: Gulf capital, accumulated through hydrocarbon rents and now held in sovereign wealth funds, is being recycled into AI infrastructure through the vehicles of U.S. alternative asset managers who supply the deal flow, the operational expertise, and the market-making capacity that sovereign funds cannot easily replicate on their own. This mirrors the 1970s dynamic in which petrodollar surpluses were intermediated through New York and London banks into sovereign loans and corporate debt—except that the asset class is now physical AI infrastructure rather than developing-country balance of payments, and the intermediaries are Blackstone and KKR rather than Citibank and Chase. All of this may depend on what happens with the war of course.

A new ownership layer is being constructed over the physical substrate of AI

Compute infrastructure, power infrastructure, and industrial manufacturing capacity are being brought under common ownership by a small number of extremely well-capitalized firms using sovereign and institutional capital as their anchor. In a market “correction,” they could acquire stressed assets on the cheap. In continued expansion, they capture a disproportionate share of the asset appreciation. Whether or not the anticipated recession materializes at scale, the vehicles being built now may very well reshape the ownership structure of AI infrastructure regardless. The asymmetry is the point—and the historical record suggests it is not accidental.

References

1. Daggett, Stuart. 1908. Railroad Reorganization, Volume IV. New York: The Riverside Press. eBook available here: https://www.gutenberg.org/files/55397/55397-h/55397-h.htm.

2. Noyes, Alexander D. 1913. “The Money Trust.” The Atlantic, May 1, 1913. Available at: https://www.theatlantic.com/magazine/archive/1913/05/the-money-trust/645558/.

3. Untermyer, Samuel. 1914. “Reasons and Remedies for Our Business Troubles: An Address Delivered before the Commercial Club and the Pittsburgh Industrial Development Commission.” Untermyer, Samuel (1911-1928), Entry 168, Box 15, Folder 3. Available at: https://fraser.stlouisfed.org/archival-collection/committee-history-federal-reserve-system-1342/untermyer-samuel-1911-1928-459351.

4. Lenin, Vladimir Ilʹich. 1999 [1917]. Imperialism: The Highest Stage of Capitalism. Resistance Books; Hilferding, Rudolf. 1985. Finance Capital. Translated by Tom Bottomore. London, England: Routledge.
   

When the smoke cleared from the Iranian retaliatory strikes on Amazon data centers in early March 2026, Bloomberg noted that digital infrastructure had become a “casualty of war.” But that framing is far too passive. These infrastructures—now openly and contentiously integrated into the unclear military ambitions of firms like OpenAI and Anthropic—aren’t just caught in the crossfire, they are the physical substrate of techno-statecraft. As I’ve argued previously, your cloud account isn’t just hosting data—it’s funding a new era of digitized warfare. To understand why these high-stakes targets sit in the line of fire, we must look past the hardware and into the broader geopolitical architecture of the AI boom where many of the Gulf states aren’t just a “deep pocket” for Silicon Valley, but foundational partners in an accumulation frontier that bridges our carbon past with a silicon future.

In Part I of this series, I explored how inflated AI load forecasts have been leveraged to justify a massive, financialized overbuild of data centers and energy systems. In this article, we take a look at the specific capital that makes this expansion possible. What we are seeing resembles a new phase of petrodollar recycling, in which Gulf oil surpluses are being redirected through sovereign wealth funds and infrastructure finance into the global AI buildout. In the 1970s, oil surpluses stabilized the dollar order through commercial bank lending. Today, those same rents are being channeled through Gulf sovereign wealth funds directly into the “compute + power” asset stack. By trading carbon rents for “safe,” AI-linked cash flows, Gulf states are securing a structural position within the U.S.-led technological order.

The AI economy is not a departure from the fossil fuel era—it is its most ambitious extension. The circuits of accumulation that defined the twentieth century are not being replaced; they are being recycled, acquiring new institutional forms while preserving the same geopolitical hierarchies and geographic vulnerabilities.

Oil surpluses once built the dollar order through US-centered finance

When oil prices quadrupled after the 1973 embargo, the global financial system confronted what economists of the era called a “transfer problem.”¹ Oil-exporting states accumulated enormous dollar surpluses while importing nations ran mounting deficits. The more systemic question was how “the oil countries’ surplus could in principle be recycled to deficit countries by private financial intermediation” without triggering a collapse in trade and finance.² The solution that emerged was petrodollar recycling, and the institutional architecture through which it operated reveals something important about how global capitalism under US hegemony reproduces itself across technological eras.

The mechanism operated through two principal channels. First, OPEC’s rapidly accumulating dollar surpluses were deposited in major commercial banks, particularly in the offshore Eurocurrency market. These funds were then intermediated outward as international bank lending, including syndicated loans to governments and corporations in deficit countries. IMF documentation from the period observed that the buildup of foreign-exchange claims by oil exporters was largely held in the form of dollar assets, including “identified Euro-dollar deposits” and claims on U.S. markets, which became an important source of financing for oil-importing economies.³ Alongside private banking, multilateral institutions expanded official financing mechanisms. The IMF introduced temporary oil facilities beginning in 1974, enabling member states to borrow to offset the sudden deterioration in their external accounts following the oil price increase. Between 1974 and 1976, dozens of countries drew on these facilities to manage balance-of-payments pressures created by higher import costs.⁴

What the historical record makes clear is that petrodollar recycling was never a neutral plumbing operation. It was institutionally structured around US- and UK-centered finance, anchored in the Eurodollar market as an offshore extension of dollar hegemony. In effect, the recycling circuit was governed by US banks, supported by the US government, and shaped by US monetary policy incentives.⁵ The effect was not simply to prevent a payments crisis but to reinforce the dollar-centered financial order by routing the world’s most important commodity surplus through its infrastructure.

That reinforcement carried costs that were not immediately apparent. Bank-intermediated recycling sustained global demand and payments in the short term, but it also created the conditions for the debt crises that swept Latin America and much of the Global South in the early 1980s, when US interest rates rose sharply and dollar liquidity contracted. The political economy of petrodollar recycling thus bound oil-export rents, sovereign borrowing, and structural vulnerability into a single circuit whose fragilities became visible only after a decade of apparent stability.

The logic described in this history has not disappeared. The institutional form has changed substantially. The underlying dynamic, oil-export surpluses channeled through US-centered infrastructure to stabilize and accelerate a new accumulation regime, is being replicated today in a context that most observers have not yet named with adequate precision. The AI boom, widely narrated as a story of Silicon Valley innovation and venture capital, has a different material foundation than most accounts acknowledge.

Gulf sovereign wealth funds are financing the AI buildout directly

Indeed, the financial architecture of the contemporary artificial intelligence boom has attracted considerable commentary, but much of it has focused on Silicon Valley firms, institutional venture capital, and a handful of hyperscale technology companies (my own commentary included). This misses the structural role of Gulf sovereign wealth funds within a renewed petrodollar recycling mechanism. In the 1970s, oil surpluses were intermediated primarily through commercial bank balance sheets and syndicated lending. Today a significant share flows through sovereign wealth funds, state-linked AI investment vehicles, and infrastructure partnerships that structurally resemble private equity and infrastructure finance far more than traditional banking. Yet, this follows the same logic—oil-export rents are channeled through US-centered infrastructure to stabilize and extend a new accumulation frontier.

Gulf sovereign wealth funds collectively control more than $3 trillion in assets, most of it derived from decades of hydrocarbon export revenues. The Saudi Public Investment Fund manages approximately $900 billion. The Abu Dhabi Investment Authority, Mubadala Investment Company, and ADQ together hold over $1.5 trillion. The Qatar Investment Authority manages roughly $475 billion. These are strategic capital pools pursuing deliberate technology investment mandates, and AI has become their primary target.

In May 2025, President Trump’s Gulf trip secured over $2 trillion in “deals,” anchored by a claimed $600 billion Saudi investment commitment, a $1.2 trillion “economic exchange” framework with Qatar, and $200 billion in U.S.–UAE commercial deals—linking Gulf capital commitments to the expansion of AI infrastructure and energy. The White House reported that Saudi firm DataVolt was “moving forward” with plans to invest $20 billion in AI data centers and energy infrastructure in the United States, while also listing an additional $80 billion in tech commitments spanning Google, Oracle, Salesforce, AMD, and Uber in both countries. On the UAE side, the White House ties the $200 billion slate to a previously announced 10-year, $1.4 trillion UAE investment framework aimed at AI infrastructure, semiconductors, and energy, and said a same-day U.S.–UAE AI agreement includes UAE commitments to “invest in, build, or finance” U.S. data centers comparable in scale to those in the UAE, alongside tighter alignment on national-security controls for U.S.-origin technology.

In May 2025, the White House framed President Trump’s Gulf trip as securing more than $2 trillion in investment frameworks and commercial deals, including a claimed $600 billion Saudi commitment, a $1.2 trillion U.S.–Qatar “economic exchange” framework, and $200 billion in U.S.–UAE commercial deals, with repeated emphasis on AI, data centers, and energy infrastructure. The Saudi fact sheet highlighted DataVolt’s stated plan to invest $20 billion in U.S. AI data centers and energy infrastructure, alongside a separate headline figure of $80 billion in cross-border technology commitments involving firms such as Google, Oracle, Salesforce, AMD, and Uber. Saudi Arabia also created Humain, a Public Investment Fund–owned AI vehicle intended to build an integrated AI stack spanning data centers, cloud, and models. In 2026, Humain invested $3 billion in xAI and publicly linked the partnership to building roughly 500 MW of AI data center capacity, a load comparable to a medium-sized U.S. city. It also signed a $10 billion agreement with AMD and partnered with AWS on a joint $5 billion AI zone in the kingdom.

On the UAE side, the $200 billion deal package was presented as part of a wider 10-year, $1.4 trillion UAE investment framework aimed at AI infrastructure, semiconductors, and energy, and the administration said a same-day U.S.–UAE AI agreement included UAE commitments to invest in, build, or finance large U.S. data centers while tightening controls around U.S.-origin technology. Abu Dhabi’s MGX has emerged as a key conduit for sovereign capital into both AI firms and the U.S. compute real-estate layer. MGX participated in major OpenAI-related secondary financing, and it co-led a consortium acquisition of Aligned Data Centers (reported at roughly $40 billion), placing significant hyperscale capacity under a Gulf-linked ownership structure alongside U.S. partners including Microsoft and Nvidia and financial co-investors tied to BlackRock’s infrastructure platform and other sovereign funds. In parallel, MGX joined Microsoft and BlackRock (via Global Infrastructure Partners) in the AI Infrastructure Partnership, which targets roughly $30 billion in equity and up to $100 billion with debt to finance AI data centers and the power infrastructure that supplies them.

In sum, Gulf hydrocarbon surpluses are being recycled not only into AI companies, but into the coupled “compute + power” asset stack that makes the AI boom buildable, financeable, and scalable, with sovereign wealth increasingly underwriting both hyperscale data center platforms and the energy systems that feed them (see table above). Because Saudi Arabia and the UAE anticipate a long-run plateau in oil demand, they are positioning AI and digital infrastructure as replacement growth sectors within diversification agendas like Vision 2030. Investment in frontier AI firms provides strategic ownership stakes in the systems, chips, and research ecosystems that will define the next technological era. And AI development now requires capital commitments measured in tens or hundreds of billions of dollars, meaning sovereign wealth funds built from decades of oil rents are among the very few actors capable of financing projects at this scale. The financial architecture of the AI boom rests on a three-party coalition of US technology monopolies, asset managers and infrastructure funds, and sovereign wealth funds built from fossil-fuel revenues.

Chips for capital has also locked the gulf into the US technology order

The investment flows described above do not operate through arms-length market transactions. They are embedded in explicit geopolitical bargains that have structured the terms of Gulf integration into US-led AI infrastructure. The Commerce Department’s approval in late 2025 of tens of thousands of advanced AI chips to the UAE and Saudi Arabia marked a turning point, with G42 and Humain each receiving clearance to buy semiconductors equivalent to 35,000 of Nvidia’s most powerful Blackwell GB300 processors. These approvals did not happen automatically. They followed G42’s severing of ties with Chinese tech giant Huawei, its divestment from ByteDance, and Humain’s pledge not to purchase Huawei equipment. Washington had, in effect, locked major Chinese firms out of the region’s AI expansion. However, given the complexities of the global supply chain, it is unlikely that the region will fully decouple its technology ambitions from Beijing.

In January 2026, the partnership aimed to go further with the Pax Silica initiative, bringing the UAE and Qatar into a US-led effort to keep advanced semiconductors from reaching China. The deals carry strict security protocols requiring companies to declare planned uses, storage locations, and security measures against unauthorized transfers. Abu Dhabi’s G42 and Saudi Arabia’s Humain are now among the largest non-US operators of Nvidia’s newest hardware globally, positioned to serve AI computing needs for emerging markets across Asia and Africa. OpenAI, G42, Oracle, and Nvidia announced Stargate UAE, a planned one-gigawatt AI campus in Abu Dhabi that would be the largest of its kind outside the United States. OpenAI has stated the campus could eventually serve half the world’s population, underscoring the degree to which Gulf-based AI infrastructure is being positioned as a node in a global compute architecture with planetary reach.

This arrangement is a contemporary version of what the 1970s recycling circuit would recognize as a capital-for-system-access bargain. Then, the price of entry into the dollar-centered financial order was routing oil surpluses through US-centered financial intermediaries. Now, the price of entry into the US-led AI order is cutting ties with Chinese suppliers and accepting US security conditions around hardware use. Gulf sovereign wealth supplies capital. US firms supply chips and system access. AI infrastructure expands in both regions under negotiated terms. These deals are anchored on the premise that political alignment with Washington, abundant sovereign capital, and world-class infrastructure would make the region the third global center for AI alongside the United States and China. The Gulf’s AI ambitions are real, and so is its leverage. By trading Chinese suppliers for US chips, these states have converted their financial surplus into something they hope is more durable: structural position within the global AI order that the United States is working to build.

War has exposed the physical fragility of the new petrodollar circuit

However, the deals, pledges, and geopolitical alignments described above rested on another shared premise. Gulf governments had effectively offered Silicon Valley a bargain: bring your data, models, and chips, and the region would provide capital, energy, and political stability. That assumption was shaken in early March 2026 when Iranian retaliatory strikes across the Gulf damaged Amazon Web Services facilities in the United Arab Emirates and Bahrain. Two AWS data centers in the UAE were directly struck while another in Bahrain was damaged by a nearby drone strike, triggering fires, power disruptions, and service outages across parts of the company’s cloud infrastructure, which serve governments and businesses across the Middle East, South Asia, and parts of Africa.

The incident exposed a structural gap in the security assumptions surrounding the Gulf’s emerging AI hub, a vulnerability that some analysts already identified shortly after the wave of investment announcements last year. Yet, policy debates around Gulf technology expansion had largely focused on export controls preventing advanced chips from reaching China. Far less attention had been paid to the possibility that the physical sites where those chips operate—large, power-intensive industrial complexes—might themselves become targets in regional conflict. Commercial data centers are designed for redundancy against cyberattacks and equipment failures, but not for sustained missile or drone strikes. As one security analyst observed in the immediate aftermath, it is “cheaper to attack than to defend.”

The conflict also threatened the fiber-optic cables that carry the data these facilities process. Approximately seventeen submarine cables pass through the Red Sea, carrying the vast majority of data traffic between Europe, Asia, and Africa. Additional cables run through the Strait of Hormuz, serving Iran, Iraq, Kuwait, Bahrain, and Qatar. Iran’s Revolutionary Guard Corps declared Hormuz shut on March 3, threatening to destroy any vessel attempting passage. Houthi militants simultaneously resumed attacks on Red Sea shipping, ending a ceasefire that had held since late 2025. For the first time, both choke points that underpin Gulf data connectivity were effectively closed to commercial traffic simultaneously. The specialized repair ships required to fix severed submarine cables could not safely reach either passage.

In February 2024, a foretaste of this scenario had already materialized when three Red Sea cables were cut by the dragging anchor of a cargo ship struck by a Houthi missile, disrupting 25% of traffic between Asia, Europe, and the Middle East. One cable took five months to repair because vessels could not safely access the area. The 2026 conflict introduced the possibility of damage at a scale that no redundancy architecture has been designed to absorb. The crisis exposes a fundamental asymmetry in how Washington had approached its Gulf technology expansion. Security frameworks were designed to prevent advanced chips from reaching China—not necessarily to protect physical infrastructure from missiles and fiber cutting.

The deeper structural implication is that the AI economy has reproduced a vulnerability the oil economy required decades to even partially address. The Strait of Hormuz first became a geopolitical chokepoint in the context of oil flows, and US military planning around it developed slowly across the Cold War period. Oil pipelines capable of bypassing Hormuz exist as a partial hedge for energy in this respect. No equivalent bypass routes exist for undersea cables yet. The Gulf’s structural advantages remain intact for now, and the investments sunk into its AI architecture are too large to unwind overnight. But the conflict has validated the idea that the physical substrate of the AI boom is geographically constrained and militarily exposed in ways that planners underestimated.

The fragility of the interregnum

Antonio Gramsci famously observed of the disintegrating liberal-capitalist order of the 19th century that “the crisis consists precisely in the fact that the old is dying and the new cannot be born; in this interregnum a great variety of morbid symptoms appear.”⁶ This sense of being caught between eras perfectly captures the current state of global power. We are witnessing a transition where the digital future of the 21st century is being grafted onto the bones of the fossil-fueled 20th century, creating a hybrid order that is as technologically advanced as it is physically vulnerable.

The computational systems now reshaping labor, knowledge production, and strategic power are being built, in measurable part, with money made from burning carbon. They are being housed in facilities that sit within missile range of active conflict zones. The circuits of accumulation that defined the twentieth century are not being replaced by artificial intelligence, they are being extended through it—acquiring new institutional forms while preserving the geopolitical hierarchies, capital flows, and geographic vulnerabilities from which they emerged.

This extension carries significant consequences for the global financial order. It is reasonable to read U.S. control over frontier semiconductors and the “American AI technology stack” as a new axis of geopolitical leverage that is analogous to, and in some arenas complementary with, earlier forms of power exercised through energy security and oil-market politics. In this reading, access to the next technological era is increasingly mediated through U.S.-defined governance and security conditions, binding AI’s expansion to the institutional architecture of American power, including the dollar-centered financial system and its enforcement capacities.

Yet, this arrangement is inherently fragile. While the dollar still benefits from powerful network effects, fiscal instability and the frequent use of sanctions are eroding global trust. This has created a “financial interregnum” where states and financial actors seek to diversify their assets even without a clear replacement for the dollar.⁷ While escalations in the Gulf may temporarily reinforce the dollar as a safe haven, those same conflicts accelerate the long-term drive to bypass American-controlled infrastructure. The AI economy has effectively inherited the structural tensions of the oil order, built on the inequalities of exchange and the surpluses of a fossil-fuel past while increasingly embedded in a volatile security architecture. As these systems expand, will the very institutions that made the AI buildout possible eventually generate the pressures that drive the world to seek alternatives to them?

References

1. James, Harold. 1996. “The 1970s Capital Markets Versus the New International Economic Order.” In International Monetary Cooperation Since Bretton Woods. Oxford, England: Oxford University Press: 313.

2. Bank for International Settlements. 1974. “Fourty-Fourth Annual Report: 1st April 1973 – 31st March 1974.” Bank for International Settlements: 198. https://www.bis.org/publ/arpdf/archive/ar1974_en.pdf.

3. International Monetary Fund. 1975. “Annual Report of the Executive Directors for the Fiscal Year Ended April 30, 1975.” Washington, D.C.: International Monetary Fund: 35. https://www.imf.org/external/pubs/ft/ar/archive/pdf/ar1975.pdf.

4. International Monetary Fund. 1996. “A Second Oil Facility.” In IMF History (1972-1978), Volume 1. International Monetary Fund.

5. The BIS observed that “US banks played a major part in the recycling of funds” and described policy and regulatory changes that “tie the Euro-market even closer to the US market,” strengthening US-based banks’ capacity to compete in Euro-loan syndicates. See Bank for International Settlements. 1975. “Forty-Fifth Annual Report: 1st April 1974 – 31st March 1975.” Bank for International Settlements: 84; 132. https://www.bis.org/publ/arpdf/archive/ar1975_en.pdf.

6. Gramsci, Antonio. 1977. Quaderni del Carcere, vol. 1, Quaderni 1–5 Turin: Giulio Einaudi Editore: 311; English translation in 1971 Selections from the Prison Notebooks of Antonio Gramsci, ed. and trans. Quintin Hoare and Geoffrey Nowell-Smith. London: Lawrence & Wishart: 276.

7. Pforr, Tobias, Fabian Pape, and Johannes Petry. 2025. “Dollar Diminished: The Unmaking of US Financial Hegemony under Trump.” International Organization 79 (S1): S117–33.

The failed 2002 coup in Venezuela was quickly succeeded by a more insidious form of sabotage via networked infrastructure. This episode starkly revealed the vulnerabilities inherent in the commercial interdependence that binds global corporate networks today. However, to understand how these infrastructural ties—originally designed to bind nations together—became weapons of coercion, we must return to their ideological origins. Let’s briefly revisit the “pacific system” of commerce from the late eighteenth and early nineteenth centuries (see Part I for an introduction). This concept, which served as a precursor to the Monroe Doctrine, rested on the premise that peace could be secured through institutions and incentives rather than warfare. However, in espousing the pacific system, the American founders also harbored explicit aims to craft new governing arrangements across the hemisphere in the image of the United States (a government of powerful merchants, lawyers, and property owners). Writing in the immediate wake of the Napoleonic Wars and the War of 1812, and during the height of the Latin American wars of independence, William Thornton—the architect of the U.S. Capitol—identified the inherent risks of this pacific ambition in his 1815 text Outlines of a Constitution for United North and South Columbia. He proposed a hemispheric confederation across the Americas precisely because he feared that if the republic could not ensure its hegemony, the pursuit of peace would signal weakness and invite predation:

[I]f governments form themselves around us, essentially different; if daring chiefs, at the head of armies, and ambitious politicians, disturb our repose, it will be vain to offer the branch of peace. Our pacific system, if continued, would then but offer temptations to aggression, and we should repine at the necessity of armies and warfare, now so justly deprecated.¹

Thornton also feared that the vast internal distances of the continent would turn administrative delay into political fragmentation. He argued that the government must act with speed to avoid offering temptations to aggression. The emerging technology of the telegraph offered a solution to this dilemma by compressing territory into governable time. Thornton marveled that “intelligence can now be given with ease twenty miles a minute.” Once perfected, he believed telegraphs could “convey from the remotest bounds of this vast Empire, any communication to the supreme government” and enable measures to be taken “as rapid as the occasion may require.”² In theory, it allowed the state to secure unity and responsiveness without defaulting to the permanent militarized presence that the pacific ideal sought to avoid. Networked communication technologies were thus viewed from the outset (~200 years ago) as critical to command and control. It also enabled a new form of corporate power by coordinating the transnational geography of early oil companies, and today, serves as a key mechanism of imperial control.

This essay, Part II, explores how U.S. actors deployed these technologies when Hugo Chávez attempted to reform the Venezuelan oil economy. Chávez certainly embodies Thornton’s figure of an “ambitious politician disturbing the repose” of the United States. His anti-neoliberal Agenda Alternativa Bolivariana platform aimed to dismantle the established Puntufijo order and restructure the political system to be more democratically responsive. He argued that while oil must remain the productive core of the nation, the then-existing export model (Apertura Petrolero) entrenched a quasi-colonial dependency. To cure this, he proposed a strategy of re-nationalization and sovereign industrialization. This approach prioritized domestic scientific development to rebuild national power rather than yielding to transnational capture.³ He articulated this long-term vision clearly at his inauguration:

The economic battle, the transformation of the economic model, and the building up of productive economic activities to move away from the oil-dependent rentier model is a challenge that lies before us.⁴

The 2001 Hydrocarbon Organic Law served as the opening salvo of this agenda, but it immediately triggered a coup and sabotage campaign designed to topple his administration. This event is quite useful for understanding the trajectory of U.S. intervention in Venezuela. It provides the historical blueprint for the weaponization of networked technologies, a tactic that has since evolved into the more robust system of “weaponized interdependence” visible today.⁵

Trojan Horse—The Intesa Affair and the Meta-State

The events that transpired in Venezuela between December 2002 and February 2003 constitute a watershed moment in the history of technological warfare and state sovereignty. The “Oil Strike” of December 2002 may easily be misunderstood as a labor dispute or a civic protest (as the name evokes). In reality, it was a management lockout combined with a sophisticated technological sabotage operation—a kinetic assault on the nation’s economic foundation. Oil production collapsed precipitously from three million barrels per day to less than 25,000. For the first time in the digital age, a sovereign nation’s primary economic engine, Petróleos de Venezuela, S.A. (PDVSA), was brought to a near-total standstill not by external military bombardment, but by the weaponization of its own information technology infrastructure. This paralysis demonstrated to the Chávez government that sovereignty over natural resources is meaningless without sovereignty over the technology used to extract them.

To understand how the Venezuelan state was locked out of its own industry, one must understand the management architecture of the Apertura Petrolera of the 1990s (also introduced in Part I). Aside from looking for foreign investment, this policy introduced new management philosophies within PDVSA itself, forming a group known later as the “Meritocracy.” This faction was led by figures such as Luis Giusti—then-CEO of PDVSA who later became a key outside architect of the Bush administration’s National Energy Policy during the 2000–2001 transition. Under this leadership, the Meritocracy operated as a ‘meta-state’: a powerful corporate bureaucracy that functioned with near-total autonomy from the central government.⁶ They advocated for the internationalization of the industry, which in practice meant the privatization of non-core functions and the adoption of corporate structures that mirrored multinational oil giants rather than state-owned enterprises. The logic was that information technology was just a commodity to be purchased—a support service rather than a strategic asset to be guarded. Prior to 1996, PDVSA’s IT landscape was fragmented, consisting of disparate systems that the management sought to standardize for efficiency. This drive for standardization provided the rationale for a massive outsourcing project that would ultimately place the keys to the kingdom in foreign hands.

In 1997, this strategy materialized with the creation of INTESA (Informática, Negocios y Tecnología, S.A.). This joint venture was established to manage the entirety of PDVSA’s information infrastructure, from help desks to complex reservoir modeling systems. However, the equity structure was critically unbalanced: 60% was owned by Science Applications International Corporation (SAIC) and only 40% by PDVSA. This split effectively transferred decision-making power and technical control of Venezuela’s most critical infrastructure to a foreign entity.⁷ By 2000, INTESA managed nearly all of PDVSA’s data processing and transmission, creating a monopoly on the flow of information within the company.

The selection of SAIC as the majority partner remains the most contentious and revealing aspect of this arrangement. SAIC was not a standard commercial IT provider like IBM. It was, and remains, a primary contractor for the United States national security apparatus. SAIC’s leadership was heavily populated by former intelligence officers from the NSA, CIA, and the Department of Defense, with deep ties to the development of information warfare capabilities. By handing over its IT infrastructure to a firm embedded in the “Deep State” of a foreign power—one that would soon become hostile to the Bolivarian Revolution—PDVSA’s pre-Chávez management effectively placed the country’s economic nervous system under the influence of the U.S. imperial apparatus.

This transfer was not just administrative—it had material and cultural elements too. Approximately 1,600 PDVSA IT employees were transferred to INTESA, where they were subjected to a deliberate reshaping of their professional identity. The bureaucratic culture of the state oil company was replaced with a Silicon Valley mentality that emphasized stock ownership in the parent company. SAIC implemented an “Employee Stock Ownership Plan” (ESOP), and by 2002, 88% of INTESA personnel held stock in the U.S. defense contractor.⁸ This financial entanglement created a fatal conflict of interest. Many IT workers identified more with transnational corporate objectives than with the national mandate of the new Venezuelan government. When the political order came down to execute the strike, the “brain” of PDVSA had already been privatized, not just technically, but ideologically.

Black Box—Anatomy of a Cyber-Sabotage

Juan Fernández served as the visible face of Gente del Petróleo, coordinating the public spectacle of the strike, its effectiveness relied entirely on the covert maneuvers of the INTESA workforce. As Fernández declared political victory in press conferences, the technicians behind the scenes were executing the digital lockouts that made the stoppage a reality. This sabotage was made possible by the weaponization of the very infrastructure INTESA had built for efficiency; the centralization of the network transformed it into a mechanism of control through the “black box” phenomenon. Because INTESA retained exclusive possession of proprietary codes and administrative privileges, the Venezuelan government found itself locked out of its own industry. A rogue team of high-level administrators exploited their root access to dismantle the system from within, shutting down servers, changing passwords, and disabling remote access to ensure that the state remained paralyzed.

The sabotage was executed through a sophisticated combination of physical and digital vectors. Saboteurs installed hidden modems in walls and false floors to bypass physical security, allowing them to control and shut down systems from remote locations. They modified access keys and user privileges to physically lock loyalist engineers out of the digital systems required to run the plants. Crucially, they attacked the Supervisory Control and Data Acquisition (SCADA) systems that monitor pressure, flow, and temperature in pipelines and refineries. By blinding operators to critical safety data, they created the risk of over-pressure events and explosions, effectively holding the physical infrastructure hostage.⁹

The most crippling aspect of the offensive was the “total and unexpected collapse” of the SAP system.¹⁰ PDVSA had migrated its entire administrative, financial, and logistical workflow to SAP under INTESA’s guidance. This software managed everything from the procurement of spare parts to the generation of bills of lading for supertankers. When INTESA cut access, the blackout was absolute. PDVSA could not issue invoices for oil exports, meaning that even if oil could be physically loaded, it could not legally be sold under international maritime law. The company was unable to process Value Added Tax declarations, leading to a legal crisis with the national tax authority.¹¹ Logistical automated systems for spare parts went offline, forcing operators to cannibalize machinery to keep refineries running. Critical databases containing historical geological data were erased or encrypted, permanently damaging the management of high-pressure extraction environments.

The recovery of PDVSA, mythologized in Venezuelan political culture as the Rescate del Cerebro (Rescue of the Brain), was a desperate, improvised operation involving loyalist engineers, the National Armed Forces, and volunteer IT specialists. Resistance was led by a cadre of professionals who refused to join the strike, such as Socorro Hernández, who coordinated a team of “patriotic technicians” and hackers. Their strategy was multi-pronged. The National Guard physically occupied INTESA offices to prevent the destruction of hardware, while a collective of Venezuelan hackers worked around the clock to crack administrative passwords, boot servers into single-user modes, and reconstruct corrupted partition tables. In the absence of automated systems, retirees and volunteers operated refineries manually, physically turning valves and calculating flow rates with pencil and paper—a highly risky process that saved the industry from total collapse.¹²

The devastation was nonetheless immense. Official estimates place the direct economic losses between $14 billion and $20 billion. For the first time in its history, Venezuela ran out of gasoline, forcing the government to import fuel to keep food and medicine moving. The abrupt shutdown of wells caused irreversible damage to reservoirs in Lake Maracaibo, permanently reducing production capacity. In the aftermath, the Supreme Tribunal of Justice ruled that INTESA’s actions constituted a deliberate violation of the constitution, noting that the company had withheld data for “purely political reasons.”¹³ This legal battle confirmed what the government already knew—that the “meritocracy” had prioritized political damage over the preservation of the assets they claimed to protect.

The U.S.’s Overseas Private Investment Corporation (a federal agency that provided political risk insurance and development finance, later reorganized into the U.S. International Development Finance Corporation) explicitly rejected the defense that INTESA sabotaged PDVSA’s infrastructure, determining that the seizure of assets and operational control constituted a “total expropriation” by the state. It ruled that the takeover was uncompensated and confirmed that SAIC qualified for a payout under its insurance contract. PDVSA responded days later by demanding independent international arbitration. Thomas Wilner, the company’s counsel at Shearman & Sterling, dismissed the ruling as a political maneuver arguing that the decision ignored the facts to appease a corporation with powerful connections to the Bush administration.

Siege of Hardware—From Sabotage to Sanctions

From the ashes of the INTESA affair, the government institutionalized the rescue effort within a new internal department known as AIT. Staffed by the loyalists and volunteers who had saved the industry, this unit focused on stabilizing recovered systems and decoupling PDVSA from foreign contractors. The trauma of the shutdown drove a radical shift in state strategy as officials recognized that proprietary software was a black box that could hide backdoors or be revoked by external powers. President Chávez signed Decree 3390 in 2004 to mandate the use of free and open-source software across the public administration in an effort to eliminate the technological rent paid to U.S. corporations. The state launched projects like the Canaima GNU/Linux distribution to socialize a new generation away from Silicon Valley norms, and this quest for independence soon extended to physical infrastructure. By 2007, the government nationalized strategic sectors and partnered with China to launch satellites in an attempt to construct a technological stack immune to Washington’s kill switches. Yet implementation struggled to match this ambition. By 2013, AIT workers were still warning that a deepening dependence on proprietary systems like SAP compromised sovereignty. They argued that externalizing control over core commercial operations merely recreated the vulnerability of the Intesa era by handing expertise and access back to foreign vendors.¹⁴

While Venezuela made efforts to migrate its software, it still remained trapped by its hardware. Following the death of Chávez in 2013 and the subsequent collapse in oil prices, the U.S. strategy shifted from covert sabotage to overt economic warfare. The sanctions regime, particularly under Trump 1.0 (2017–2021), transformed the global supply chain into a weapon of mass destruction. As one official from the State Department put it:

The pressure campaign is working. The financial sanctions we have placed on the Venezuelan Government has forced it to begin becoming in default, both on sovereign and PDVSA, its oil company’s, debt. And what we are seeing because of the bad choices of the Maduro regime is a total economic collapse in Venezuela. So our policy is working, our strategy is working and we’re going to keep it on the Venezuelans.

The phrase “bad choices” invites a dual reading. While it overtly serves the propaganda narrative of administrative incompetence, it simultaneously operates with the cynical logic of a protection racket—implying that the regime’s fatal error was simply its refusal to capitulate to external demands, thereby forcing the US to inflict the damage. Meanwhile, the impact of sanctions on Venezuela’s infrastructure was fundamentally technological while its consequences spread to the nearly entire population. Modern infrastructure relies on a continuous stream of proprietary spare parts and updates from original equipment manufacturers, most of whom are U.S. or European firms like General Electric and Siemens.

Executive Order 13884, issued in 2019, induced a state of “overcompliance” in the global compliance departments of major corporations. Fearing U.S. Treasury penalties, these companies severed all links with Venezuela, creating a “technological embargo.” Adobe announced it would deactivate all accounts in Venezuela to comply. Oracle followed suit and threatened the database operations of Venezuelan banks and insurance firms.

The decay of the electrical grid provides the starkest example of technological warfare. Venezuela’s thermal power plants run on turbines that require proprietary parts and service contracts. Unable to purchase these, the state was forced to cannibalize machines, drastically reducing thermal generation capacity and forcing dangerous over-reliance on the Guri Dam’s hydroelectric output (which provides ~75% of Venezuela’s power). When the Guri transmission lines failed in March 2019, plunging the country into a massive, multi-day blackout, there was no thermal backup to stabilize the grid. While the government alleged a U.S. cyber-attack on the Guri’s SCADA systems—a plausible scenario given the vulnerability of legacy systems to cyber-physical attacks—the collapse could have been indirectly attributable to sanctions-induced decay. The timing of the blackout, coinciding precisely with the U.S.-backed attempt to install Juan Guaidó as interim president, and the immediate, detailed commentary from U.S. officials, suggested a coordinated campaign where infrastructure failure was weaponized to delegitimize the Maduro government.

(Left) Satellite view around Caracas, Venezuela, March 3, 2019. (Right) Satellite view of blackout around Caracas, Venezuela, March 8, 2019. Source: NASA.

Digital Enclosure and the Lawfare Machine

The INTESA episode serves as a historical prologue to a far more pervasive form of containment that has coalesced in recent years. This earlier intervention produced a reactive dialectic in which the Venezuelan state sought to secure its internal survival through alternative dependencies. Yet the struggle for sovereignty is not confined to the digital infrastructures of hybrid warfare—the legal domain remains a critical layer of the imperial “full stack.” A significant portion of the resource conflict has unfolded within arbitration panels and legal tribunals over the very sites where the architecture of the oil empire was first codified.

Major players that departed following the end of the 1% royalty rate during the Chávez era pursued international arbitration to seize or monetize Venezuelan-linked assets abroad. Beginning in 2007, firms like ConocoPhillips and ExxonMobil utilized these legal claims to create pressure. More recently, the litigation surrounding Citgo illustrates how arbitration enforcement transforms into asset seizure. Crystallex sought to collect a roughly 1.2 billion dollar award by attaching the United States holding company chain of PDVSA that owns Citgo. The courts facilitated this by treating the state oil company as the “alter ego” of the Venezuelan government. That ruling triggered a court-supervised auction in November 2025 of the shares held by PDV Holding. A judge approved a bid of roughly 5.9 billion dollars by Amber Energy which is owned by Elliott Management and led by Paul Singer. While the sale remains pending due to Office of Foreign Assets Control clearance and appeal processes regarding alleged conflicts of interest, the legal system still serves as the initial mechanism of dispossession.

Washington has since constructed a lattice of oversight around Venezuelan commodities that operates in tandem with these judicial maneuvers. Through sanctions and licenses, the United States permeates the global plumbing of trade and influences the movement of capital and cargo through banks, insurers, and shipping registries. This complex regime effectively determines who is allowed to interact with Venezuelan wealth. It shapes the rules and chokepoints of global logistics so that control over finance becomes the primary method to steer physical resources. Combined with technological sovereignty pursued since the early 2000s, the resulting economic asphyxiation creates a desperate domestic environment that forces the Venezuelan state to seek technological lifelines elsewhere. The government moved to weaponize digital tools for social control and utilized the Carnet de la Patria system to track beneficiaries of social programs—built with Chinese technology provided by ZTE. In other domains too, the state continues to shift its technological reliance from the United States to China and Russia in a bid to construct a parallel stack.

The resilience of this new architecture was tested in December 2025 when PDVSA faced a U.S. blockade alongside a coordinated cyberattack intended to paralyze operations. The true extent of the damage remains ambiguous. Juan Fernández of Gente del Petróleo argued that the industry is effectively blind and unable to endure the pressure because of deep logistical and financial voids. Venezuelan cybersecurity specialists rebuffed such claims and criticized the actions as a militarized campaign of denial-of-service distractions and targeted intrusions that echoed the sabotage of 2002. Whether the attack successfully penetrated the core of this new sovereign stack or was repelled by it remains an open question in a conflict now fought over server access and software protocols.

This technological and financial siege produced a profound social consequence that fed back into the machinery of American power. The economic collapse precipitated by sanctions created a desperate labor pool. As the bolívar lost its value, large numbers of Venezuelans turned to precarious data work.¹⁵ They performed image and video labeling, content moderation, and categorization through platforms that supply training data to major vendors like Scale AI (owned by Meta). Hao and Hernandez tell the story of a Venezuelan college student who prepared for a well-paid job in the oil industry but had to switch to full-time AI training when the economy collapsed. Venezuelans formed an unusually large share of the annotator workforce around 2018 and comprised up to 75% of the labor at firms servicing computer-vision pipelines which rely on microtasks that pay only cents per label.

In the years spanning 2024 and 2025, the Pentagon effectively industrialized this supply chain. AI was paid at massive scale to turn raw data into “AI-ready” training and evaluation services for the military while GenAI.mil pushed frontier models from Google, OpenAI, Anthropic, and xAI into military networks with the aim of being deployed for intelligence triage, operational planning, and targeting. In effect, the low-wage annotation economy reappears downstream as the perceptual substrate of surveillance and targeting. The everyday classification of objects helps military systems identify what constitutes a radar or air defense site from satellite or airborne imagery. In this sense, the concept of digital enclosure encompasses more than just sanctions, tribunals, and server access. It describes how imperial power routinizes human poverty into training data and then routinizes that data into the capacity to see, classify, and strike.

Perhaps the bombing and abduction of the Venezuelan President in January 2026 may be seen in a new light—certainly an escalation—a contingent conjuncture at the center of these converging systems rather than a definitive endpoint. The degradation of infrastructure through sanctions and the delegitimization of sovereignty through arbitration aimed to weaken the target long before troops landed. The possible use of AI-enhanced military capacities played a distinct role in these kinetic operations. The argument that Venezuela had stolen American assets was an inversion of reality. After decades of weaponizing software, banking codes, sanctions, and legal measures to asphyxiate the country, the so-called “Donroe Doctrine” simply migrated into the battlefield. Yet, such overt violence may reveal the weakness in the American imperial project rather than its strength.

This case has offered a stark warning to the Global South regarding the reality of weaponized interdependence. It demonstrates that national sovereignty is impossible without technological autonomy and robust military capacity. Ownership of the subsoil effectively means nothing if a hostile power owns the code that runs the extraction machinery and dictates the laws of commerce. The tragic irony is that this persistent conflict with the United States has only calcified the status of Venezuela as a petro-state and trapped it within the very extraction model it sought to escape—a victory for fossil capital. Ultimately, the January 2026 incursion simply makes the underlying geopolitical logic more visible. It is an attempt to preserve a global power structure that remains fatally tethered to fossil energy.

Notes

1. William Thornton, “Thornton’s Outlines of a Constitution for United North and South Columbia,” Hispanic American Historical Review 12, no. 2 May 1932: 206.

2. Ibid.: 214.

3. Chávez, Hugo. 1996. Agenda Alternativa Bolivariana. Caracas, Venezuela: Ediciones Correo del Orinoco.

4. Translation mine, from the original: “[L]a batalla económica, la transformación del modelo económico, levantar las actividades económicas productivas para ir saliendo progresivamente del modelo rentista petrolero, es un reto que tenemos por delante.” Hugo Chávez, speech at his inauguration as President of the Bolivarian Republic of Venezuela for the period (2000-2006), Legislative Palace, August 19, 2000.

5. Farrell, Henry, and Abraham Newman. 2023. Underground Empire: How America Weaponized the World Economy. Henry Holt and Company.

6. Ibañez, Pedro. 2012. “2 de Diciembre: A Diez Años Del «paro Petrolero», Sabotaje Criminal de Una Meritocracia Contra El Pueblo [December 2: Ten Years after the ‘Oil Strike’, a Criminal Act of Sabotage by a Meritocracy against the People].” United Socialist Party of Venezuela. December 2, 2012. http://www.psuv.org.ve/portada/2-diciembre-diez-anos-paro-petrolero-sabotaje-criminal-una-meritocracia-contra-pueblo/.

7. Sánchez, Eleazar Mujica. 2025. “A 50 Años de una Traición Petrolera (Parte III) [50 Years After an Oil Betrayal (Part III)].” Intersaber, September 28. https://intersaber.org/2025/09/28/a-50-anos-de-una-traicion-petrolera-parte-iii/.Velutini, Magdalena, and Robert Bottome. 2002. “Rechazando el éxito: El increíble caso de Intesa [Rejecting success: The incredible case of Intesa].” VenEconomía Hemeroteca 20 (1). https://luiscastellanos.wordpress.com/wp-content/uploads/2007/09/veneconomia-rechazando-el-exito-caso-intesa.pdf.

8. See Rodríguez, Miguel Ángel. Del Terrorismo Petrolero al Golpe Económico [From Oil Terrorism to the Economic Coup]. Reported in Toledo, Yuleidys Hernández. 2024. “Tellechea es vinculado con empresa involucrada en golpe contra Hugo Chávez [Tellechea is linked to a company involved in the coup against Hugo Chávez].” DiarioVea, October 22, 2024. https://diariovea.com.ve/tellechea-es-vinculado-con-empresa-involucrada-en-golpe-contra-hugo-chavez/.

9. See Rodríguez, Miguel Ángel. Del Terrorismo Petrolero al Golpe Económico [From Oil Terrorism to the Economic Coup]. Reported in Toledo, Yuleidys Hernández. 2024. “Tellechea es vinculado con empresa involucrada en golpe contra Hugo Chávez [Tellechea is linked to a company involved in the coup against Hugo Chávez].” DiarioVea, October 22, 2024. https://diariovea.com.ve/tellechea-es-vinculado-con-empresa-involucrada-en-golpe-contra-hugo-chavez/.

10. Translation mine. Seventh Superior Contentious Tax Court of the Judicial District of the Metropolitan Area of Caracas, Final Judgment No. 1740 (Caracas, June 30, 2014), case AP41-U-2004-000102. https://caracas.tsj.gob.ve/DECISIONES/2014/JUNIO/2101-30-AP41-U-2004-000102-SENTENCIADEFINITIVAN%C2%BA1740.HTML.

11. Ibid.

12. A detailed account is available in: Albarrán, Franklin, Samuel Carvajal, Carmen Chirinos, Maryann Hanson, Carlos M. Mujica, Haydee Nava, Eric Omaña, Carlos Polanco, and Tibisay Hung. 2013. Testimonios del rescate de PDVSA [Testimonies of the Rescue of PDVSA]. Caracas, Venezuela: Fondo Editorial Ipasme. http://ipasme.gob.ve/images/Varias/Fondo%20Editorial/Ciencias%20Humanas/Coleccion%20contra%20el%20Olvido/Libros/testimonio%20PDVSA%20TOMO%20I.pdf.

13. Constitutional Chamber of the Supreme Tribunal of Justice (Venezuela), Decision No. 827 (May 6, 2004). Cited in Seventh Superior Contentious Tax Court, Final Judgment No. 1740.

14. Roa, Luigino Bracci. 2013. “Trabajadores de PDVSA Rotundamente Opuestos a Que Empresa Alemana SAP Controle Sus Sistemas Informáticos.” Blogger. October 28, 2013. https://lubrio.blogspot.com/2013/10/trabajadores-de-pdvsa-rotundamente.html.

15. Posada, Julian. 2024. “Deeply Embedded Wages: Navigating Digital Payments in Data Work.” Big Data & Society 11 (2). https://doi.org/10.1177/20539517241242446.

Happy New Year…? Since December I’ve been working on a research piece on the U.S.-Venezuela conflict and recent events have compelled me to finish it. I will publish it in two parts, Part I (this) with Part II out in a few days.

The January 2026 kidnapping of Nicolás Maduro and bombing of Venezuela marked the kinetic flashpoint of a Venezuela policy explicitly focused on oil “restitution.” In mid-December 2025, Trump ordered what he called a “total and complete blockade” of sanctioned oil tankers moving in and out of Venezuela—an interdiction campaign Caracas condemned as piracy—and, after the raid, the White House and allied commentators moved quickly to cast “rebuilding” Venezuela’s oil system as the central prize of the operation, with U.S. firms positioned as the primary beneficiaries.

Major outlets have read the escalation through a revived Monroe Doctrine frame, reinforced by the administration’s December 2025 National Security Strategy, which elevates the Western Hemisphere as a priority theater and explicitly links regional strategy to protecting and jointly developing “strategic resources” while denying rivals access. In some ways it may represent an acknowledgement of a multipolar world order with the U.S. empire on the back foot, retreating into domination of its own “back yard.” Taking perhaps a more systemic view, Grace Blakely argues that Trump’s Venezuela stance is a blunt oil-power strategy—use coercion and sanctions leverage to redirect Venezuelan crude away from China and toward U.S. firms. She frames this as a response to the shale contradiction—lower prices undermine U.S. frackers—so Venezuela becomes a geopolitical substitute for weakening U.S. energy independence. This is certainly plausible: see my article on Fossil AI on the rising involvement of natural gas for AI data centers. Yet it is probably more likely that the through-line is less a single coherent master plan than a convergence of aims—oil industry restoration, hemispheric primacy signaling, and China-denial—assembled opportunistically around a spectacle of force.

With all this in mind, let’s take a step back to see the broader arc of U.S. interest in Venezuela—to look past the chaos of the present and trace the invisible wires that have bound Venezuela to the north for over a century. As this two-part essay explores, U.S. influence in Venezuela was a slow and deliberate project achieved by rewriting laws, co-opting elites, and weaponizing the technology of extraction itself. Part I examines the historical foundations of this empire and the class divisions that set the stage for the Chávez-era conflicts. Part II will then analyze the sophisticated sabotage of the state-owned oil company by managers linked to U.S. firms and Venezuela’s subsequent struggle for technological sovereignty amidst relentless pressure by the U.S. security apparatus and global capital.

The Pacific System and the Architecture of Extraction

Long before the Monroe Doctrine hardened into a rigid policy of intervention, it existed as a sentiment known then as the “pacific system” of commerce (meaning rejecting the use of force as a policy instrument rather than the Pacific Ocean). While the contradictions of this system are evident in hindsight, it’s worth mentioning that many early U.S. statesmen were eager to compete with European imperial powers but remained too vulnerable to challenge them militarily. The idea was also not without diverging interpretations. Thomas Paine claimed that commerce was a pacific system where nations would be useful to one another through commercial interdependence rather than warring over claims.¹ Following the Louisiana Purchase in 1803, the U.S. Minister to the United Kingdom, James Monroe, sought to place the pacific system on a secure basis through reciprocal respect from other powers (particularly Britain in this case). As president in 1823, Monroe shifted from a defensive posture to reveal a more assertive Doctrine that aimed to transform the Western Hemisphere into a closed sphere of influence.

The initial aspirations of the Bolivarian revolution initially mirrored the northern ambitions of the pacific system (at least as it existed in rhetoric). Simón Bolívar’s project of Gran Colombia—encompassing present-day Colombia, Venezuela, Panama, and Ecuador—was an attempt to create a sovereign counterweight to external European powers. Yet, Venezuela was birthed in the interstices of imperial conflict. The region was a theater for Spanish, French, British, and Dutch rivalries, its internal class struggles between elites and non-elites exacerbated by the shockwaves of the American, French, and Haitian revolutions. This fragility led to the faltering of the First Republic (1811–1812) in civil war, the collapse of the Second Republic (1813–1814) under Spanish reconquest, and the eventual dissolution of Gran Colombia (1819–1831) itself. The fragmentation of this project was not just a failure of governance but a geopolitical necessity for rising powers; the wedging away of Panama from Colombia in 1903 with U.S. support to facilitate plans for an isthmian canal remains the most blatant example of how U.S. empire-building relied on the atomization of Latin American sovereignty.

By the turn of the twentieth century, Venezuela had been integrated into the global capitalist system by competing imperialist powers, yet not as a sovereign partner. It was constructed as a specialized extractive enclave. While British and Dutch capital, represented by Royal Dutch Shell, had established the early footholds, the strategic imperatives of World War I fundamentally altered the landscape. The conflict laid bare the vulnerability of the West’s reliance on Middle Eastern or Asian oil. The United States, operating under an increasingly expansive interpretation of the Monroe Doctrine, sought to secure a proximate, defensible reserve of hydrocarbons. British hegemony was displaced by a U.S. imperative to turn the Caribbean Basin into an energy security lake. By this time, communications technologies like the radio and telegraph had transformed the corporate world by allowing for communications to extend across oceans and continents and into areas with strategic resources much easier.

The key instruments were still institutional and legal, and a prime initiating factor was the dictatorship of Juan Vicente Gómez (1908–1935). Gómez provided the ruthless stability required for capital investment, trading national resource sovereignty for the political and financial backing of the United States. This was an active, strategic collaboration; by sedulously avoiding conflict with foreign powers and paying off Venezuela’s external debt, Gómez positioned his regime as a “reliable partner.” In exchange, Washington insulated his dictatorship from the democratic waves sweeping other parts of the world. However, the most profound act of colonization during the Gómez era was not the drilling of wells, but the writing of laws.

In the United States, the “rule of first possession” and landowner mineral rights encouraged a chaotic proliferation of wildcatters. In Venezuela, the Spanish colonial tradition vested subsurface rights in the state, making petroleum development a matter of administrative concession. American giants such as Standard Oil of New Jersey and Gulf recognized that this centralized ownership structure offered a strategic advantage by providing a single regulatory focal point for negotiation. The Petroleum Law of 1922 marked the consolidation of this bargain. Drafted with direct input from industry representatives, the legislation incentivized a network of regime loyalists who monetized their access by transferring paper rights to foreign firms. The resulting framework was exceptionally accommodating to international capital. It regularized concession rights and structured royalties to ensure Venezuelan oil remained competitive while granting companies wide latitude in their daily operations. Although the subsoil formally remained state property, the law effectively constructed an extractive order where the government served as a passive collector of fiscal rents while strategic control resided entirely with the concessionaires.² Oil became Venezuela’s most valuable export, and by the 1930s the country was the world’s largest exporter of oil.

Metabolic Rift—Corporate Colonization of Consumption and Agriculture

As the legal architecture of extraction solidified, the physical and economic landscape of Venezuela underwent a radical, engineered transformation. The result was the creation of an enclave economy—a sector physically located within Venezuela but economically and legally detached from it. The infrastructure built by Standard Oil (Creole Petroleum) and Shell reflected this extractive logic as pipelines bypassed Venezuelan population centers, and refineries were deliberately constructed offshore in the Dutch Antilles (Aruba and Curaçao) to insulate high-value assets from mainland political instability. This ensured that Venezuela remained an exporter of low-value crude and an importer of high-value refined products, establishing a structural trade deficit in value-added goods that persists to this day.³

The massive influx of petrodollars led to currency appreciation to such a degree that traditional agricultural exports like coffee and cocoa became uncompetitive. By 1935, the agrarian economy had collapsed. The peasantry, displaced from the land and drawn by rumors of wealth, migrated to the periphery of oil camps and cities, forming a nascent urban proletariat that the capital-intensive oil industry could not absorb. However, the colonization of Venezuela in the mid-twentieth century extended far beyond the oil fields; it penetrated the metabolic processes of daily life—how Venezuelans ate, shopped, and moved.

The post-World War II era, characterized by the rhetoric of “modernization,” saw the deliberate dismantling of indigenous economic models in favor of systems requiring constant inputs from U.S. corporations. No figure better embodies this nexus of state power and corporate interest than Nelson Rockefeller. A scion of the Standard Oil dynasty, Rockefeller viewed Venezuela as a laboratory for “enlightened capitalism.”⁴ Through the Venezuela Basic Economy Corporation (VBEC), established in 1947, he sought to revolutionize the food supply chain in the image of the U.S. While framed as developmental aid as part of a modernization agenda, VBEC served as a mechanism for market capture.

VBEC established the Compañía Anónima Distribuidora de Alimentos (CADA), the first U.S.-style supermarket chain in South America. CADA fundamentally altered the retail landscape, displacing traditional open-air markets that relied on local peasant networks. Because CADA required standardized, mass-produced products that local agriculture could not supply, it turned to imports. By the 1950s, up to 80% of goods sold in CADA were imported from the United States, creating a downstream channel for U.S. agricultural surpluses and socializing Venezuelan consumers into preferring imported wheat and canned goods over traditional staples.⁵ This transformation extended to the processing of corn, the most basic national staple. The introduction of pre-cooked corn flour (Harina P.A.N.) by Empresas Polar in the 1960s revolutionized the diet but centralized control.⁶ The industrial process required specific strains of corn and milling machinery imported from the U.S., thus placing the nation’s food security increasingly into the hands of foreign firms.

The inherent volatility of this transition—from agrarian self-sufficiency to urban precarity—demanded a political containment strategy. Venezuelan politics had long been shaped by repeated cycles of military dominance and abrupt regime change, culminating in the fall of the Pérez Jiménez dictatorship in January 1958. With no credible mechanism to arbitrate distributional conflict in an oil-dependent economy, the elite forged the Puntofijo pact. This was the stabilization bargain for the modern rentier state. Its material basis was a class compromise: party elites agreed to share power, while labor and business reached accommodations to secure continuous production. Crucially, U.S. policy reinforced this pact, backing constitutional civilian rule only insofar as it preserved a climate of security for foreign private capital. In practice, Puntofijo institutionalized a rent-mediated order where oil revenues financed political incorporation, masking the deepening dependency.

Under the umbrella of this U.S.-backed stability, the industrialization policy of the 1960s and 1970s, known as “Import Substitution Industrialization,” was effectively captured by U.S. corporations. Automakers like General Motors and Ford established assembly plants in cities like Valencia, but these were not genuine manufacturing hubs. They relied on “Complete Knock Down” kits manufactured in Detroit and shipped to Venezuela for final assembly. This “truncated industrialization” meant that every car “made in Venezuela” actually consumed foreign exchange to import parts.⁷ Thus, the industrial sector did not reduce dependency on oil; it exacerbated it. By the end of the 1970s, Venezuela had transitioned from a productive agrarian society to a consumption-based society wholly dependent on the U.S. for everything from textiles to the engine blocks in their cars.

Neoliberal Trap—Financialization and the Crisis of Legitimacy

The illusion of prosperity maintained by the Puntofijo order began to fracture as the mechanisms of control shifted from direct corporate ownership of production to the disciplining power of finance. The 1976 nationalization of the oil industry is considered a moment of sovereignty. Yet, while the state assumed legal ownership, operational control remained with transnational firms through “service contracts” and “technology assistance agreements.” The newly formed state affiliates were placed under the new Petróleos de Venezuela S.A. (PDVSA), but they still mirrored the old foreign structures,⁸ ensuring that technological rent continued to flow North even after the physical rent was nationalized. Juan Pablo Pérez Alfonzo, a Venezuelan politician who is considered one of the “fathers of OPEC,” would later call the process a “sham nationalization” (nacionalización chucuta).

Meanwhile, the distributional glue that held the Puntofijo pact together—high oil revenues—began to dissolve as Venezuela moved into the 1980s. U.S. and European banks had aggressively lent recycled petrodollars to the Venezuelan state during the boom years. When U.S. Federal Reserve Chairman Paul Volcker raised interest rates in 1979 to combat U.S. inflation (the “Volcker Shock”), the service costs on Venezuela’s variable-rate debt skyrocketed. Combined with a decline in oil prices, this external shock pushed the nation into insolvency.

The ensuing debt crisis was treated less as a commercial bankruptcy than as a geopolitical opportunity. The “Washington Consensus” mandated structural adjustment as the price for bailouts, requiring the dismantling of subsidies, privatization, and trade liberalization. When President Carlos Andrés Pérez began to implement these measures in 1989, the social contract snapped. The Caracazo of February 1989 was a massive popular uprising against the abrupt rise in transport and food prices. The brutal military repression that followed, leaving hundreds dead, delegitimized the existing order and shattered any lingering myth of Venezuelan exceptionalism.

This fragility was exacerbated by the reckless deregulation of the financial sector—a key tenet of the World Bank/IMF program.⁹ In the early 1990s, the state aggressively liberalized interest rates and relaxed restrictions on foreign participation, yet oversight lagged fatally behind. This regulatory vacuum unleashed a speculative frenzy, as politically connected banks exploited debt arbitrage and “related-party lending” to funnel deposits into their own high-risk ventures. Amidst this institutional decay, Lieutenant Colonel Hugo Chávez led a failed military uprising in February 1992. Although the operation collapsed, the episode propelled Chávez into public view as the most legible antagonist to a discredited political class.

The neoliberal ruling order received its final blow when the banking sector’s house of cards finally collapsed in 1994. Beginning with the intervention in Banco Latino, the state was forced to socialize the losses, paying an exorbitant price to clean up a mess created by private greed while using the crisis as a pretext to denationalize and privatize the financial sector. The crisis stripped the domestic economy of the capital it needed to rebuild and facilitated the flight of wealth abroad. In the aftermath, the hollowed-out Venezuelan banks became easy prey for foreign capital. Under pressure to recapitalize, the government permitted the acquisition of the country’s largest financial institutions by U.S. and Spanish giants like Santander, BBVA, and Citibank.

To fill the void left by domestic capital flight, the state accelerated the Apertura Petrolera, or “Oil Opening.” This policy included auctioning operating service agreements (in 1992, 1993, and 1997) and establishing large Orinoco Belt “strategic association” megaprojects—such as Petrozuata, Hamaca, and Cerro Negro—built around upgrading extra-heavy crude for export markets. Designed to lure transnational capital back with the promise of exceptional legal and tax incentives, the state slashed royalties from 16.67% to as low as 1% and cut income taxes from 67.7% to 34%.¹⁰ This strategy effectively risked future revenue to secure immediate foreign technology and finance, prioritizing investor returns over the national interest. While it succeeded in raising production, the surge contributed to a market oversupply that crashed oil prices in 1998.¹¹ This economic catastrophe, combined with the deepening corruption scandals of the Puntofijo era, opened the final electoral pathway for an outsider coalition that could plausibly claim to refound the state.

Rupture—Sovereignty, Sabotage, and the Weaponization of Infrastructure

In 1998 Hugo Chávez won the election by a substantial margin. At the outset, the new administration attempted to rewire the fundamental circuitry of how authority and distribution were mediated in Venezuela. Following an antipoverty program, two major reforms were consequential for the coordinated resistance the Chávez government would soon face. First, a new constitution was approved in 1999 by popular referendum which reallocated power away from the party-mediated Puntofijo institutional order and toward a plebiscitary executive armed with new recall and referendum mechanisms, while weakening elite control over institutional chokepoints by redesigning the state around new oversight and electoral authorities that included more of the people. The new constitution also hardened resource sovereignty by reserving petroleum to the state and requiring state ownership of PDVSA, directly threatening the autonomy of the oil technocracy and the contractual expectations of foreign partners. And it constitutionally elevated the state’s directive role in the economy, treating property as protected but explicitly subordinated to “general interest” obligations and social-purpose expropriation, which signaled a more interventionist terrain for land, finance, and investment. Second, The “49 laws” (decree-laws) issued in 2001, were designed to rapidly restructure key domains and avoid legislative bargaining with elite opposition leaders—largely because the reforms represented a frontal assault on the accumulated layers of imperial integration—legal, territorial, and financial—that had been sedimented over the previous century.

A core part of the 49 laws was the Organic Law of Hydrocarbons. This legislation challenged the foundational logic of the 1922 concessions and the neoliberal Apertura of the 1990s. It reasserted state control over upstream activity, limited participation to joint ventures where the state held a majority, and significantly raised royalty rates. In parallel, the Land and Agrarian Development Law targeted the large, idle or uncultivated estates (a 5,000-hectare threshold in specified land classes), while bringing public and private agricultural lands into a regime oriented to national food security planning. This was a direct challenge to the landed property structures that had persisted since the colonial era. These laws did not just annoy the opposition—they threatened the material basis of the elite bargain. The opposition bloc that coalesced in response—comprising business federations (Fedecámaras), traditional unions (CTV), media owners, and the “meritocracy” of PDVSA managers—was united by the existential threat these reforms posed to their integration with global capital.

The institutional response to this reassertion of sovereignty was the coup attempt of April 2002. Declassified U.S. intelligence reporting from April 6, 2002 described “conditions ripening for coup attempt,” indicating that dissident military factions were stepping up efforts and might act within the month, and that plotters could try to exploit unrest from planned opposition demonstrations to provoke military action, with scenarios including detaining Chávez and key officials. When the interim government of Pedro Carmona briefly seized power for 47 hours in April, their first move was to dissolve the Supreme Court and the National Assembly and, crucially, to nullify the 49 Laws, revealing the true nature of the coup: a restorationist project designed to reset the rules of accumulation and re-secure the oil rent channel for the old elite.

However, the failure of the coup led to a so-called Paro Petrolero, or “Oil Strike,” of 2002–2003. It was during this conflict that the weaponization of the technological stack was first fully deployed, and the contours of the U.S. digital surveillance apparatus during the Bush administration took shape. This is covered in Part II.

Notes

1. Paine, Thomas. 1974. The Life and Major Writings of Thomas Paine. Secaucus, NJ: Citadel Press: 400.

2. On this saga, see: McBeth, B. S. 1983. Juan Vicente Gómez and the Oil Companies in Venezuela, 1908–1935. Cambridge University Press.

3. Earlier scholarship on “unequal exchange” provided a broad outline of this sort of trade relationship: Raul Prebisch, The Economic Development of Latin America and Its Principal Problems (United Nations, 1950); Emmanuel Arghiri, Unequal Exchange: A Study of the Imperialism of Trade (Monthly Review Press, 1969).

4. Rivas, Darlene. 2002. Missionary Capitalist: Nelson Rockefeller in Venezuela. Chapel Hill, NC: University of North Carolina Press.

5. Hamilton, Shane. 2018. “Supermercado USA.” In Supermarket USA: Food and Power in the Cold War Farms Race, 70–96. New Haven, CT: Yale University Press.

6. Felicien, Ana, Christina Schiavoni, and Liccia Romero. 2018. “The Politics of Food in Venezuela.” Monthly Review 70 (2). https://monthlyreview.org/articles/the-politics-of-food-in-venezuela/.

7. Coronil, Fernando, and Julie Skurski. 1982. “Reproducing Dependency: Auto Industry Policy and Petrodollar Circulation in Venezuela.” International Organization 36 (1): 61–94.

8. For example Lagoven was the old Creole/Exxon and Maraven was the old Shell.

9. For more details, see the 1995 Project Completion Report: https://documents1.worldbank.org/curated/en/502991468315002614/pdf/multi-page.pdf.

10. Mommer, Bernard. 1998. “The New Governance of Venezuelan Oil.” Oxford Institute for Energy Studies. https://www.oxfordenergy.org/wpcms/wp-content/uploads/2010/11/WPM23-TheNewGovernanceOfVenezuelanOil-BMommer-1998.pdf.

11. Boué, Juan Carlos. 2013. “Conoco-Phillips and Exxon-Mobil v. Venezuela: Using Investment Arbitration to Rewrite a Contract.” Investment Treaty News, September 20, 2013. https://www.iisd.org/itn/2013/09/20/conoco-phillips-and-exxon-mobil-v-venezuela-using-investment-arbitration-to-rewrite-a-contract/.

H.R. 4776, the “SPEED Act,” just passed the house yesterday, December 18, 2025. It’s basically a major overhaul bill that would speed up and lock in federal approvals for big energy and infrastructure projects (most likely for data centers) by shrinking environmental review and making it much harder for communities or courts to slow or stop projects once they’re approved. If this passes the Senate, I will do a deeper dive into its implications. Yet, permitting overhaul has been a years-long project to fast-track data centers. In this article, I look at the role of FERC and corporate influence in its governance of the grid as it also explores a nationalized framework that could diminish local and state authority over energy permitting.

AI data centers are making the electric grid the new choke point for growth, and inside FERC’s rulemaking and leadership politics, the biggest companies are increasingly able to turn “grid access” into guaranteed, preferential power.

Many parts of the U.S. power grid are already under stress. Old power plants are retiring, key transmission corridors are crowded, and it can take years for new projects to get approval to connect. Into that system come AI data centers and other hyperscale facilities that need enormous amounts of electricity, sometimes as much as a small city. Their developers want firm dates for when power will be available, and they often want to match their use to particular power sources. That scale creates hard choices with real consequences for everyone else. If the grid has limited capacity, regulators have to decide who connects first, what upgrades must be built, and who pays when new demand raises costs or increases the risk of outages. Those decisions shape household bills, local reliability, and whether communities end up subsidizing private growth.

The Federal Energy Regulatory Commission, or FERC, is the federal agency that sets many of the rules that make these outcomes possible. It oversees the technical and legal terms for connecting to the grid, using the transmission system, and paying for upgrades. Most of this happens through formal filings and written comments, not public meetings. For companies racing to build AI capacity, these behind-the-scenes processes can determine whether a project moves forward on schedule or gets stuck for years. Large tech firms increasingly show up in these proceedings, pushing for faster connections and clearer pathways to secure power. Google and Amazon, for example, have submitted comments and participated through energy affiliates. Some grid operators (like PJM) also partner with tech firms (like Google) on AI-driven interconnection planning tools which further ties corporate interests to the administrative machinery of grid expansion.

This is why personnel and priorities inside FERC matter. It is not a detached tribunal or impartial experts. It is a politically appointed institution whose judgments about urgency and fairness shape which reforms happen, and whose interests are protected. In that sense, the fine print of grid governance becomes a major lever for AI expansion, and a quiet site where public costs, private timelines, and system risk are negotiated.

This article is part of ongoing commentary related to AI industrial policy (parts I, II, and III) and the rise of Fossil AI. Please check out these articles for further background.

Share

FERC’s rulemaking ecosystem

Several recent FERC actions show how the push to power AI and data centers turns into binding rules that shape who gets electricity, how fast, and on what terms.

Let’s begin with the interconnection queue—the long line of power plants and grid-scale projects waiting for permission to connect to the grid. In Order 2023, FERC overhauled the generator interconnection process. Instead of studying new generation requests one by one in a “first-come, first-served” line, transmission providers must study them in clusters under a first-ready, first-served approach. FERC also tightened “commercial readiness” rules to reduce speculative requests—requiring stronger site control and larger readiness deposits to enter and remain in the queue. The aim was to fix an interconnection system that had become slow, backlogged, and difficult to manage. But this also means that projects with more money, land secured, and staff capacity can move forward more easily than smaller or less prepared developers. The Clean Energy Buyers Association, which represents large corporate buyers such as Google, Amazon, Microsoft, and Meta, argued that faster interconnection was essential to meeting clean energy goals. The final rule adopted many of the changes it supported, including grouped studies, stricter screens, and tighter timelines.

Transmission planning moved at the same time. Order 1920, issued in 2024 after a proceeding that began in 2022, requires 20-year regional transmission planning using multiple scenarios and directs regions to establish clearer up-front methods for allocating the costs of major new regional lines. For many customers, this matters because transmission is expensive and hard to build. Decisions about planning horizons and cost allocation can shift costs across states, utilities, and ratepayers for decades. Investor-backed developers and industry groups such as WIRES supported clearer rules because they reduce uncertainty for investment. Private equity has also adjusted its strategy around these constraints as it buys up energy and data center assets. Blackstone’s purchase of the 620 MW Hill Top gas plant in Pennsylvania has been described as a way to pair a power source with data center demand, reduce exposure to long queue delays, and deliver power fast enough for AI buildouts.

Relatedly, the most contested issue has been “co-location,” which is when a large data center sits next to a power plant and tries to take power directly, with less reliance on the wider grid. In 2024, FERC rejected amendments tied to a Talen and Amazon arrangement at the Susquehanna nuclear site, and the dispute moved into appellate review. The case became a test of where to draw the line between private arrangements and shared public systems. For developers, co-location can look like a shortcut around delays and network upgrade charges. For the grid, it raises a basic fairness problem. If a large private load uses the plant as an anchor but still depends on the broader system for backup, balancing, and contingencies, who is responsible for the resulting costs and risks?

That question sharpened in 2025. On February 20, 2025, FERC opened a show-cause proceeding involving PJM Interconnection and PJM transmission owners, drawing on a November 2024 technical conference and a complaint by Constellation against PJM. FERC asked whether PJM’s tariff, which is the rulebook that governs rates and service terms, clearly covers co-located load. Beneath that procedural issue was a practical one. Can a power plant effectively serve as a private power source for adjacent computing facilities while the regional grid still absorbs volatility, reliability obligations, and the financial consequences of system stress.

Two additional pressures pushed the issue toward a decision. PJM’s independent market monitor warned that approving large new data center loads without enforceable service conditions could lead to curtailments when the system tightens. Curtailments could fall on data centers, existing customers, or both. At the same time, federal officials pushed for speed. On October 23, 2025, the Department of Energy sent a Section 403 “Large Loads” letter directing FERC to start a rulemaking to accelerate the interconnection of large loads, explicitly including AI data centers. In that framing, delay is no longer just a technical inconvenience. It becomes a national constraint, and interconnection becomes a tool of industrial policy.

On December 18, 2025, FERC acted by ordering PJM to establish clear rules for co-located AI data centers and other large loads—one day after PJM reported record-high capacity auction prices linked to rising data center demand. It found PJM’s tariff unjust and unreasonable because it did not provide clear, consistent terms for customers serving co-located load and for customers taking transmission service on their behalf. In practical terms, FERC pushed PJM to adopt enforceable requirements, tighten accounting practices that can make loads look smaller on paper, and create faster pathways for new generation to interconnect. The aim is to move the buildout forward, but also to make the costs and reliability duties harder to shift onto everyone else.

Personnel is policy

The December order matters on its own terms, yet its technical language can obscure how nominally “independent” commissions like FERC may work in practice. Recent public arguments by the chair of the FCC have also underscored how contested that independence is, especially when executive priorities of the Trump administration pressure regulatory agendas. What happens next will depend on how FERC, through its five commissioners, continues to define the problem, set priorities, and balance speed for new large loads against reliability, cost shifting, and the obligations of the shared grid.

FERC is run by five commissioners who vote on major orders and rulemakings. Their professional backgrounds influence what they see as the core problem, which risks they prioritize, and what solutions they consider practical. The chair also matters because the chair sets internal management priorities and helps decide what comes onto the agenda.

In November 2024, then-Chair Willie Phillips dissented when FERC rejected PJM’s proposed approach to the Susquehanna co-location arrangement. He warned that the decision could threaten reliability and raised national security concerns in the context of growing data center demand. In February 2025, he issued a concurrence that placed that dissent within a growing Commission record on co-location. Soon after, Chair Mark Christie led a unanimous vote to open a formal review of co-located large loads in PJM, framing the issue as the need for clearer tariff terms, reliable operations, and fair cost allocation. From the outside, it is hard to see exactly what internal judgments shifted at each step, but the larger point is clear: leadership turnover can change the Commission’s priorities and pace, and therefore who has influence over what comes next.

Phillips resigned in April 2025 after the White House asked him to step down, and Trump later nominated David A. LaCerte to fill the seat, placing a new commissioner directly into FERC’s intensifying fights over data centers. Christie’s term expired June 30, 2025; the administration chose not to renominate him and moved to replace him with Laura Swett, amid reported White House pressure on FERC to move faster on “baseload” and fossil fuel infrastructure to meet rising demand. These moves preserved a 3–2 Republican-Democratic split, but they also shifted the Commission’s internal balance and direction, especially as Christie publicly criticized DOE’s large-load interconnection push as an overreach into state authority.

LaCerte’s profile is not the standard power market resume. FERC’s biography describes him as having two decades across public service, law, and regulatory policy, including senior roles at the U.S. Office of Personnel Management, the U.S. Chemical Safety and Hazard Investigation Board, and in a Louisiana cabinet position. He was confirmed on October 7, 2025, for a term expiring June 30, 2026. Reporting also highlighted links to Project 2025 and treated the nomination as part of a broader push to reshape regulatory governance. For lay readers, the takeaway is practical. FERC is now dealing with problems that look less like routine market tuning and more like physical bottlenecks. Interconnection delays, siting conflict, and tight reliability margins have become constraints on rapid AI expansion. A commissioner with a strong executive-branch management background may approach those constraints with a different style and different priorities.

Swett’s rise to chair soon after LaCerte’s confirmation may also point toward a wider energy buildout approach that includes natural gas infrastructure (see my article on Fossil AI). FERC’s biography describes her as having spent fifteen years litigating FERC matters for generating utilities, transmission owners, and pipelines, most recently at Vinson & Elkins. She decided not to recuse from certain data center-related matters involving a former-firm client and appointed a former Vinson & Elkins colleague to general counsel. These details matter because they connect FERC’s leadership to a professional world built around tariff design, regulatory litigation, and the legal work that makes large infrastructure projects possible—particularly for fossil fuel interests.

Vinson & Elkins’ was founded in Houston in 1917 during a Texas oil boom and has long specialized in the legal architecture of energy development. As is (I hope) evident at this point, energy systems require more than physical construction—they require legal machinery that secures access to land and capital, obtains permits, and stabilizes price formation through enforceable rules. The firm describes representing clients before FERC, regional grid operators, and state commissions in multi-billion-dollar proceedings, with clients across utilities, pipelines, producers, generators, and power marketers. It also emphasizes its long involvement in U.S. LNG export and import projects and its experience across the federal agencies that oversee LNG among other fossil fuel industries.

As the grid becomes the main permitting bottleneck for AI, energy regulatory law firms are becoming key intermediaries between infrastructure and capital. They help turn network capacity into a contractual right and an investable asset—credible to investors and regulators alike. Their influence becomes most visible when new loads arrive at scale and FERC has to decide what kind of certainty it will recognize for those loads, and what costs and risks it will allow to fall on everyone else.In PJM, those decisions are unfolding alongside a push to meet data center demand with new natural gas production, new power generation, and related infrastructure. This puts FERC’s December order directing PJM to clarify tariff terms into a broader perspective as these leadership changes may shape how FERC balances rapid AI load growth against the terms of a gas buildout that increasingly underwrites it.

The art of techno-statecraft

Across these dockets, it’s not just that AI uses a lot of electricity; it’s that a small group of powerful companies is learning how to turn that demand into enforceable advantages inside the grid’s rule system. Hyperscalers submit comments and push for faster ways to connect. Developers and transmission industry coalitions push planning changes that make new lines easier to approve and finance. Private equity looks for ways to secure power without waiting at the back of the line. The tactics differ, but the approach is the same. They do not merely build around regulation. They work through the rulebook so the rulebook works for them.

The “market” is not an external environment that law reacts to. Electricity is not as simple as supply and demand meeting on their own. This is what regulated infrastructure looks like in practice. The “market” is built through tariffs, cost-sharing rules, interconnection procedures, and the written records that justify them. These are not obscure technical details. They decide who gets priority, what counts as reliable service, who can rely on self-supply, and who stays exposed to congestion, price swings, and the costs of new upgrades.

Co-location makes the stakes easiest to see. FERC treated this as a tariff issue in its February 2025 show-cause proceeding, and that is a big deal because tariffs are where obligations become enforceable. In December 2025, FERC ordered PJM to write clearer terms for how co-located load is served and how transmission service works when someone takes service on the customer’s behalf. Whatever PJM writes now could become a model for other regions, shaping how future data centers get powered nationwide. This also matters because FERC (following the DOE directive) is also considering a national large-load interconnection framework (RM26-4-000) that could shift authority away from states. This follows suggestions by the Data Center Coalition (which includes the largest hyperscalers and others) in a Senate hearing back in February 2025 which advocated for the empowerment of FERC to override state authority on permitting projects deemed “in the national interest.”

Hyperscalers like Google have asked FERC for party status in RM26-4-000, arguing that its nationwide data centers give it a direct stake in the outcome (you can search theirs and other comments here). It also argued that in order to address the “foundational transmission issue” the U.S. should “adopt a holistic transmission planning process that optimizes grid development by accounting for both generators and the large loads planning to interconnect. … Absent a change of course, electric infrastructure appears to be the one pillar of our economy and our global competitiveness where we are increasingly falling behind, a result that would have devastating consequences for our economic growth and national security.” (Emphasis added, “Reply Comments of Google LLC” on Docket No. RM26-4-000: 3–4) Holistic planning may be sensible on its face, but it would also likely strengthen the position of the largest corporate buyers by making their load forecasts and timelines central to what gets built, while “competitiveness” and “national security” rhetoric serves to obscure this.

This is why FERC leadership and process matter. FERC does not govern through public speeches and open public hearings. It governs through what it puts on the agenda, how it runs formal proceedings, and how it writes orders that can survive rehearing and court review. That style of governing favors the parties that can hire specialist lawyers and experts, stay in long proceedings, and shape what counts as “reasonable” evidence and convincing argument—like the appeals to national security by Google quoted above. If you want to understand why outcomes tilt toward certain players, it helps to pay attention to who is in the room (literally and figuratively), who can afford to participate, and what professional networks and industry experience top officials come from. Not unrelated to its ongoing rulemaking efforts, FERC runs commissioner-led technical conferences on resource adequacy and reliability that have become tightly coupled to data center growth.

In sum, once new rules and pathways are created, speed and certainty become their own allocation mechanism. The parties who benefit most are those that can move immediately when a regulatory opening appears, locking in priority positions before competitors or publics can respond. Hyperscale operators can do this because they arrive with capital, repeatable site models, and a willingness to pay for accelerated outcomes, including bespoke infrastructure and contract structures that smaller customers cannot match. The long-run risk is a grid that increasingly serves large corporate customers as the default priority, while everyday customers and smaller businesses absorb more of the uncertainty, costs, and reliability risks that come with rapid buildout.

The AI boom has turned data centers into one of the fastest growing categories of industrial development, with a resource footprint that is hard to hide once it arrives as buildings, substations, transmission corridors, diesel generators, and new water demands. Federal analysts now estimate that data centers used about 4.4 percent of total US electricity in 2023 and could reach roughly 7 to 12 percent by 2028, depending on how quickly AI-driven demand grows and what happens to overall national load.

Over the past year, community resistance to data centers has stopped being a niche zoning or permitting dispute and become a recurring national storyline (if not a growing political liability):

• The Wall Street Journal has treated opposition as a social and political story, including a May 18, 2025 report on a small West Virginia community organizing against what residents were told could become one of the world’s largest data center complexes.

• The Washington Post has followed Northern Virginia backlash as both a legal fight and an electoral issue, including court action that invalidated Prince William County’s approvals for the “Digital Gateway” project and reopened a debate about the region’s development model.

• Bloomberg has framed the same region as a global capital of data centers where residents feel increasingly boxed in, with quality-of-life conflict pushing localities toward tighter rules and limits.

• Financial Times has elevated resistance as a national political economy story, including a recent report on a multibillion-dollar proposal in Georgia that drew opposition organized around water use, road burdens, and fears about who pays for grid upgrades.

Most recently, on December 12, 2025, Chandler’s City Council unanimously rejected a rezoning request that would have enabled New York developer Active Infrastructure to build a 422,000-square-foot AI data center campus, ending a high-profile local fight marked by intense community opposition and heavy outside lobbying. The dispute drew national attention after former Senator Kyrsten Sinema publicly urged approval while warning officials that “federal preemption” would soon limit local control. The rejection came one day after the Trump administration signed an executive order that adopts a preemption-style posture toward state AI regulation. Even though it does not directly override data center siting rules, it provides fresh rhetorical and legal leverage for proponents of rapid data center and energy expansion.

Yet there is a broader and much more contested battlefield within the national AI and energy buildout. Tax incentives for data centers are having a major impact on state budgets (not to mention energy costs). Meanwhile, dozens of states are moving in different directions, with different institutional structures, different partisan coalitions, and different positions over data center and energy siting. As communities begin to resist more and more, that resistance is now colliding with a governance regime designed to reduce the practical power of local refusal.

This article is the third in a series on AI industrial policy under Trump 2.0. Part I outlined early moves to define AI through a national security lens. Part II tracked a program of infrastructural acceleration and techno-industrial enclosure. Here, the focus narrows to the state and local politics of data center expansion, and to the legal and institutional tools that increasingly bypass local authority. Preemption is the key instrument in this shift, not as a technical doctrine, but as a way of reorganizing who gets to decide how the AI economy is built.

Share

Preemption as a governing regime in the AI buildout

In the United States, “local control” has been an important aspiration in political culture but is always conditional. Municipal authority exists because states delegate it. States can thus retract it or constrain its practical power. In earlier eras states used preemption to advance projects framed as public necessities, including highways, pipelines, large energy systems—and even more recently for housing. Similar arguments are returning under the aegis of AI competitiveness, grid reliability, and national economic security. In a political climate that currently promotes deregulation (read: corporate or market power), accelerated permitting, and industrial expansion, preemption becomes less an exception and more a governing regime. It aims to convert land and infrastructure into platforms for rapid buildout, while reducing the friction of local negotiations that can slow projects or raise costs—at least for the investors and corporate actors involved.

Preemption also operates selectively. In some policy domains, it is justified as a way to prevent local exclusion, as in state overrides aimed at loosening barriers to housing. In the AI and energy arena, it often functions as a ceiling rather than a floor. It blocks cities and counties from adopting stricter environmental standards, labor requirements, or climate-oriented energy rules. It also reflects a political calculation about influence. State-level policymaking is easier for concentrated industries to shape than hundreds of local hearings, ordinances, and planning commission meetings. The result is a governance pipeline that makes AI infrastructure more deployable at scale.

That pipeline is now visible because the local stakes have become too large to ignore. The AI economy is not landing softly on existing territory. It is arriving through discrete, bulky interventions, in the form of rezoning requests, new substations, new generation permits, and new transmission lines. The conflicts that follow are not simply about aesthetics or parochialism. They are about who bears costs, who captures benefits, and what democratic voice means in a buildout framed as urgent and strategic.

The current wave of state preemption is not a single instrument. It is a family of strategies that reorganize where decisions happen and whose decisions matter. Three patterns stand out.

Some states explicitly remove local authority. West Virginia offers one of the clearest examples. Its One-stop Shop Permitting Program (HB 2002) centralizes and accelerates permitting timelines while its Power Generation and Consumption Act (HB 2014) promotes “high-impact data center” development tied to on-site generation and microgrid districts—exempting projects from county and municipal zoning and limiting the reach of local ordinances related to noise, lighting, and land use. Residents in Tucker County learned about a proposed gas-linked complex only after a permit notice circulated. The organizing that followed was intense, but the local government’s formal leverage had already been removed. In this structure, public participation is reduced to mere expression rather than a deliberative feature, a way to register harm without the power to condition it.

Louisiana pursued a similar outcome through deal-specific lawmaking designed to secure Meta’s roughly $10 billion Hyperion campus. Legislators rewrote tax rules to create data-center-specific exemptions and advanced a site that had already cleared environmental review, limiting the scope of new local scrutiny. Entergy Louisiana was fast-tracked to build three gas plants and a transmission corridor through a state-centered process with constrained local input. The effect is not only a project approval. It is the creation of a de facto entitlement zone for digital infrastructure, where the state pre-arranges the conditions of development.

Other states have considered similar moves. Pennsylvania has recently proposed legislation that would fast-track data center permitting and craft special Commonwealth Opportunity Zones for AI infrastructure, among others. North Carolina’s 2023 “energy choice” law takes a different route by barring local restrictions on particular fuel sources in buildings, preempting municipal efforts to steer electrification or decarbonization. Across these cases, the language of reliability and competitiveness does political work. It redefines the “public interest” in ways that consolidate authority upward.

A second strategy is a de facto bypass through utility and grid governance. Even where local zoning remains intact, states can hollow out local influence by placing the most decisive approvals inside utility commissions, grid operators, and energy permitting channels. Virginia, home to the world’s largest concentration of data centers, shows this split clearly. Counties such as Loudoun and Prince William can tighten zoning for data center parcels, but the approvals that matter for power plants, transmission lines, and network upgrades sit with the State Corporation Commission and related state-level institutions. This is why resistance in Virginia so often becomes a fight about the broader development model, not a single building, and why court challenges to land use approvals can still leave the larger growth pressures intact.

Texas illustrates a variant of this as well. Counties often have limited zoning authority, and statewide planning levers run through electricity regulation and interconnection governance. Where the decisive approvals sit inside grid institutions, the distribution of power shifts. Developers that can finance new generation, negotiate special arrangements, or move quickly through interconnection queues gain leverage. Communities sometimes respond by seeking the limited tools available to them, including incorporation or changes in municipal status, simply to gain basic land use authority. When the state’s strongest levers sit in energy regulation and economic development, local control exists at the margins of a system built to prioritize capacity and investment.

A third strategy works through fiscal dependence and legal pressure. Incentive architectures can discipline local governments even without formal preemption. Indiana’s data center tax exemption, which can run up to fifty years, removes key local revenue streams and encourages counties to compete for projects even when public benefits are limited. Near Indianapolis, residents opposing Google’s proposed 460-acre campus won a temporary victory, but the state’s incentive logic makes repeat proposals likely. Fiscal dependence does not merely reward compliance. It trains local governance to anticipate compliance.

Courts can tighten the vise. In Georgia, developers have used litigation to challenge local denials and force approvals. After Monroe County rejected a billion-dollar rezoning, the developer sued, arguing the denial was arbitrary and inconsistent. Meanwhile, counties that approved comparable projects have faced resident lawsuits alleging procedural violations. This two-sided legal squeeze narrows the practical space for democratic contestation. Local governments face risk whether they approve or deny, and that risk itself becomes a disciplining force.

The distributive stakes of accelerated siting

Once decisions are relocated upward, the equity implications become harder to avoid. The burdens of large energy and tech projects land where the infrastructure is built, while the financial upside often flows elsewhere to firms, state budgets, and distant consumers. Preemptive siting can therefore function as an extraction mechanism that is legal rather than illicit. It externalizes costs tied to land conversion, air emissions, water demand, and the wear placed on local services, while preserving a narrative of statewide competitiveness and national necessity.

Fiscal design is one obvious fault line. West Virginia’s approach is instructive because it makes the revenue question explicit. Reporting on the state’s microgrid and data center program describes a structure in which host counties receive only a minority share of property tax revenue from facilities, with a large portion routed into state-controlled funds. Even if formulas shift under public pressure, the underlying message remains. Localities will host the industrial footprint, but they should not assume they will capture the fiscal return that would traditionally justify industrial development. That mismatch is not a side effect. It is often part of the deal architecture that makes projects attractive to investors.

Utility governance creates a second channel for cost shifting. When data centers drive new generation and transmission investments, the central question becomes who pays. If utilities spread upgrade costs across all customers, households subsidize corporate operations through higher rates. That concern is now prominent enough to shape state politics, not only local hearings. Georgia’s surge in proposed capacity additions and the controversy over whether ordinary customers will be left with the bill have become central to public debate, with regulators and advocates explicitly contesting the allocation of infrastructure costs.

Some states are experimenting with ways to limit cross-subsidies by requiring very large loads to internalize supply costs. Utah’s 2025 SB 132 is an example. It defines large loads at the scale of 100 megawatts or more and establishes a process in which the utility evaluates whether it can serve the request without major new investments, with the option for the customer to procure supply elsewhere if the utility cannot. In principle, this shifts risk away from existing ratepayers. Yet it can also deepen a privatized energy landscape where new generation is sited to serve private campuses, potentially increasing localized pollution burdens even as broader grid pressures ease. A policy meant to stop cross-subsidies can still produce local externalities, only now through more privatized power islands.

Environmental impacts sharpen the distributive conflict because they concentrate in place. West Virginia’s fights have featured residents describing how a gas-linked data center complex would alter air quality, noise, and the sensory life of a scenic region with an outdoor recreation economy. Under a preemptive framework that removes zoning leverage, accelerated siting becomes a kind of relocation. It moves decision-making away from those who live with the consequences, while leaving them with the everyday burdens of the built result.

Water politics intensify the same dynamic in arid regions, where scarcity makes every new demand legible as a trade-off. Tucson’s Project Blue fight turned on exactly this leverage. A proposed $3.6 billion development required annexation and municipal water access, prompting Amazon to back out and the city to halt annexation talks after public outcry and to pursue stronger local conditions on large water users. The struggle was not simply about a single facility. It was about whether water, once treated as a background utility, can remain a public governance object in an era when compute tries to behave like an entitled industrial load. Nevertheless, the data center developers are hoping to continue with the project by seeking Pima County and Arizona Corporation Commission (ACC) approval.

The Great Salt Lake Basin shows what happens when the accounting itself is contested. In Utah, recent reporting on proposed transparency legislation notes that an analysis cited in legislative debate found the National Security Agency’s Bluffdale facility consumed more than 23.5 million gallons of water in a single month, a figure that becomes politically explosive when set against the region’s broader drying crisis. The basin’s vulnerability is not theoretical. An economic analysis for the Great Salt Lake Advisory Council estimates annual monetized costs in the range of roughly $1.7 to $2.2 billion associated with declining water levels, driven in part by dust and health impacts from exposed lakebed. When water use is treated as proprietary and disclosure is weak, the public cannot evaluate whether sustainability claims match reality or whether a shared resource is being quietly converted into an industrial input.

These material patterns are why the AI boom is extractive. Value and strategic advantage flow outward, while pollution, landscape change, and service burdens stick locally. Data centers can deliver short-term construction spending, but their long-term job base is often thin relative to their land, water, and power demands. The computing they produce is sold to distant firms and markets, so the main beneficiaries are rarely the host community. Data center expansion is also an environmental justice story about leverage. Affluent, well-networked places can hire counsel, mobilize quickly, and slow projects down, while rural and lower-income jurisdictions are often pursued precisely because land is cheaper and resistance is easier to contain.

Federal preemption as tailwind, not bulldozer

The politics of preemption extend beyond statehouses. An executive order titled “Accelerating Federal Permitting of Data Center Infrastructure” was already made back in July. On December 11, 2025, President Trump signed another executive order titled “Ensuring a National Policy Framework for Artificial Intelligence,” aimed at challenging or deterring certain state AI laws and pushing toward a single federal posture. The order does not directly rewrite local zoning codes or automatically override state statutes. Its practical impact depends on agency action, funding conditions grounded in statutory authority, and litigation outcomes. But its political effect is immediate. It frames a fragmented state landscape as a national competitiveness problem and offers a federal vocabulary that states and industry can use to discipline local resistance.

The order directs the federal government to identify state AI laws it considers “onerous,” including laws that require AI models to alter “truthful outputs” or that compel certain disclosures, and it establishes a posture of coordinated federal challenge through the Department of Justice and related agencies. It also signals a fiscal approach, instructing Commerce to review whether certain federal funds can be conditioned on state compliance, a move that has already generated legal debate about federal authority and constitutional limits.

This is where the connection to water and energy transparency laws becomes plausible even if it is not explicit. The executive order does not name water, energy, data centers, or environmental reporting as direct targets. Yet it does target a category of state rules framed as compelled disclosure, and at the same time statehouses are actively debating transparency requirements for data center water use. At least eight states introduced legislation in 2025 to require data centers to report water use, precisely because communities and regulators are struggling to govern what they cannot see. The likely near-term effect is therefore not federal field preemption that sweeps away local authority overnight. It is a federal tailwind that strengthens the hand of state-level centralizers and industry advocates by treating local and state constraints as obstacles to a national AI project.

Preemption aligns state power with a new generation of energy-intensive, capital-heavy infrastructure, and it redefines the public interest in terms of speed, scale, and investor certainty. The question this raises is not simply whether communities can plan better. It is whether democratic control over territory can survive an industrial policy that increasingly treats land, energy, and water as inputs to a national AI project, while relocating decision-making away from the places asked to absorb its costs.

Over the past year, the financial press now routinely links the data center construction wave to broader fears of an AI bubble, extending beyond tech equities into power markets and infrastructure finance. At the same time, analysts estimate that global AI infrastructure spending could reach trillions of dollars over the next five years, a substantial share of it financed with debt on the assumption of robust and enduring cash flows. A survey of major fund managers reported that many institutional investors believe firms are overspending on AI infrastructure. The global head of digital Infrastructure for KKR even cautioned clients to “look beyond the bragawatts”: to treat headline announcements of data center capacity less as evidence of guaranteed demand and more as marketing in a crowded capital-raising environment. The warnings echo earlier technology booms where future growth was capitalized long before it was realized.

A second strand of reporting follows this logic downstream into utilities and data center operators. In the United States, utility stocks rallied on the back of an AI-power narrative that promised decades of rising demand and guaranteed cost recovery through ratepayers. That rally has already shown signs of strain as “AI bubble” concerns mount. Analysts now warn that if the most aggressive electricity demand scenarios fail to materialize, utilities will find themselves overvalued and holding stranded generation, pipelines, and grid upgrades built for a load that never arrived. Data center executives, for their part, publicly acknowledge the risk. Profiles of firms in India and Europe describe executives trying to “thread the needle” between capturing AI-related growth and avoiding overbuilding in what they frankly characterize as a “bubbly” market.

A series of widely syndicated Associated Press pieces has brought these concerns into the public regulatory arena. One article highlights “eye-popping” claims by U.S. utilities that they will need two to three times more electricity within a few years to serve new AI data centers, and asks whether these forecasts are padded with projects that will never break ground. Another reports growing skepticism among lawmakers and regulators in states such as Pennsylvania, Texas, and Ohio about whether speculative data center deals should be embedded in “base case” forecasts at all. Grid Strategies, a consultancy closely watched in energy circles, estimates that utilities now project roughly 90 gigawatts of added data center load by 2030—equivalent to nearly 10 percent of anticipated U.S. peak demand—while independent market analysts place a more realistic figure in the 60–65 gigawatt range. The wedge between those numbers represents not just methodological disagreement, but a vast volume of potentially overbuilt infrastructure.

Forecasts have also been revised upward with startling speed. BloombergNEF’s estimate of U.S. data center demand by 2035 has jumped by more than a third in a short period, and similar upward revisions are now steering plans for multi-billion-dollar generation, transmission, and pipeline investments. Advocacy and research groups such as IEEFA and As You Sow warn that utilities and private developers are using these exaggerated projections to justify a new wave of fossil gas plants and related infrastructure, even as the technical trajectory of AI remains deeply uncertain. The industry itself concedes that substantial gains in hardware efficiency, changes in model architecture, or a slowdown in training could sharply reduce load growth. Yet, in practice, today’s high-end forecasts are often treated as inevitabilities rather than contingent scenarios.

This combination of exuberant projections, rapid forecast inflation, and willingness to treat speculative demand as a planning baseline is the core problem. AI and data center loads are growing, but the financial and policy discourse around them is increasingly wrapping real technological change in a narrative that licenses overbuilding of fossil-based energy and socializes risks of a downturn.

This is part I of a multipart series. See part II on petrodollars and the AI infrastructure boom.

Private equity is using the AI boom to fuse data centers with fossil-heavy energy portfolios

If financial journalism frames AI infrastructure as a potential bubble, private equity and infrastructure funds treat it as an opportunity to reshape control over energy systems. Their pitch to clients is deceptively simple: AI, data centers, and electric vehicles will drive a structural surge in electricity demand; grids are constrained; therefore, owning flexible, dispatchable power is the surest way to capture the upside. Even worse—the narrative often launders an “energy transition” narrative that functions, in practice, as a strategy to lock in and revalue fossil assets under a variety of rationales from economic competitiveness and national security to the inevitability of technological revolution.

Let’s take Energy Capital Partners’ 2023 ESG report (which it calls “A Responsible Approach to Value Creation”) as an example—though many other investment firm reports are equally representative of the broad narrative framing. It presents the rise of AI, data centers, and EVs as a naturalized, quasi-inevitable source of load growth. That demand is not framed as the outcome of strategic corporate choices, regulatory decisions, and public subsidies, but as a kind of secular force to which investors must adapt. On that basis, the firm defends a “balanced portfolio” in which long-lived gas (and even coal) assets are recoded as responsible contributions to grid reliability in an era of AI-driven strain and interconnection bottlenecks. The effect is to turn what might otherwise appear as aging, carbon-intensive plants into future-proofed infrastructure keyed to a glamorous new demand narrative.

BlackRock’s recent moves demonstrate the scale at which this logic is being pushed. Its $12.5 billion acquisition of Global Infrastructure Partners (GIP), followed by GIP’s roughly $40 billion purchase of Aligned Data Centers in partnership with Nvidia, Microsoft, xAI, and MGX, has effectively positioned BlackRock at the center of the AI infrastructure buildout. The firm now stands among the largest private owners of gas-fired power plants globally and is expanding its control over regulated utilities and generation assets through deals such as the $6.2 billion acquisition of Allete and advanced talks to acquire AES. Together, these moves would give BlackRock direct or indirect stakes in around fifty fossil-fuel plants with more than 30,000 megawatts of capacity across multiple countries.

By consolidating ownership of both AI data centers and the fossil-heavy power portfolios that feed them, BlackRock is not simply “responding” to AI demand—it is helping to coordinate how and where that demand is translated into concrete energy infrastructure. The firm’s strategy relies on a particular vision of the future: that AI load will keep rising, that regulators will tolerate continued dependence on fossil generation in the name of reliability, and that data centers will remain sticky, long-term customers willing to sign contracts that underwrite new plants or extend the life of old ones. In such a world, owning both the server halls and the smokestacks becomes a vertically integrated play for a fossil-fueled digital capitalism.

Other asset managers are pursuing similar paths, stitching together portfolios of data center capacity, gas plants, and transmission assets under the rubric of ‘supporting the AI revolution.’ In early 2025, Blackstone announced its purchase of the 744-megawatt Potomac Energy Center gas plant in Loudoun County, Virginia—“Data Center Alley,” home to more than 130 data centers and roughly a quarter of global data center capacity. The firm explicitly described the acquisition as part of a strategy to back power infrastructure serving data centers and AI, calling it one of its “highest-conviction areas.” The plant is thus reimagined as a critical asset in a global digital hub, its future profitability secured by the presumed inevitability of AI expansion.

These are part of a string of large-scale partnerships. KKR and Energy Capital Partners have announced a $50 billion strategic partnership to invest in data centers and power generation linked to AI growth. A consortium including KKR and PSP Investments has taken a nearly 20 percent stake in transmission companies serving Midwestern states, tying high-voltage infrastructure more closely to private equity’s AI thesis. Brookfield has unveiled a $5 billion “AI infrastructure” partnership with Bloom Energy and joined with Cameco and the U.S. government in an $80 billion deal leveraging Westinghouse nuclear technology. Together, these deals map a coordinated effort to build portfolios in which AI demand is the justificatory glue binding fossil-heavy generation, nuclear assets, transmission networks, and hyperscale data centers into a single investment narrative.

In effect, AI is a marketing device that bundles old and new infrastructure into a single growth frontier for finance capital. But the assumption that AI workloads will reliably fill gigawatts of capacity for decades is exactly what analysts and regulators are beginning to question—and those doubts come into sharpest focus in utility planning.

Utility planning models are turning speculative AI demand into public obligations for new (fossil-heavy) capacity

In Georgia Power’s 2025 Integrated Resource Plan, probabilistic hype about data centers is converted into a concrete build-out. Between its October 2023 IRP update and Q3 2024, its projected “Large Load Pipeline” grew from 16 GW to 34.6 GW, with 8 GW now coming from customers that have formally committed. This is a sharp revision from its 2023 plans, with the majority attributed to AI-oriented data centers. To meet this, the near- and medium-term portfolio is anchored in new and extended gas units and legacy coal, supplemented by battery storage, incremental nuclear uprates, hydro modernization, and a largely RFP-driven expansion of solar and wind. At stake is how a highly uncertain wave of AI and data center demand is being operationalized inside the IRP—and, in particular, how Georgia Power’s load forecast and its Large Load Realization Model are used to justify this new tranche of infrastructure.

On May 5, 2025, expert testimony poked holes in Georgia Power’s 2025 IRP load forecast—the projection it uses to justify new power plants and grid upgrades—and, specifically, its Large Load Realization Model. The expert witnesses accepted the broad architecture of the model but demonstrates that its innards are biased toward overestimation. Short-term commercial demand models have systematically overshot actual usage, and these inflated figures are then propagated into longer-term projections, baking excess into the planning process. The evidence is concrete: Georgia Power’s 2024 summer peak came in significantly below what its 2025 forecast had anticipated, even as the company cited rising AI and data center loads as reasons to accelerate capacity additions.

The sharpest critique focuses on how the model treats data centers. A “pipeline” of proposed large loads is used to estimate how many projects will ultimately connect to the grid. In the 2025 forecast, the model “unreasonably biases” the likelihood that data center projects will proceed, despite internal evidence that such projects cancel more often than other industrial loads (page 6, 51). By 2037, roughly 83 percent of the company’s large-load pipeline, measured on an announced-load basis, is made up of data centers. The forecast is therefore extraordinarily sensitive to the fate of a single volatile industry. Yet the model does not transparently account for higher cancellation rates, shorter contract terms, the shift from energy-intensive training to less demanding inference, or the possibility that many of today’s headline AI projects will never operate at full scale or for very long.

The testimony situates this modelling problem within a wider instability of AI-driven demand. It points to examples such as DeepSeek, a Chinese AI firm reportedly leaving up to 80 percent of its computing resources unused at times, and whose models are claimed to be far more energy-efficient than competitors. Industry analysts expect roughly a third of generative AI projects launched in the early wave to be abandoned or substantially scaled back by 2025 (page 78). With this in mind, building multi-decade infrastructure plans around the most bullish AI and data center scenarios begins to look less like prudent planning and more like a speculative gamble with public money. Nevertheless, in mid-2025, Georgia’s PSC voted to approve Georgia Power’s 2025 IRP.

*Georgia has adopted some ratepayer protections—primarily the PSC’s January 2025 large-load rule and a base-rate freeze approved just before the 2025 IRP—but a stronger statutory fix (see SB 34) has stalled, and other PSC decisions have simultaneously reduced transparency and left significant room for data-center costs to show up on bills later.

Across states, speculative AI load forecasts are colliding with growing regulatory pushback

As I’ve alluded to at the outset, Georgia is just one of many states in which this dynamic is playing out. GridUnity’s CEO reports that one large utility client saw roughly 30 percent of big-load proposals cancelled in 2024. A Grid Strategies report found that utilities in states such as Virginia, North Carolina, South Carolina, and Georgia have filed demand forecasts tied to data centers that exceed independent industry assessments by as much as a factor of four.

Despite this, the underlying AI business model remains far from settled. Scrutiny of OpenAI’s finances suggests that even the largest AI firms may struggle to grow revenues quickly enough to cover their enormous compute and infrastructure costs, especially if capital markets tighten. With early signs that the broader AI boom is slowing, doubts are rising about whether today’s aggressive load forecasts will ever translate into sustained, billable demand.

There is notable pushback. In Virginia, multiple parties have challenged Dominion Energy’s IRPs for leaning heavily on data center-driven growth. The Piedmont Environmental Council’s testimony shows that residential bills in 2039 could vary by about $100 per month depending on whether the projected data center load actually materializes, while independent reviews describe Dominion’s assumption that data centers will consume more than 80 percent of commercial sales and over half of total sales by 2038 as “improbable.”

Elsewhere, commissions are experimenting with ways to reinsert uncertainty into planning. In Oregon, proceedings on Portland’s fast-growing data center cluster emphasize the need for explicit high, reference, and low load cases. PacifiCorp’s 2025 IRP goes further, explicitly excluding certain large data center loads from its main retail forecast on the premise that those customers will procure and pay for their own dedicated resources—a move enabled by recent legislation on “large new loads” in Oregon and Utah. Kentucky’s Public Service Commission is using tariff design as a brake: expert testimony supports an “Extremely High Load Factor” rate for mega–data centers, paired with collateral and exit fees, as a precondition for approving new infrastructure built on anticipated demand. In California, staff and intervenors in IRP proceedings have begun citing research and journalism on AI data centers’ bill and blackout risks as they question how much speculative AI load should drive long-term procurement.

These cases show a regulatory field in tension. On one side are utilities and financiers eager to convert AI projections into concrete assets and regulated revenue streams—on the other are advocates, analysts, and some commissions insisting that AI load must be treated as contingent.

Can we avoid a stranded, fossil-heavy grid? Or is this the beginning of a fossil-fueled AI boom?

Throughout all of these developments, AI and data center loads are being invoked as an inevitability in order to justify a rapid, capital-intensive expansion of largely fossil-based energy infrastructure. Financial markets increasingly recognize the bubble-like features of this expansion, even as private equity and infrastructure funds use the AI story to revalue fossil assets and consolidate control over electricity systems. Within utilities, generous assumptions about data center realization rates and long-term AI demand are being embedded into planning models, transforming speculative forecasts into obligations to build plants, lines, and pipelines whose costs will fall on ratepayers for decades.

The question is not whether AI will use more energy—it clearly will—but who bears the risk that its most exuberant projections prove wrong. A more cautious, genuinely transitional approach would treat AI load as deeply uncertain, require robust high–medium–low scenarios, and insist that large data center customers shoulder a meaningful share of the financial risk through tariffs, collateral, and off-balance-sheet procurement. It would treat efficiency, demand management, and genuinely low-carbon generation as preconditions rather than afterthoughts. Above all, it would resist the seduction of inevitability that turns contingent corporate strategies into excuses for locking in fossil-heavy infrastructure under ambiguous allusions to “innovation” or “competitiveness.”

Arizona likes to tell a simple story about itself—a desert that willed a high-tech future out of dust and sunshine. Semiconductor fabs rise out of the Sonoran scrub, data lines pulse under the sand, and articles now rightly point out that “Arizona is at the center of two of the biggest technological transformations of the decade: the semiconductor boom and the rapid rise of AI-driven data centers.” Until recently, there has been a reassuring narrative in the air—that after decades of offshoring and financial crisis, the state is finally building things again.

Behind the headlines about the CHIPS and Science Act, record foreign investment, and an AI data center boom is a different kind of project. A massive restructuring of land, water, and public institutions to absorb risk on behalf of some of the most powerful firms on earth. What is being built in Arizona is not just a cluster of fabs (and data centers). It is a way of governing territory that relies on subnational agencies, utilities, universities, and developers to make the desert “investable” for capital that can leave as quickly as it arrived. I will focus solely on semiconductor development and leave data centers for another time.

This essay offers a brief introduction to my recent research, including my open-access article: “Techno-Statecraft and Industrial Strategy: Semiconductor Development in Arizona,” published in Contemporary Social Science.

Part of the argument I want to make is that what is often celebrated (or derided) as the “return” of U.S. industrial policy looks, on the ground, less like a centralized plan and more like a fragmented (but coordinated) campaign to remake particular places as platforms for high-tech accumulation. Arizona is a pretty good example of this model par excellence. To understand what is at stake, we have to situate today’s AI and semiconductor boom within the state’s longer history of speculative growth, and then watch how that history is being folded into the day-to-day work of building fabs, pipelines, and roads for companies like Intel and TSMC.

Silicon Desert stories

The national story starts in Washington. In 2022, Congress passed the CHIPS and Science Act, promising more than $50 billion in subsidies and tax credits to re-shore semiconductor manufacturing and reduce dependence on East Asian supply chains. The law is routinely framed as a sovereign response to the “rise of China,” proof that the U.S. state has rediscovered its industrial muscles after decades of neoliberal neglect.

But federal legislation is only the preface. The CHIPS Act does not build a single access road, wastewater plant, or cooling tower. Those are the purview of state agencies, municipal planners, water utilities, and development boards who have to make the basic arithmetic of high-tech production work in specific places. In the U.S., that means working through a state apparatus that is fragmented by design. Authority over land use, utilities, permitting, and taxation is dispersed across levels of government and specialist agencies. Industrial strategy, such as it is, becomes a patchwork of experiments and improvisations rather than a single coherent program.

Arizona has been unusually aggressive in turning that patchwork into an advantage. Since the mid-2010s, institutions like the Arizona Commerce Authority (ACA), the Greater Phoenix Economic Council (GPEC), and Arizona State University have converged around a shared anticipatory posture. They have streamlined permitting, prepared industrial parks in advance of firm commitments, and aligned workforce programs with projected labor needs. They have also learned to perform for federal and corporate audiences, marketing Arizona as indispensable to national security and technological sovereignty.

This is what I call techno-statecraft: the work of assembling state power through infrastructural provisioning, institutional adaptation, and coalition politics in order to produce “investable” territory. Readers of this newsletter will no doubt be familiar with all of its implications. Rather than a single developmental blueprint, techno-statecraft is a conjunctural formation—a way of aligning public capacities and private imperatives in the name of resilience, innovation, or security. In Arizona, it has coalesced around semiconductors.

To see how, we need to rewind.

A long history in speculative growth

A nation which does not expand is marked for decay… When opportunities for expansion present themselves they must be taken advantage of at once or the opportunities may not come again.¹

Arizona did not wake up one morning and discover industrial strategy. Its semiconductor moment sits on top of a century of experimental development models, speculative booms, and institutional inventions that trained the state to treat land, infrastructure, and external capital as raw materials for growth.

The pattern was visible as early as the 1910s, when reports of oil seeps triggered a flurry of petroleum speculation. Promoters sold stock by promising that Arizona—wedged between Texas and California—would share in their hydrocarbon wealth. The oil never came; the wells were largely dry by 1918. What remained, however, was a political and business class accustomed to thinking of territory as an object of promotion and financial leverage.

That basic logic resurfaced in the Cold War. From the 1950s onward, Phoenix was transformed from a peripheral desert city into a node in a dispersed defense-industrial geography. Federal housing policies swelled the metropolitan population, while Pentagon contracts and presidential directives encouraging industrial dispersal drew electronics and aerospace firms into the region. Motorola opened a federally backed semiconductor lab in 1949. By the early 1960s, manufacturing output had increased nearly tenfold, and firms like Goodyear Aerospace, General Electric, and later Intel and IBM, embedded Phoenix in national electronics supply chains.

This new growth was not spread evenly. Many of these plants operated as branch facilities with limited local supply chains and externalized decision-making. Jobs and fixed capital arrived, but procurement and higher-end activities remained elsewhere. By the 1980s, local elites found themselves displaced by multinational executives whose allegiances were global rather than regional. Inter-municipal competition over water and tax base intensified.² The same period saw the rise of speculative real estate as Phoenix’s dominant economic engine: deregulated savings and loans funnelled billions into risky land deals, culminating in catastrophic collapses and taxpayer-backed bailouts.

The response was not to abandon speculative development but to retool it. Business leaders and politicians turned to high-tech manufacturing as a more respectable face of the same growth model. Organizations like the Greater Phoenix Economic Council were created to coordinate recruitment of tech firms and polish the region’s image as a “business-friendly” innovation hub. After the 2008 housing crash wiped out hundreds of thousands of jobs, this strategy took on new urgency. The Arizona Commerce Authority was established in 2011 as a public-private agency charged with centralizing economic development, deploying incentives, and courting globally competitive industries.

By the time CHIPS arrived, then, Arizona had decades of experience in mobilizing land, infrastructure, and political access to attract mobile capital. What changed in the late 2010s was the scale and geopolitical framing of the opportunity. Semiconductors promised not just jobs or tax base, but a role in the national drama of technological rivalry with China. That promise gave state and regional actors a powerful narrative to attach to their longstanding growth machinery.

How to make a territory investable

The work of techno-statecraft is rarely spectacular. It looks less like a dramatic policy announcement and more like a series of negotiated decisions about pipes, parcels, and bond issuances. Intel’s long-standing presence in Chandler and TSMC’s arrival in North Phoenix make this process a bit more concrete.

Chandler was an early laboratory. leveraging the ecosystem built around Motorola, Intel began operations there in the 1980s and has expanded its Ocotillo campus through multiple generations of fabs. Over time, the city’s planning apparatus was effectively reshaped around the firm’s needs. Chandler pursued free-trade zone status, extended water and wastewater networks, and built subsidized utility corridors to serve the campus. As Intel prepared for new production lines, it secured up to $7.9 billion in CHIPS funding—nearly $4 billion of that for its Arizona operations. The city, in turn, embarked on major capital outlays for water and wastewater systems, road upgrades, and energy infrastructure aligned with Intel’s growth trajectory.

By 2024, Intel’s facilities drew roughly 15 percent of Chandler’s water supply, even as the city’s overall capital program for water infrastructure more than tripled compared to 2019.³ Planners responded by passing ordinances that require large industrial users to self-fund water access beyond a certain threshold—an attempt to contain the fiscal exposure of building bespoke infrastructure for a single mega-customer. But the fundamental asymmetry remains as Chandler is deeply invested in Intel’s presence, while Intel retains the ability to reallocate production across its global footprint.

North Phoenix illustrates a newer, more expansive variant of the same logic. In 2020, TSMC announced that it would build a fabrication facility in Phoenix. That initial $12 billion commitment has since ballooned into more than $165 billion in planned investment, making it the largest foreign direct investment package in U.S. history. Federal support has been substantial—$6.6 billion in CHIPS subsidies, plus tax credits—but the deal rests just as heavily on state and municipal undertakings.

From the outset, Phoenix, ACA, and GPEC moved to prepare what would become a sprawling science and technology park in the city’s far north. The city pledged over $200 million for roads, sewers, and water systems; expedited land-use approvals; and pursued a Free Trade Zone designation to secure customs and tax advantages for TSMC and its suppliers. In 2023, Phoenix committed an additional $300 million to reviving the long-dormant Cave Creek water treatment plant, boosting capacity by 6.7 million gallons per day—theoretically enough to serve 25,000 households—in order to free up potable supplies for industrial use and enable large-scale real estate development around the fabs.

The Arizona State Land Department, closely linked to ACA, has complemented these moves by auctioning and leasing state trust lands surrounding the TSMC campus. Proceeds are channelled into infrastructure and workforce training directly tied to semiconductor expansion. University research parks have launched chip-focused training programs while suppliers like Applied Materials, ASM, and LCY Chemical have been courted to build adjacent facilities, knitting the area into a vertically integrated ecosystem.

Territoriality of North Phoenix Technology and Real Estate Expansion Complex

Combined, these interventions look less like isolated “incentives” and more like an infrastructural choreography. Water plants, roads, zoning overlays, university curricula, and land auctions are aligned to pre-empt bottlenecks and reassure firms that Arizona can shoulder the physical and institutional risks of expansion. This is techno-statecraft in practice—not a single plan, but a series of coordinated moves that transform a patch of desert into a landscape formatted for high-tech production.

The question is not whether this is skillful. It is. The question that matters for most people is…

Who carries the risk?

The conventional way of celebrating Arizona’s semiconductor boom is to tally jobs, GDP contributions, and headline investment figures. That accounting rarely asks how risks are distributed across firms, governments, communities, and ecologies.

At the most basic level, cities like Chandler and Phoenix are taking on long-term liabilities to support fabs whose productive lifespans and future profitability are uncertain. They issue bonds for water and wastewater plants, expand road networks, and reconfigure zoning to accommodate industrial campuses and their real estate halos. These decisions lock in infrastructures that will require maintenance and upgrading long after initial subsidies expire. All of this is not to mention the long-term impacts and speculative overbuilding of suburban sprawl that has long characterized Phoenix’s growth.

Resource risk is layered on top. In a semi-arid region already grappling with over-allocated rivers and intensifying drought, dedicating large and growing volumes of water to fabs is a political choice. Intel’s 15 percent share of Chandler’s water portfolio, and Phoenix’s commitment to repurpose the Cave Creek plant for industrial demand and associated growth, effectively shift hydrological risk away from firms and onto public systems and, ultimately, residents. The technical language of “reclaimed water” and “efficiency” often obscures the blunt fact that some uses are being prioritized over others, now and in the future.

Fiscally, the model builds in asymmetry. Corporations negotiate site-specific deals, secure federal subsidies, and benefit from state-built infrastructure. If market conditions or corporate strategies change, they can scale back investment or shift production elsewhere, as Arizona learned when Motorola gradually disassembled its Phoenix operations in the 1990s and 2000s. Municipal governments, by contrast, are left with sunk costs and debt obligations denominated in concrete and pipe.

There is also a democratic dimension. Many of the key decisions that shape Arizona’s semiconductor landscape are made through quasi-public or public-private bodies—the ACA, GPEC, technology councils—whose accountability is far thinner than that of elected councils. State audits have already raised concerns about opaque spending and elite courting, even as these organizations position themselves as stewards of a new “strategic” economy. Infrastructure-led growth advances through extra-democratic channels precisely because speed and certainty are central selling points to mobile capital. That is not a side effect—it is part of the pitch.

To be clear, the point is not that industrial strategy is inherently undesirable, nor that semiconductor manufacturing should somehow revert to the pre-CHIPS status quo. The point is that the current model of techno-statecraft—coalition-driven, infrastructure-intensive, framed by geopolitical urgency—is being built with remarkably little public debate about its long-term implications. Industrial strategy is being treated as a technical problem of capacity and competitiveness, when it is also a political problem of territorial reorganization and risk distribution.

A brief contrast helps clarify this. The very firm now reshaping North Phoenix, TSMC, emerged from a different trajectory in Taiwan. There, semiconductor development was anchored in a state-led strategy centered on national science parks, with utilities and land-use policies organized to secure firms’ growth as a matter of national priority. That model had its own profound social and ecological costs—from farmland conversion to water conflicts—but it rested on a more centralized institutional architecture. Arizona’s model, by contrast, is fragmented and coalition-driven, relying more on a web of agencies, utilities, and developers to do the work of translation between federal ambition and local landscapes.

It’s not that one model is better than the other in the abstract. It is that techno-statecraft is not a neutral or inevitable response to technological change. It is a political choice about how to organize state power, infrastructure, and territory around certain industries and narratives. Those choices can, in principle, be made differently.

For now, Arizona’s leaders are doubling down for what is surely a massive boon to state economic growth. They have opened a trade office in Taipei, signed supply-chain agreements with Taiwan’s Bureau of Foreign Trade, and secured a CHIPS R&D Flagship designation for Arizona State University. The state’s development boards convene industry executives and diplomats as matter-of-factly as earlier generations hosted land promoters and oil men.

Nevertheless, the fabs rising over the desert are real. So are the jobs they bring. But so, too, are the pipelines, bonds, and political arrangements that make them possible. High-tech industrial strategy has become the latest chapter in a much longer story of speculative growth that continues to this day. The massive growth in suburban sprawl is a testament to the continuation of this model, but also its fragility.

References

1. Phoenix Gazette, circa 1900. Quoted in Luckingham, Bradford. 1989. Phoenix: The History of a Southwestern Metropolis. Tucson, AZ: University of Arizona Press: 48.

2. Shermer, Elizabeth Tandy. 2013. Sunbelt Capitalism: Phoenix and the Transformation of American Politics. Politics and Culture in Modern America. Philadelphia, PA: University of Pennsylvania Press: 306–307.

3. Intel. 2024. “2023-24 Corporate Social Responsibility Report.” https://csrreportbuilder.intel.com/pdfbuilder/pdfs/CSR-2023-24-Full-Report.pdf; Also see: City of Chandler. 2020. “2020-21 Annual Budget: ‘The Future’s in Sight’”; City of Chandler. 2023. “2023-24 Annual Budget: ‘Innovation at Work.’”

Comparing Cisco and Nvidia

At the height of the dot-com boom, Cisco Systems briefly became the world’s most valuable company. Its routers and switches were seen as indispensable to the internet, and its stock rose nearly fortyfold in the 1990s. When the bubble burst, demand collapsed, carriers defaulted under massive debts, and Cisco’s market value plunged by almost 90%. Investors, pension funds, and workers bore the losses. Yet the infrastructure endured. The glut of idle “dark fiber” from the crash eventually lowered bandwidth costs and supported the growth of broadband and the cloud. Cisco’s trajectory revealed how speculative booms can destroy financial capital while leaving lasting material legacies and a powerful firm intact.

Two decades later, Nvidia occupies a comparable position in the artificial intelligence surge. Originally known for gaming chips, its GPUs became the essential hardware for training large AI models—controlling more than 98% of the AI accelerator market by 2023. By 2025 its valuation had climbed past $4 trillion (about 14% of U.S. GDP). As concerns about a bubble mounted, the financial press drew constant parallels to Cisco:

• Financial Times, Aug 13, 2023: “Nvidia circa 2023, Cisco circa 2000”

• Morningstar, December 7, 2023: “Nvidia 2023 vs. Cisco 1999: Will History Repeat?”

• Financial Times (again), Feb 13 2024: “AI hype has echoes of the telecoms boom and bust”

• Bloomberg, March 11, 2024: “Nvidia Vs. Cisco — A Bubble by Any Other Metric”

• The Wall Street Journal, June 22, 2024: “Nvidia Is No Cisco, but It Is Getting Expensive”

Even the New York Times weighed in when Nvidia briefly surpassed Microsoft as the world’s most valuable company. Its rise has been driven by a twentyfold increase in data center revenues since 2020, fueled by orders from hyperscalers such as Microsoft, Amazon, Google, and Meta. Like Cisco in its era, Nvidia is cast as the “picks and shovels” of a digital gold rush. But its ascent has also been reinforced by state intervention—subsidies, export controls, and executive orders that frame AI infrastructure as a national security priority.

These comparisons are often invoked as warnings, but they also show how speculative bubbles are structured. The dot-com bust was not only the collapse of equity-funded start-ups; it was tied to the implosion of debt-laden telecom infrastructure, where Cisco played a central role. Today’s AI start-ups remain unprofitable (and may never be), yet Nvidia is highly profitable, its growth fueled not just by chips but by an unprecedented expansion of global infrastructure—data centers, energy, and land use—on a scale far larger than the fiber buildouts of the 1990s. If the AI bubble deflates, the financial fallout could be severe, and while Nvidia itself may weather the storm (as Cisco did), the systemic and ecological consequences could be even greater.

The sprawling Cisco campus in San Jose, CA, 2025. Photo by author.

Cisco and the Internet Buildout (1997–2002)

When people recall the dot-com bubble, they often think of quirky websites, speculative start-ups, and the collapse of a nascent internet culture. What receives less attention is the enormous infrastructure built during this period and the companies that supplied it. At the center stood Cisco Systems. By 2000, Cisco controlled more than 70% of the global market for high-performance networking equipment and, for a moment, was the most valuable company in the world, with a market capitalization exceeding half a trillion dollars (roughly 4% of U.S. GDP).

Cisco’s rise cannot be explained by private ingenuity alone; it was propelled by a broader political-economic shift. The Telecommunications Act of 1996, the first major overhaul of U.S. communications law in six decades, was marketed as deregulation but functioned as industrial policy. By privatizing the internet backbone and dismantling barriers between local and long-distance carriers, phone and cable companies, and telecom and internet services, it rewrote the rules to channel capital into fiber, switching, and long-haul infrastructure. The Act, presented as serving “the public interest,” opened new markets and spurred unprecedented private investment.¹ Yet rather than fostering lasting competition, as its boosters claimed, it accelerated consolidation. Mergers among major telecommunications firms soon concentrated control of the internet’s backbone in the hands of a few corporations—an outcome industry insiders had anticipated. As Network World observed in 1995, “[u]ltimately, the Internet will boil down to a few big providers with bilateral agreements on how to interoperate, service and support one another’s networks.”²

Cisco quickly emerged as a major beneficiary. Its revenues surged as carriers expanded the internet backbone with debt-fueled investment, amplified by stock market enthusiasm. Between 1996 and 2001, U.S. carriers spent more than $500 billion on new networks, justified by the oft-repeated claim that internet traffic was doubling every hundred days. Though more myth than reality, this figure shaped expectations of limitless demand. Every new mile of fiber required routers and switches, and Cisco was the dominant supplier. Its reputation for reliability and innovation reinforced its position, while its advertising sought to make visible what was otherwise hidden infrastructure. A 2000 commercial evoked its tag line “empowering the internet generation,” depicting diverse global settings while asserting that virtually all internet traffic passed through Cisco’s systems.

Beneath this commanding position, however, lay fragility. Cisco outsourced much of its production to contract manufacturers such as Flextronics and relied heavily on component suppliers like Corning and JDS Uniphase. More critical still was its dependence on downstream carriers such as WorldCom, Global Crossing, and Qwest, which borrowed heavily to finance their buildouts. That wave of debt-financed bandwidth expansion most closely foreshadows today’s debt-financed push for computing power in AI data centers. As an article in Wired observed in 1998, “Qwest is operating under an if-you-build-it-they-will-come vision. Bandwidth restrictions, the company believes, have held back development of all manner of innovation.” Cisco reinforced this dynamic through vendor financing. Its subsidiary, Cisco Capital, extended billions of dollars in credit to telecom firms and start-ups, booking these loans as revenue even when buyers lacked cash. By 2000, roughly 10% of Cisco’s $20 billion revenues came from such financing. The practice inflated sales in the short term but tied Cisco’s fortunes to the solvency of customers already burdened by massive debt.

By 1999 it was clear to many insiders that dot-com start-ups were failing to generate sustainable revenue. Investor enthusiasm cooled, and debt-laden carriers saw growth slow. As lock-ups expired in early 2000, insiders across dot-com firms sold more than $43 billion in shares—twice the volume of the previous two years. The unraveling accelerated as WorldCom collapsed in an accounting scandal and Global Crossing, after laying vast undersea cables, went bankrupt. Within a few years more than twenty major telecom companies had failed, leaving the sector burdened with over a trillion dollars in unrecoverable debt. Cisco was pulled into the crisis, writing off nearly $900 million in bad loans by 2001.³ Its apparent boom had been underpinned by vendor financing and customer borrowing, a pattern resembling what economist Hyman Minsky described as the slide from hedge finance into speculative and Ponzi finance.⁴

The fallout was severe. Cisco announced a $2.2 billion inventory write-down and laid off 8,500 employees. Revenues stalled, and its share price fell from nearly $80 to under $10 by late 2002—an 88% decline. Investors lost hundreds of billions in market value. Pension and mutual funds that had treated Cisco as a core holding in tech-heavy indexes were hit hard. Vanguard’s U.S. Growth Fund saw its value fall sharply because of its large Cisco position. Vanguard’s U.S. Growth Fund dropped sharply, and public pension systems across the country reported weakened funded ratios. Years of litigation followed, with little real consequence or remedy. The collapse reverberated through the labor market as well, with more than half a million telecom jobs disappearing between 2001 and 2002.

Yet bubbles rarely leave only destruction. While financial capital was destroyed, fixed infrastructure remained.⁵ Billions had been sunk into fiber optic cables, much of it sitting idle as “dark fiber.” This apparent excess eventually drove down bandwidth costs for years, enabling broadband expansion, the rise of streaming, and the growth of cloud computing. The bust was both a disaster and a foundation. It also accelerated industry concentration. As smaller players liquidated, larger firms consolidated control of the digital economy’s backbone. Cisco anticipated this shift. In 2001, just after reporting heavy losses, it predicted that as these “networks of networks” (i.e. cloud) expanded, enterprises and service providers would standardize on only a handful of vendors. Two or three companies, it argued, would likely emerge as strategic partners.

The legacies of the dot-com boom and bust extended beyond unused cables. Many of today’s senior executives built their early careers at Cisco, carrying its practices into the next generation of firms. Cisco itself remains significant. Despite years of financialization through buybacks and acquisitions, by 2022 its revenues and workforce were more than double their 2001 levels. Its trajectory is illustrative of how speculative buildouts typically unfold: favorable policies create conditions for rapid expansion, narratives of exponential growth attract capital, suppliers scale to meet imagined demand, and collapse follows. Yet the wreckage leaves behind both financial losses and durable infrastructure that forms the basis for future growth. The dot-com and telecom crises fit this pattern, and their networks of networks prepared the ground for the cloud. By the early 2020s, a new story of limitless demand was already taking shape within the cloud—not around internet traffic, but artificial intelligence.

Nvidia and the AI Buildout (2021–2025)

In the span of just a few years, Nvidia transformed from a niche graphics and gaming company into the defining firm of the AI era. Breakthroughs in large language models in 2022 and 2023 suggested that the future of work, communication, and national security would depend on ever larger systems with unprecedented computing power. Once known mainly for gaming chips, Nvidia now supplied the GPUs that trained trillion-parameter models. By 2024 its market value topped $1 trillion, and by 2025 it surpassed $4 trillion, making the company the emblem of the AI boom.

Nvidia’s dominance in AI computing rests on a dual advantage. Its GPUs are optimized for the parallel processing needed to train massive models, outpacing competitors such as AMD and Intel. Equally important, the company developed CUDA, a programming framework that became the standard environment for machine learning. Because porting CUDA-based code to alternative hardware requires rewriting complex libraries, the framework locked developers into Nvidia’s ecosystem. Over time, Nvidia hired more software engineers than hardware designers, ensuring that its chips were paired with performance-enhancing tools. This ecosystem strategy gave the company an unrivaled moat. By 2023 Nvidia controlled more than 98% of the data center GPU market, with its Blackwell line alone accounting for two-thirds of quarterly data-center sales.

Data center revenue rose from under $6 billion per quarter in 2020 to more than $147 billion by Q3 2025, accounting for nearly 90% of total revenue. Gross margins hovered around 73%, with operating margins close to 60%. Free cash flow exceeded $25 billion each quarter, supporting stock buybacks of more than $60 billion. Yet this success rests on a narrow customer base. Nearly 40% of data center revenue came from just two buyers—widely believed to be Microsoft and Meta—and about half from hyperscalers overall. The remainder was buoyed by sovereign-backed projects in Europe, the Middle East, and Asia, along with U.S. government procurement.

Nvidia’s Revenue by Segment, 2020–2025

Industrial policy has once again reinforced this dominance, but in a different form. The CHIPS and Science Act directed billions in grants and loans toward domestic semiconductor research and fabrication, while recent executive actions under Trump 2.0 have declared AI infrastructure critical to competitiveness—fast-tracking new data centers and energy projects under a sweeping deregulatory agenda. The state is not simply enabling markets; it is reorganizing them, prioritizing the expansion of compute capacity through massive infrastructure investments. Where the Telecommunications Act of 1996 created markets for carriers and equipment vendors, today’s interventions restructure energy, land use, and permitting systems to accelerate the AI buildout.

The scale of capital expenditure is staggering. Nvidia executives estimate that global AI infrastructure spending could reach $3–4 trillion by the end of the decade. A single hyperscale data center, costing as much as $60 billion, could yield $35 billion in chip sales for Nvidia. Such figures fuel the conviction that the company can sustain valuations exceeding even those of Microsoft or Apple.

Nvidia has built a portfolio of investments in AI startups and infrastructure firms—including CoreWeave, Inflection AI, Mistral, AI21 Labs, Wayve, and Figure AI—aimed at extending its GPU dominance across the ecosystem. The strategy serves two purposes: first, to ensure these companies use Nvidia hardware, locking in chip demand; and second, to gain early insight into emerging workloads that inform future chip design and software platforms. Many of these firms, however, remain unprofitable and financially strained. CoreWeave, for example, reported nearly $1.9 billion in revenue in 2024 but posted losses of more than $860 million, weighed down by over $8 billion in debt and quarterly interest payments above $250 million. Growth is also slowing: after five quarters of triple-digit expansion in 2023 and 2024, revenue growth dropped to 56% year-over-year by mid-2025. Profitable applications beyond enterprise search and cloud services are still uncertain. Corporate deployments of generative AI tools show limited productivity gains, while governments and defense agencies increasingly appear to act as backstops.

The physical buildout of data centers is also magnifying these stakes. Training AI models requires enormous GPU clusters, often numbering in the hundreds of thousands, and the facilities that house them consume staggering amounts of electricity and water. Data centers are spreading into rural areas, industrial zones, and even former steel mill sites, reshaping landscapes in search of cheap power and available land. In Northern Virginia, home to the world’s largest concentration of data centers, grid operators warn of surging demand that could overwhelm the regional power system. In Arizona, cooling requirements have already sparked conflicts over water rights leading to successful resistance against data center construction.

While data centers could sit idle if demand slows, much like unused fiber-optic cables, their economics are far more resource-intensive. Unlike fiber, they require continuous inputs of electricity and cooling, often supported by new gas or nuclear plants built specifically for their needs. Utilities are already planning dozens of new gas facilities to meet projected demand, potentially locking in emissions for decades while passing costs on to average ratepayers. To accelerate construction, environmental protections are being rolled back, leaving communities with higher bills, more pollution, and contested land use. Nvidia’s success is thus tied directly to ecological and infrastructural burdens borne locally. These costs—actual and potential—could make the AI buildout even riskier than the telecom overbuild of the 1990s.

The Structure of High-tech Bubble Economies

Comparisons between the dot-com bubble and the current AI boom are proliferating, but they are often framed too narrowly as stories of firms and markets. What is at stake is the political economy of infrastructure—the ways states, regulations, and financial systems shape speculative expansion. In the late 1990s, start-up mania was only one part of the story. As the Cisco case shows, the more destabilizing crisis was the telecom crash, when carriers such as WorldCom and Global Crossing amassed nearly $1 trillion in debt to build fiber networks. This debt-fueled expansion was enabled by the Telecommunications Act of 1996, a deliberate act of techno-statecraft that privatized the internet backbone, reinforced monopoly power, and spurred speculative infrastructure investment. When carriers collapsed in 2001–2002, the shock rippled through banks, pension funds, equipment vendors, and workers. To speak only of a dot-com bubble obscures the entanglement of policy, finance, and infrastructure that made the downturn systemic.

Viewed through Nvidia’s rise, today’s AI frenzy carries familiar risks but on a far larger scale. America’s leading tech firms are expected to spend nearly $400 billion this year on AI infrastructure, with global investment projected to exceed $3 trillion by 2028. Independent GPU cloud providers like CoreWeave are financing rapid buildouts through large credit facilities, while Wall Street has created an $11 billion debt market using Nvidia chips as collateral. At the same time, AI labs are taking on “cloud debt” through long-term compute contracts that assume exponential demand will continue. Nvidia, like Cisco two decades ago, has become the indispensable bottleneck, its valuation sustained by the belief that computing demand will rise without limit. That narrative has unleashed staggering sums of capital, yet the downstream picture is far less secure. Chipmakers may appear to be the picks and shovels of this boom, but much of the activity in AI labs risks becoming a money trap. Even some optimists concede that, even if AI delivers its most ambitious promises, many investors are still likely to lose heavily.

The geopolitical landscape further distinguishes Nvidia’s moment from Cisco’s rise. Geopolitics today is not only about rivalry abroad but also domestic struggles over land and infrastructure. Data centers now cluster in rural areas, consuming vast amounts of energy, water, and land while sending their benefits to distant metropolitan centers and shareholders. This uneven geography of extraction and reward underscores how foreign policy is inseparable from domestic territorial politics. Cisco’s ascent took place during the optimism of the 1990s unipolar moment, when U.S. liberalization was promoted as a universal model just as the outsourcing of manufacturing was accelerating. States and municipalities competed with foreign sites for investment, while Washington assumed rivals would eventually converge. Nvidia’s growth, by contrast, unfolds under open rivalry with China, framed by techno-nationalism, export controls, and the treatment of infrastructure as a strategic asset. Local resistance to data center expansion is increasingly being overridden by state legislatures stripping municipalities of permitting powers—a domestic echo of securitized competition abroad.

It is important to emphasize that the Telecommunications Act of 1996 was not simply “deregulation,” but a state policy designed to shape markets. The liberalization of telecommunications was a form of techno-statecraft—a strategy to mobilize capital and export U.S. infrastructural control worldwide. Through the WTO’s Basic Telecommunications Agreement and a broader privatization agenda, Washington pressed other nations to dismantle state monopolies and open procurement to U.S. firms. Cisco’s equipment and American fiber-optic cable became the backbone of global connectivity in the late 1990s, embedding U.S. dominance into the architecture of the internet. This was industrial policy disguised as deregulation, reinforced by the Clinton administration’s 1997 Framework for Global Electronic Commerce, which codified minimal regulation, private-sector leadership, and open markets as global standards. It was an optimistic moment when the state extended markets outward, confident the world would follow.⁶

Nvidia’s success, by contrast, is bound up with defensive industrial policy aimed at maintaining a lead over China. The CHIPS and Science Act channels billions into semiconductor research and fabrication, while export controls restrict the sale of advanced GPUs abroad, effectively enrolling Nvidia into U.S. national security strategy. Under Trump, this logic has intensified in extractive form, with environmental safeguards rolled back, national energy emergencies declared, and public lands to be reclassified as AI-ready zones to fast-track data centers and power plants. The Department of Energy has even designated Cold War sites such as the Pantex Plant in Texas for rapid AI-energy buildouts. Executive orders essentially frame AI infrastructure as a matter of emergency and export strategy, casting oversight as an obstacle to competitiveness. Where Cisco once rode a wave of liberalization to global expansion, Nvidia advances in a climate of securitized competition, with industrial policy aimed less at opening markets than at securing them.

This entanglement also exposes Nvidia to the vulnerabilities of its supply chain. Its most advanced GPUs are designed in California but fabricated by Taiwan Semiconductor Manufacturing Company (TSMC) in Taiwan, tying the company’s fortunes to one of the world’s most volatile geopolitical flashpoints. Taiwan’s security now sits at the center of U.S.–China rivalry, and Nvidia’s reliance on TSMC raises the stakes of that confrontation. Pelosi’s 2022 visit to Taiwan, staged with TSMC leadership, underscored how semiconductor production links industrial and foreign policy. Nvidia’s role as one of TSMC’s largest customers highlights this mutual dependence, while TSMC’s new Arizona fab—backed by CHIPS Act funds—will produce Nvidia’s Blackwell chips. If Cisco’s global spread reflected Washington’s confidence in unchallenged dominance, Nvidia’s trajectory is tethered to a fragile security environment that makes its industrial strategy inseparable from high-stakes diplomacy.

The fragility lies not only in geopolitics but also in the concentration of demand. Cisco’s revenues depended on a few carriers whose bankruptcies rippled through equipment markets. Nvidia faces a similar risk: much of its income comes from a narrow set of hyperscalers, sovereign funds, and utilities. If their investments slow or prove uneconomical, the shock could cascade across the sector. Unlike fiber, which retained at least some value after the crash, GPUs depreciate quickly as new generations displace old ones. The result could be stranded assets in the form of overbuilt, energy-intensive data centers with few alternative uses.

The question is not whether the AI boom will end, but how its fallout will be managed. The stakes reach beyond investor portfolios to ecological strain, territorial conflict, and geopolitical crisis. Speculative technology booms are never just about exuberant markets; they reflect how states engineer demand, how narratives mobilize capital, and how infrastructures distribute risk. Cisco’s bust left financial wreckage but also a durable networked legacy. Nvidia’s moment may leave deeper systemic costs, with infrastructures that are harder to repurpose and more entangled in geopolitical tension.

If you’re here for more bubble, talk, please consider looking at my other articles on this:

References:

1. See Aufderheide, Patricia. Communications Policy and the Public Interest: The Telecommunications Act of 1996. Guilford Press, 1999.

2. Cooney, Michael, Adam Gaffin, and Ellen Messmer. “Internet Surge Strains Already Shaky Infrastructure: Who Will Manage the ’Net’s Commercialization?” Network World 12, no. 14 (April 3, 1995): 1; 67. Also see Couper, Elise A., John Hejkal, and Alexander L. Wolman. “Boom and Bust in Telecommunications.” Economic Quarterly 89, no. 4 (October 1, 2003): 1–24.

3. Starr, Paul. “The Great Telecom Implosion.” American Prospect, September 8, 2002. Available at: https://www.princeton.edu/~starr/articles/articles02/Starr-TelecomImplosion-9-02.htm.

4. Minsky, Hyman P. May 1992. “The Financial Instability Hypothesis.” Working Paper No. 74: 6–8.

5. Perez, Carlota. Technological Revolutions and Financial Capital: The Dynamics of Bubbles and Golden Ages. Cheltenham, England: Edward Elgar Publishing, 2002.

6. For more on the 1996 Act and its political economic developments, see McChesney, Robert W. Rich Media, Poor Democracy: Communication Politics in Dubious Times. Baltimore, MD: University of Illinois Press, 1999.

On November 5, 2025, OpenAI CFO Sarah Friar suggested in a public talk that she hoped the federal government would help support future AI infrastructure investments, remarks that many interpreted as a call for a government “backstop” on the company’s roughly $1 trillion in planned data-center deals (think “too-big-to fail” logic). After a wave of critical coverage and market anxiety, Friar walked back the wording and CEO Sam Altman went further, posting that OpenAI neither has nor wants any government guarantees for its data centers, insisting that taxpayers should not bail out failed companies. The comments were aimed at securing “frontier chip” technology the company sees as instrumental in pushing ahead (see some background into this issue in a previous post). Yet, his assertion of private-sector responsibility, offered at a moment when AI stocks like Nvidia and Palantir were selling off on fears that AI spending has outrun profits and market stability, underscores a core tension in AI finance. The sector rests on enormous, interlocking infrastructure commitments and circular investment flows, coupled with an ongoing search for someone else to hold the downside risk (that’s all of us).

Share

Boom or bubble—the real question is who pays?

Across the world, banks are financing the physical architecture of artificial intelligence—rows of energy-hungry data centers, chip fabrication plants, and new corridors of power and water to keep them running. The scale is staggering. A single facility can draw as much electricity as a mid-sized city. Yet even as they accelerate the build-out, banks are already seeking protection from the risks of the very boom they have helped create.

In September 2025, Deutsche Bank published a research note titled “AI Bubble Bubble Bursts.” The report suggested that fears of an AI bubble had themselves popped. Google searches for the term had plunged, media coverage had cooled, and investors had apparently regained confidence. The authors argued that bubbles are nearly impossible to define and that hype often softens into another phase of expansion rather than a crash. They conceded that AI’s growth faces constraints—mounting costs, limited energy supply, and the slow pace of corporate adoption—but framed these as technical challenges rather than structural contradictions. The underlying message was reassuring: the boom is rational, sustainable, and self-correcting.

That framing is less a diagnosis than a performance of confidence. It casts anxiety as a market sentiment that can be measured and managed, converting uncertainty into data. By treating volatility as a sign of health, the bank naturalizes a system that must keep lending, investing, and building even when its foundations are already trembling. The thick veil of economic realism is more of a financial reflex. One must not lose sanity. At least in some part, an institution must believe in the future it is leveraged against.

How banks are manufacturing stability (for now) in the AI bubble economy

Deutsche Bank has lent billions to data-center developers serving clients like Microsoft, Amazon, and Alphabet, and is now looking to hedge that exposure. It is reportedly shorting a basket of AI-linked stocks, arranging a synthetic risk transfer (a deal where the bank keeps the loans but buys protection on the riskiest slice), and preparing, via its asset-management arm DWS, to sell about €2 billion of its own data-center assets. The aim is not to exit the AI economy but to stay deeply embedded in it while shedding as much downside risk as possible.

Synthetic risk transfers and securitizations are what really make this balancing act workable. In these deals, investors such as pension funds, insurers, and hedge funds agree to absorb a defined share of potential losses in exchange for yield. Banks shift credit risk into capital markets, freeing space on their balance sheets for new lending while keeping the client relationships, fee income, and appearance of safety. Each new layer of “protection” supports another round of expansion, and many regulators now treat these instruments as ordinary tools for managing capital requirements. Yet the International Monetary Fund estimates estimates that since 2016 more than $1 trillion in assets have been synthetically securitized in this way and warns that data gaps make it difficult to see where that risk ultimately ends up.

Other banks are moving in the same direction. Germany’s Aareal Bank completed its first €2 billion synthetic risk transfer this year, and UniCredit Bulbank shifted €2.1 billion in loan risk to the Dutch pension manager PGGM. Toronto-Dominion (TD) Bank executed one of the first North American transfers backed by data center exposures. A report by TD likens today’s data center buildout to past railway and telecom booms and describes their crashes as “digestion periods,” a framing that reassures investors by normalizing those downturns while obscuring how losses were actually borne by ordinary savers and workers.

This architecture supports the ongoing build-out of AI infrastructure despite public fears of a bubble. In effect, deals like UniCredit’s with PGGM and Deutsche’s co-lending with the Canada Pension Plan Investment Board pull long-term retirement savings directly into the financing stack that sustains the data center economy. Pension and insurance capital becomes the cushion that lets banks keep lending to increasingly leveraged projects. The Bank of England recently warned that AI-related infrastructure will require trillions in investment over the next five years, much of it financed through debt. At the same time, the same banks funding these projects are quietly exporting their risk to the very institutions meant to safeguard the public’s future.

Risk management for banks means risk exposure for your retirement savings

In market language, this is called efficiency. In political terms, it is displacement. Long-term, capital-intensive assets—data centers, transmission networks, energy contracts—are being converted into streams of tradable exposure. Future rents and power purchase payments are bundled into synthetic securities that promise steady yields to investors who may never see the infrastructure they finance. Land, water, and energy systems are reorganized to serve the requirements of these contracts, with the built environment increasingly shaped by where credit can be placed rather than by what communities need.

I’ve been arguing that the crucial question is where this risk actually ends up. When banks use synthetic risk transfers, the counterparty on the other side of the trade isn’t some speculative day trader. It is more often a pension fund, an insurance company, or a sovereign fund managing retirement savings and public reserves. In other words, the liabilities of the AI build-out are being laid on top of the balance sheets that are supposed to guarantee income in old age. Dutch healthcare workers, Canadian public employees, teachers and civil servants across Europe and North America become indirect backers of AI infrastructure when their pension managers buy into these structures. What looks like a safe, diversified fixed-income allocation may in practice be a thin slice of concentrated risk in a sector whose economics are not yet proven.

This changes who absorbs the shock if AI infrastructure underperforms or is rapidly rendered obsolete. An overbuilt or stranded data center might be written down on a bank’s books, but the first losses will often fall on the pension vehicle that sold protection or bought a subordinated tranche. Banks keep the origination fees, advisory roles, and reputational benefits of financing “innovation,” while the downside is scattered across diffuse pools of retirement capital. Regulatory frameworks treat this as sound practice. Risk has been “transferred,” capital has been “freed,” and the banking system appears more resilient. In reality, the exposure has moved from institutions designed to take credit risk to institutions supposedly designed to provide long-term security—and this isn’t even getting into the risks of 401(k)s.

The reassuring narratives of the AI bubble leave all of this out. The real issue is not whether an AI bubble can be statistically proven, but how its continuation is being underwritten by people who never chose to invest in it. By fixating on search trends and investor sentiment, Deutsche Bank’s report turns a structural question about who funds and who absorbs the costs of AI infrastructure into a passing question of market mood. In reality, the tools of risk management are being used to extend a speculative cycle, with pension funds, retirement systems, and quite possibly taxpayers acting as the backstop. And if what follows the boom is politely described as a “digestion period,” we should be clear about whose futures are being chewed up and broken down in that vat of speculative volatility.

To call this “structural violence” is to name this transfer of risk accurately. The AI boom proceeds as a project of accumulation secured in part by the same savings that should protect people from insecurity. Banks present their hedging and capital relief as prudence, and for a small, wealthy slice of the system, it is. But for those whose retirement savings sit behind these trades, prudence looks a little different. It would mean not having their retirement tied to the fate of data centers they will never see, in places they will never visit, built to service a technological frontier whose winners are already well insulated from loss.

See related articles:

Industrial policy is not decreed by a central committee or a cadre of autonomous bureaucrats. In the U.S., it is crafted through sprawling coalitions linking firms and financiers to “experts,” policy advocates, staffers, and even self-proclaimed “activists” who translate corporate needs into public policy. What we rarely see—and what this article only begins to map—is who these actors are and how their coordinated efforts shape policy across municipal, state, and federal scales.

An unscheduled visit by Kirsten Sinema…

Chandler’s Price Corridor was not supposed to host a new data-center campus. City staff and segments of the public had pressed for uses that fit the district’s established “high-wage, knowledge-industry” vision, and skeptics argued that a data center—low headcount, heavy utility footprint—would undercut that model. In my own research, Chandler’s planners told me they had a system in place to reserve the city’s water for higher-value uses which, at the time, included Intel’s CHIPS-era expansion (what they called “quality of life” water). Given data centers’ low headcounts and heavy utility demand, many proposals have been rejected. These proposals shifted to Mesa and west Phoenix, yet Chandler’s Intel-driven water-system upgrades remain a coveted draw for data-center developers. Into this fraught, locally bounded land-use debate stepped Kyrsten Sinema on October 16, 2025, appearing during the council’s unscheduled public comment period to advocate for what she called an “AI data hub.” No longer a senator, she came, rather, as the founder and co-chair of the AI Infrastructure Coalition which she says was working with the Trump administration.

No longer a senator, Sinema led with her new credentials, the “founder and chair of the AI Infrastructure Coalition,” describing AI as a “national security imperative” and positioning Arizona as one of five states “best poised to lead.” The city’s 2016 Price Corridor plan, she argued, pre-dated today’s technology stack and therefore the corridor needed to adapt: “Back nine years ago, the general plan didn’t even mention AI or AI data centers or AI hubs.” What looked like a conventional “data center” to local planners—“a regular data center that holds cat memes and your 7-year-old bank accounts”—was, in her framing, categorically different from AI-class compute: “AI data centers are the Lamborghini when compared to the old data centers’ Pintos.”

She tethered this reframing to labor and competitiveness. The corridor’s existing employers, including several financial-services call centers, would see rapid automation: “In less than five years, those will be obsolete, completely replaced by AI and AI companies.” The proposed hub, by contrast, would anchor the infrastructure that future firms require: “They create a massive amount of compute power—the very power that AI companies of tomorrow need in order to relocate here in Arizona.”

The rhetorical pivot came in her close, where local zoning gave way to federal authority. Citing a prior Planning Commission vote to advance the project, Sinema warned that rejecting the plan would only delay the inevitable:

“That site will remain empty if not developed under this plan—until federal preemption comes, which is coming, and it is coming soon.”

What does a suburban growth-district fight over use mix and job quality have to do with national AI industrial policy? Sinema’s remarks convert a municipal siting dispute into a question of preemption, grid capacity, and ‘AI-ready’ infrastructure—the preferred language of a new coalition politics in the Trump era.

What is the AI Infrastructure Coalition?

A search for the AI Infrastructure Coalition (AIC) yields almost nothing. Digitally, it’s a ghost. Though, what little appears on these sorts of organizations’ websites that do exist hardly conveys the extent of what they actually do.

The AI Infrastructure Coalition (AIC) first surfaced publicly through a formal response to the federal AI Action Plan request for information, submitted in March 2025. The twelve-page memo—filed “on behalf of the AI Infrastructure Coalition” by the policy consultancy CO2EFFICIENT—lays out an agenda to retool U.S. infrastructure policy around AI-driven energy and compute demand. Its language reads like an industrial blueprint put forward by many others in the same RFI: whole-of-government coordination, new “AI Development Zones,” interagency reforms to accelerate power, water, and cooling infrastructure for advanced data centers, etc. Beneath the rhetoric of innovation and national security, the filing sketches a new regulatory perimeter that treats compute as the organizing logic of territorial development.

The AIC describes itself as “a national coalition of companies and experts representing the digital infrastructure ecosystem,” with a deliberately expansive remit spanning “AI platform providers, microchips, server hardware, data-center equipment, software developers, data-center operators, and power-management services.” Yet it names no members, only sectors. The coalition’s public face, Kyrsten Sinema, apparently chairs and speaks for it at municipal hearings and recent industry conferences. Her own framing tracks the AIC memo: AI is a “national security imperative” requiring “a whole-of-government approach,” led by a handful of “strategic states.” Arizona, she argues, is one of them. This logic reframes local siting disputes as pieces of a geopolitical campaign to preserve U.S. technological dominance.

The coalition’s memo centers on infrastructure bottlenecks—grid congestion, permitting delays, and inadequate cooling—and prescribes policy tools keyed to corporate investment timelines. It urges the federal government to extend FERC Order 2023’s interconnection reforms and implement Order 1920-A on long-term transmission planning—positions hyperscalers have endorsed (context I cover in a Trump 2.0 AI industrial strategy post). It calls for federally brokered “collaboration” between utilities and large-load customers to scale tariff structures that enable new business models, steer siting toward available capacity, harmonize data-center efficiency standards, and expand DOE-led R&D on cooling and heat-reuse. Most ambitiously, it proposes a dedicated executive-branch “policy home” for data-center policy, consolidating oversight now dispersed across local utilities, environmental agencies, and planning departments.

Based on the available evidence, this agenda functions as preemption-by-policy already taking hold in multiple states (I’ll cover this in a forthcoming post). It advances a framework that narrows municipal authority—limiting cities like Chandler from setting land use or fully accounting for water and energy costs. The “AI Development Zone” language (akin to proposals from the Institute for Progress) reprises familiar instruments—enterprise zones, opportunity zones, and other deregulatory regimes that fast-track private capital under the banner of national competitiveness. By grafting compute infrastructure to national-security rhetoric, the AIC recasts industrial lobbying as patriotic duty. In Sinema’s framing, resisting a local data-center project becomes tantamount to opposing America’s bid for AI “global dominance.”

AIC’s affiliates tell us more about how industrial strategy is being crafted today.

Behind the AIC is a constellation of expert networks and trade groups organized by CO2EFFICIENT, each aimed at a specific bottleneck in the AI buildout. CO2EFFICIENT’s team includes ex-BP legal/public-affairs leadership, a House Energy and Commerce deputy counsel, FERC and congressional energy staff veterans, a leading advocate for large electricity buyers, and AWS’s former senior corporate counsel on AI/data-center regulation, among others. In short, these are specialists in utilities, pipelines, gas, transmission, and hyperscale data-center policy.

The AI Supply Chain Alliance (AI-SCA) frames itself as an effort to secure the “resilience and competitiveness” of the AI manufacturing base. Its membership reads like a supply-chain schematic—Dell Technologies, NVIDIA, Supermicro, Schneider Electric, Shell, Trane Technologies, Flex, ENEOS, Prometheus Hyperscale, and others—spanning chip design, server hardware, energy systems, and industrial cooling that directly support hyperscale AI facilities. The alliance’s own RFI response to the AI Action Plan describes AI as critical national infrastructure and warns of permitting and power barriers that are “stymying the digital infrastructure ecosystem.” The argument mirrors AIC’s—AI capacity should be treated as strategic infrastructure deserving of streamlined approvals and targeted incentives.

The Electricity Customer Alliance (ECA) operates as the energy-facing branch of this ecosystem. It represents “large electricity customers” and counts Google, General Motors, Freeport-McMoRan, RWE, and Steelcase among its members (ECA Members). Its senior advisor Tom Hassenboehler—a partner at CO2EFFICIENT—has testified before Congress on “Powering AI: Examining America’s Energy and Technology Future.” ECA advocates for “customer-centric” grid modernization, a euphemism for ensuring hyperscale data centers can secure dedicated transmission capacity, flexible tariffs, and interconnection priority. In practice, the group functions as a policy arm for power-hungry AI clients, embedding their load demands within ongoing federal grid reforms under FERC Order 1920.

If ECA covers electricity, the Liquid Cooling Coalition (LCC) handles heat. Founded by Intel, Shell, ENEOS, Supermicro, Vertiv, Submer, and Ada Infrastructure, the LCC promotes liquid-cooling standards and R&D programs for high-density AI servers. Its filings for the AI Action Plan argue that advanced cooling is essential to meet the thermal demands of AI systems. The group pushes for DOE and national-lab partnerships to expand cooling technology deployment, reinforcing the notion that AI’s physical intensity is an engineering inevitability rather than a policy choice.

Finally, the Natural Gas Innovation Network (NGIN) links the digital buildout to the fossil fuel sector, building on the Differentiated Gas Coordinating Council (DGCC). Public member listings include Sempra, Southern Company, Williams, BKV Corp., Kuva Systems, and Project Canary. The network is “dedicated to enhancing the transparency of the natural gas value chain, fostering technological innovation, and reducing emissions,” positioning gas as a cleaner, more traceable firm-power option for new AI loads. Remarks by Brad Crabtree—then at DOE, now an ExxonMobil executive—at a DGCC workshop hosted by Tom Hassenboehler and CO2EFFICIENT and posted on DOE’s website promote a U.S. energy policy tailored to AI demand, legitimizing a subsidized public–private alliance—“differentiated” gas, carbon capture, and hyperscalers—to secure gas-fired power for data centers under a decarbonization veneer.

These coalitions form an integrated policy stack: AI-SCA builds, ECA powers, LCC cools, and NGIN fuels. Each advances a partial solution to the same problem—how to guarantee uninterrupted compute growth under the auspices ‘innovation’ and ‘security.’ Collectively, they constitute a deliberate architecture of techno-industrial power—a compute-industrial complex that rewrites energy and environmental governance around AI’s physical demands.

These coalitions are just a small part of a much larger organizational apparatus.

At the macro level, this coalitional architecture signals a new phase of state-capital entanglement—what I’ve been referring to as techno-statecraft. Unlike earlier state-led industrial planning that at least nominally weighed public benefit against private interest, today’s model is driven by “policy entrepreneurs” like CO2EFFICIENT, which broker coordination across corporate and financial actors. By managing multiple coalitions—AIC, ECA, LCC, AI-SCA, and NGIN—the firm operates as a clearinghouse for capital, translating corporate requirements into national policy language. In practice, this privatizes the infrastructure agenda—drafting rules, shaping regulatory frameworks, and embedding proposals through public comments, legislative briefings, and closed-door negotiations.

These organizations function less as trade groups than as brokers—intermediaries between corporate interests and the state. They translate technical requirements—megawatts, megabytes, and water gallons—into the language of national policy. Their shared grammar of inevitability casts AI expansion as both unstoppable and essential to security, warranting an “all-of-government” mobilization. The result is a quiet but far-reaching restructuring of public authority on the back end of AI growth, where decisions over land, energy, and environmental limits are reframed as throughput problems for industry to solve.

The Chandler case shows that this infrastructure politics is already territorial. When Sinema told city officials that “federal preemption is coming,” she was not making an idle threat. She was articulating a governing logic that treats local planning as an obstacle to be bypassed in the name of so-called “national competitiveness”—now synonymous with corporate profit. The emerging AI industrial regime combines technological determinism with state facilitation. It represents a form of infrastructural empire building in which the state guarantees the conditions for capital accumulation through land, energy, and law, so long as it can be justified as powering AI.

The legacies of extractivism and crisis meet technology

Lithium extraction by foreign firms in Argentina is a visible emblem of global hierarchy: vast salt flats feeding distant battery supply chains while local communities see little return. The new digital extractivism is less visible in this way, but much more consequential—targeting policy and sovereignty as much as minerals. The same logic that probes the ground for lithium now probes the state for guarantees, revenue channels, and regulatory favors, aligning energy, data, and finance to reconstruct patterns of dependency through AI and the language of “innovation.”

Seen through that lens, the latest crisis reads differently. Argentina has recently dominated headlines as the peso slides, reserves turn sharply negative, and bond spreads flash default risk. In September 2025, Treasury Secretary Scott Bessen confirmed that Washington was preparing extraordinary support, a roughly $20 billion “bazooka” that included swap-style facilities and even potential U.S. purchases of Argentine sovereign bonds. Markets rallied on the signal, with bonds jumping more than 10% in a single session, underscoring how quickly this digital-extractive circuitry converts political commitments into financial gains.

Mainstream financial commentary framed this as nothing all that new. The Financial Times argued that “Argentina has a history of losing market credibility, and Milei’s government overplayed its hand by moving too quickly” while suggesting that U.S. backing could buy time until “real reforms” are completed in labor and tax policy. The Wall Street Journal warned that the public is “losing faith in Milei’s free-market revolution” and pressed for more decisive steps toward dollarization. From this vantage point, Argentina is a perennial reform laggard, and external support is a pragmatic stabilizer that gives investors breathing room.

Progressive and heterodox observers offer a different account. Analysts at the Peterson Institute warn that a U.S. rescue may invite new problems by deepening reliance on official creditors while leaving the dollar gap intact. Scholars of global debt politics add that bailouts function as disciplinary technologies that bind debtor states to austerity.¹ Protesters in Buenos Aires echoed this view in mid-2024 with chants of “no al saqueo” as they opposed Milei’s reforms. Critics then extend the argument to the digital realm, where foreign AI investment in energy and data centers resembles older patterns of mineral and agricultural extraction. For them the core issue is not failed reform but a dependency in which creditor stability and (now) high-tech investment are prioritized over social welfare.

Within this debate Milei has sought to shift the terrain. He describes himself as carrying out “chainsaw economics,” promising to slash “the parasitic state” and deliver Argentina as “the freest country in the world.” His administration has paired sweeping decrees and omnibus reform bills with a futurist discourse of technological transformation. Milei insists that austerity today is the precondition for “turning Argentina into an intelligent power,” positioning the country as a future hub for artificial intelligence and data-driven industries.

The crisis is being managed not simply through austerity, but by selling a technological future. Promises of AI hubs and data-driven growth rebrand cuts and institutional rewiring as “modernization,” while channeling external finance into privatized energy, land, and compute infrastructure. In practice, the bailout de-risks these projects for foreign creditors, asset managers, and tech firms. Announcements like OpenAI’s serve as political cover for shifting risk and discipline onto the public, while profits and control accrue abroad. Read in Argentina’s longer trajectory, this is crisis management by external dependence wrapped in techno-optimist alibis.

Dependency structures Argentina’s turn towards authoritarian neoliberalism

Argentina has long lived under the weight of its external constraint. Postwar industrialization relied on tariffs and state intervention, but imported machinery and commodity-based exports kept the economy vulnerable to balance-of-payments pressure. That structural tension set the rhythm for recurring crises. The 1976 coup deepened this dependence by liberalizing trade and finance, and under Carlos Menem in the 1990s, privatization and a one-to-one peso-dollar peg brought short-term stability at the cost of hollowing out state capacity. When global credit conditions reversed after the dot-com bust (2000) and the post-9/11 (2001) slowdown, Argentina’s debt dynamics—already fragile under the convertibility regime—unraveled. The government’s attempt to placate investors through the megacanje, a large-scale debt swap promoted as a show of fiscal discipline to maintain IMF support, backfired. It increased long-term obligations and drained credibility. By December 2001, as capital flight accelerated and fiscal targets were missed, the IMF withheld its next disbursement. Within weeks, the peso peg collapsed, unleashing default, bank freezes, and the country’s worst economic crisis in modern history.

The Kirchner governments that followed (2003–2007, 2007–2015) sought to regain autonomy through debt renegotiation, IMF repayment, and state-led redistribution, briefly easing external pressure during the commodity boom. When prices fell, dependence re-emerged. Mauricio Macri (2015–2019) reversed course, reopening capital accounts and issuing massive foreign debt. The $57 billion IMF program that followed—its largest ever—promised credibility but delivered recession and renewed default risk, proving that re-entering global markets without structural reform only reproduces crisis.

Alberto Fernández’s administration (2019–2023) pursued a centrist stabilization amid pandemic shocks and inflation but drifted into paralysis. By 2023, deficits persisted, reserves were depleted, and investor confidence had evaporated. Into this vacuum stepped Javier Milei, channeling frustration through libertarian populism. On his first day, he issued Decree 70/2023, a 366-article “mega-decree” rewriting regulation across sectors and bypassing Congress. Soon after came the Ley de Bases, an omnibus bill enabling privatizations, loosening oversight, and granting long-term guarantees to large investors. Passed in mid-2024 after fractious negotiations, it formalized his shock program and placed the old architects of liberalization back at the helm of a new, more concentrated experiment in market rule.

Rather than treading an entirely new course, neoliberalism has returned with less scruples. Milei’s Minister of Deregulation and State Transformation, Federico Sturzenegger, helped design the failed megacanje in 2001. His return—alongside other veterans of Menem’s neoliberal experiment, including Guillermo Francos and Martín Menem—underscores how Milei’s so-called libertarian revolution revives the neoliberal playbook. The “chainsaw” rhetoric may be new, but the personnel and policy logic echo the 1990s: faith in markets, external borrowing as reform discipline, and dependence on global finance that once again risked ending in collapse.

This may be why Milei’s government has been able to move so fast. The budget was cut by nearly 30% within months, and deregulation proceeded at a dizzying pace—the libertarian/conservative outlet Cato Institute counted two rollbacks a day. The emerging order resembles authoritarian neoliberalism. Under crisis conditions, executive power concentrates, legislative debate narrows, and reforms advance before opposition can organize. Courts have challenged portions of the decree, especially labor provisions, and mass protests have erupted, yet much of the framework remains intact.

Politically, Milei faces a legitimacy problem. His movement lacks a solid majority in Congress and relies on a pragmatic, transactional fragment of Peronist support (peronismo dialoguista). To compensate, he speaks the language of techno-populism—promising innovation, digital modernization, and a new era of efficiency—to attract finance and tech constituencies while deflecting social unrest. His coalition rests uneasily between global investors and domestic sectors hurt by austerity, exposing the contradictions of a project that weds radical market reform to concentrated executive rule.

Digital extractivism is at the center of Milei’s economic experiment

It’s as absurd as it sounds, AI-driven growth is being positioned as a substitute for science—more precisely, artificial intelligence is being promoted as a replacement for genuine human knowledge-making. From the start of his presidency, Milei has described the dismantling of Argentina’s research institutions as the price of becoming a “global AI hub.” Cuts to the national science agencies—CONICET, INTA, and ARSAT—are framed as modernization. In reality, they have gutted public capacity. Researchers have seen their pay collapse, funding for labs has evaporated, and the country’s most talented scientists are taking side jobs or leaving. A new decree (447/2025) has centralized decision-making in a small, politically appointed board, narrowing research priorities to those that serve private innovation. The government’s message is clear: knowledge matters only when it attracts investors.

Milei uses the language of digital transformation to give austerity a futuristic gloss. In his telling, dismantling the old state makes room for the private sector to lead a high-tech renaissance. AI becomes both the symbol and the strategy. Technology will redeem hardship by attracting foreign investment, creating jobs, and putting Argentina on the map of global innovation. Yet, what he calls “efficiency” and “liberation” amounts to converting the public sphere into a platform for private capital.

The energy sector sits at the center of this vision. Milei argues that Argentina’s traditional grid cannot power the data-intensive future he imagines, so he has turned to nuclear energy and foreign finance. His government is promoting a new small modular reactor designed by Argentina’s Invap but funded largely by U.S. investors. To make the deal possible, the state has started privatizing its nuclear company, Nucleoeléctrica Argentina, selling 44% of its shares while calling it a “nuclear resurgence.” It’s a telling move. The push for AI is used to justify privatization, and privatization is sold as the key to powering AI. The loop between narrative and policy is pretty crude but complete. Yet the tension between ambition and implementation is stark. Despite the nuclear plan, the national atomic agency (CNEA) remains stuck at 2023 nominal budgets, inflation hollowing out its resources.

Nevertheless, Patagonia is cast as the territory of this new future. With its cool climate and open land, officials describe it as perfect for data centers that need vast energy and low cooling costs. To draw foreign firms, Milei’s government passed a sweeping law—the RIGI, or Régimen de Incentivo para Grandes Inversiones—which offers long-term tax breaks and foreign-exchange privileges to investors who spend more than $200 million. The plan is to turn Argentina into a hosting hub for global computation—a place that supplies the energy and the land while control of the data and profits would likely remain offshore.

This agenda took shape in October 2025 with the announcement of Stargate Argentina, a $25 billion project between OpenAI and Sur Energy to build a 500-megawatt data-center complex in Patagonia. The deal, negotiated under the RIGI framework, binds Argentina’s energy infrastructure to OpenAI’s computing empire. Press photos of Milei standing alongside OpenAI executives were presented as proof that Argentina had “entered the AI age.” But the timing of the announcement came just as the Trump administration confirmed its ~$40 billion bailout for Argentina, suggesting that financial rescue and technological partnership are part of the same strategy. The AI corridor provides the political and ideological cover for deeper external alignment.

For Milei, this alignment is the point. The AI narrative allows him to present deregulation and privatization as forward-looking rather than regressive. It signals to U.S. and global investors that Argentina is open for business on favorable terms, even as domestic science and public employment are dismantled. Patagonia becomes a new kind of frontier—not of national development, but of hosting capacity for global (U.S.) platforms. The government’s promise of an AI-driven renaissance is built on selling energy, territory, and regulatory freedom to outside capital. Yet, research institutions are collapsing, the state is shrinking, and those most qualified to lead this “innovation revolution” are leaving. But will this story work? Can it really turn a crisis into a generative narrative? Is the dream of technology salvation amid decline?

Financial rescue shifts risk to the public and power to Washington

Announced in the middle of a government shutdown in Washington, the ~$40 billion U.S.–Argentina package was framed as a vote of confidence in Javier Milei’s reform agenda. Half of the amount comes from direct Treasury and IMF support, and half from a private-sector facility designed to attract investment. Following the announcements, the peso stabilized, bond prices rose, and investors booked quick gains, but revealed how closely Argentina’s economy now moves to signals from Washington. The bailout functioned less as a domestic policy achievement than as a display of U.S. leverage, shifting risk from private investors to the public.

The IMF provides the institutional frame for this arrangement. Its four-year lending program, approved in April 2025, conditions every tranche on fiscal tightening, higher tariffs, and reserve accumulation. The stated goal is macroeconomic stability, but the effect is subordination. Policy is disciplined through compliance reviews that privilege debt repayment, currency defense, and investor confidence over social spending or productive investment. Dependency is re-enforced through the language of fiscal responsibility. Each round of external “support” narrows Argentina’s capacity to set priorities beyond the immediate demands of its creditors.

Behind the technical design lies a clear geopolitical intent. Secretary of the Treasury, Scott Bessent, described the operation as part of an “economic Monroe Doctrine,” a strategy to reassert U.S. influence in Latin America and counter Chinese lending. Trump himself made the political dimension explicit, publicly stating “If he loses, we are not going to be generous with Argentina.” The timing ensured that the injection of dollars would stabilize the peso and lift markets in the run-up to Argentina’s legislative elections, giving Milei a short-term political dividend while anchoring his government to U.S. oversight.

Financially, the beneficiaries are also familiar. The arrangement—roughly a $20 billion dollar-peso swap and a $20 billion private debt facility—creates opportunities for U.S. banks such as JPMorgan, Goldman Sachs, and Citi, and for asset managers like BlackRock, Fidelity, and PIMCO. Rising bond prices and reduced default risk deliver immediate windfalls to these firms, while the new facilities generate fees and collateralized positions tied to Argentine assets. Treasury gains strategic leverage, Milei gains temporary currency stability, and investors lock in upside with minimal exposure.

Over the longer term, the stabilization of Argentina’s financial environment opens the door for U.S. technology and energy interests as noted above. A steadier peso and easier access to dollars favor the build-out of AI-related infrastructure and lithium extraction projects promoted under the RIGI investment regime. As journalist Pablo Calvi observed, the bailout ultimately serves elites on both sides of the equator—banks, funds, and tech corporations—rather than the Argentine public. The package thus continues a historical cycle in which crises are managed through external dependence: short-term confidence for investors, structural vulnerability for everyone else.

Something to learn here?

Although differences abound, the echoes of the 2001 crisis and today’s bailout reveal the same underlying logic—financial stability is pursued by transferring risk from creditors to the public. Each round promises a reset and delivers deeper dependence. Argentina’s current crisis is not exceptional; it is the latest iteration of an enduring pattern in which the external constraint—vulnerability to capital flows and export shocks—triggers actions fiscal retrenchment, institutional restructuring, and political concentration. What distinguishes the present moment is the fusion of austerity with a technological imaginary that casts dismantling the state as the path to digital modernization—a thin cover story that claims a smaller state will unlock a shiny digital future.

The language of AI hubs, nuclear-powered data centers, and “intelligent power” reframes foreign intervention as partnership and turns fiscal discipline into evidence of investor confidence. The narrative satisfies multiple audiences. It reassures U.S. policymakers that Argentina is a strategic ally, signals to global capital that new frontiers of data and energy are open, and offers domestic cover for the contraction of public institutions. Yet beneath the rhetoric, research budgets collapse, scientific talent drains abroad, and inequality widens. The digital future effectively functions as spectacle—an alibi for policies that deepen, rather than escape, structural dependence.

This blend of finance and futurism is not only an Argentine phenomenon. From compute corridors in the United States to AI zones in Europe, tech promises now organize how crises are managed and how states are disciplined. Argentina shows this logic in a clear and concentrated way, where fiscal fragility, geopolitical ambition, and digital aspiration meet.

The links between U.S. finance during the dot-com bust and today’s AI boom in Argentina are telling. The landscape has changed, with AI industrial policy and a more complex global economy reshaping both the shape of crises and the tools used to manage them. To see how this will play out now, we have to look beyond balance sheets and focus on the stories that make these policies seem necessary. In today’s digital capitalism, technology serves not only as infrastructure but as a means of governing through the promise of innovation while perpetuating dependency and control.

References

1. Gill, Stephen. 1995. “Globalisation, Market Civilisation, and Disciplinary Neoliberalism.” Millennium Journal of International Studies 24 (3): 399–423. Also see: Kentikelenis, Alexander E., Thomas H. Stubbs, and Lawrence P. King. 2016. “IMF Conditionality and Development Policy Space, 1985–2014.” Review of International Political Economy 23 (4): 543–82.

During the California Gold Rush of the mid-19th century, tens of thousands of prospectors streamed west in hopes of striking it rich. Most miners never found fortunes and many barely covered their costs. Some struck it rich, but it was the merchants who sold them supplies—picks, shovels, boots, food, and boarding house beds—who earned steady profits. Equipment manufacturers, saloon owners, transport providers, and land speculators often captured more value than the miners themselves. Even financiers entered the scene, extending credit and taking stakes in mining operations, profiting when mines produced and losing heavily when they did not. The Gold Rush is remembered as much for this supporting economy as for the glitter of the metal itself.

The Gold Rush is a useful shorthand for today’s AI boom. Like many observers, I use it to spotlight the key economic actors in a boom-bust system. If AI labs and startups are the prospectors—highly visible yet often unprofitable, burning cash and tied to long-term obligations—then the hyperscale clouds are the merchants, selling compute, storage, and services no matter which lab wins. Utilities and real estate investors, meanwhile, extract rents for scarce power and strategically located parcels. Profits climb further upstream to the firms that control the critical bottlenecks—Nvidia and TSMC, the shovel makers of the moment, capturing extraordinary margins in chip design and fabrication. And just as nineteenth-century financiers magnified boom-and-bust cycles, debt markets today fuel GPU purchases and speculative buildouts.

The rise of AI has led to trillion-dollar valuations and promises of revolutions in work, education, medicine, and security. Yet beneath the headlines lies an economy where value is captured unevenly—by those who provide the tools, platforms, and territory, rather than by the firms building the applications themselves (at least so far). To map this distribution, I start downstream with the model labs and move up the stack—through hyperscale clouds, utilities and landlords, to chip designers and foundries—with finance capital underwriting and shaping the whole system.

The miners burn cash while the merchants, manufacturers, and financiers cash in.

AI labs and independent clouds in the red

The most visible companies in the AI boom are often the least profitable. OpenAI, Anthropic, Cohere, and similar labs dominate headlines with their conversational models and enterprise deals. Yet their financial statements tell a more complicated story.

OpenAI, for example, is estimated to be generating revenue at an annualized pace of around $12–13 billion in 2025, yet analysts project it will also burn roughly $8 billion in cash over the same period. Anthropic faces a similar imbalance, with multi-billion-dollar annualized revenue but heavy operating losses. The gap reflects the staggering cost of compute. Training a new generation model can require investments in the hundreds of millions, while day-to-day inference demands continuous access to fleets of high-end GPUs housed in specialized data centers.

To secure this capacity, AI labs sign multi-year cloud contracts. OpenAI’s widely reported $300 billion agreement with Oracle is the most dramatic example. Economically, these contracts resemble debt—they are binding obligations to pay for compute whether or not the revenue materializes. Call it “cloud debt.” Like financial leverage, it magnifies returns if demand grows, but it can also sink firms if revenues lag behind.

The Financial Times recently reported that an $11 billion debt market has emerged around GPUs themselves. Independent providers borrow against Nvidia chips, using them as collateral for loans. CoreWeave, a fast-growing GPU cloud, has raised more than $7.5 billion in debt financing to fund expansion. In one recent quarter, it posted revenues over a billion dollars yet still recorded a net loss of nearly $300 million, mostly due to interest expenses. Debt enables rapid growth but also concentrates risk in firms whose assets and revenues are tightly tied to volatile AI demand.

In short, downstream labs and independent clouds carry the heaviest financial risk. They finance the system through multi-year contracts and leveraged debt, but the revenues they may generate leak quickly upstream.

Hyperscale clouds as intermediaries and strategists

Above the labs sit the hyperscale cloud providers—Amazon Web Services, Microsoft Azure, and Google Cloud. These companies purchase chips from Nvidia and TSMC, acquire land, secure electricity, and operate vast data centers. They then rent out “compute” to enterprises, startups, and AI labs through a mix of on-demand access, reserved instances, and managed AI services.

Financially, the clouds are on far firmer ground than their downstream customers. In its most recent quarter, AWS posted over $10 billion in operating income. Google Cloud reported nearly $3 billion, at margins above 20%. Microsoft said its cloud division’s gross margin was nearly 70%, although it noted a decline due to the heavy capital expenditures needed for AI infrastructure.

These firms are investing at a historic pace. Alphabet has guided $85 billion in capital expenditures for 2025, while Microsoft and Amazon are on similar trajectories. These budgets cover not only chips but also land, grid connections, fiber backbones, cooling systems, and transmission upgrades. Their profitability depends less on scarcity than on utilization. If customers keep servers busy around the clock, the economics work; if demand softens, depreciation and energy costs can quickly erode margins.

Yet clouds are not simply passive intermediaries. As Semianalysis has noted in its study of Oracle’s strategy, cloud providers actively shape their economics through site selection, data center design, and network efficiencies. Locating a facility near cheap power and abundant land can cut total cost of ownership dramatically. Optimizing internal networking can reduce the price of training large models by hundreds of millions. These strategic choices enable cloud providers to capture more value than their downstream tenants, who face fixed prices for capacity.

UBS analysts claim that “profitability is largely decided by performance.” The clouds that can offer the fastest, most reliable AI infrastructure command premium pricing, because enterprises value latency and throughput. This further entrenches the position of hyperscalers, who have the scale and capital to deliver best-in-class performance. Smaller players and AI labs either aren’t equipped in the first place or struggle to match these sorts of advantages.

Utilities and real estate extract resource rents

As I’ve been writing about extensively, the limits to scaling AI extend beyond silicon to include power, cooling, and water. Each data center is essentially a warehouse of GPUs that must operate continuously under heavy energy and thermal stress.

Utilities are critical in the geography of AI growth. Hyperscale firms contract for power at the scale of hundreds of megawatts, while in Northern Virginia—home to the world’s largest data center cluster—Dominion Energy is committing tens of billions of dollars to new substations and transmission lines to keep pace with demand. In Virginia, regulators have even required that transmission upgrades be completed before new loads may be connected. Similar dynamics are evident in Ohio, where AEP forecasts data center demand rising from ~600 MW in 2024 to nearly 5 GW by 2030, leading regulators to adopt new interconnection tariffs and billing rules. Because utility capital expenditures are typically recovered through regulated rates, part of the cost of private data center and AI expansion may be effectively socialized onto local ratepayers.

Real estate is increasingly strategic in the expansion of AI infrastructure. Parcels with direct access to high-capacity transmission lines, substations, and fiber backbones are in short supply, driving up competition and valuations. Developers and investors highlight proximity to substations and transmission corridors as key factors that lower construction costs and accelerate deployment. Large land acquisitions—such as recent campus-scale purchases in Ohio—are marketed as “rare and valuable” precisely because of these locational advantages.

In particular, specialized data center real estate investment trusts (REITs) have also become consistent beneficiaries of this trend, capturing higher lease rates in power- and land-constrained markets. Large players like Digital Realty and Equinix reported Q2 2025 net incomes of $1.05 billion and $368 million, respectively, though Digital Realty’s figure was lifted by one-time gains rather than recurring operations. Their economics hinge less on software breakthroughs and more on geography and infrastructure—control of scarce, grid-adjacent sites with fiber, cooling, and room to expand—so they monetize AI demand through long-term leases even as tenants shoulder most model and product risk. Hyperscalers may also lease large blocks from data center REITs, shifting some development/financing and delivery risk to the landlord.

While utilities and REITs may not command Nvidia-level margins, they capture value by controlling essential inputs like resources and land. They shape where the AI economy grows, how quickly new capacity comes online, and who pays for the supporting infrastructure.

To the chipmakers go the surplus

At the very top of the stack sit Nvidia and TSMC and similarly situated firms. Together they represent the most striking case of value capture in the AI economy.

Nvidia designs the graphics chips (GPUs) that train and run today’s AI models. In its most recent blowout quarter (Q2 2025), Nvidia reported $46.7 billion in revenue, with $41.1 billion coming from data-center sales alone. Its profit margins are unusually high for a typical hardware maker—more like what you’d expect from a software company. While its chips are both highly capable and in short supply, its success is due to its monopoly over the CUDA software framework, now the default for machine learning and effectively locks developers and organizations into its ecosystem.

TSMC is the contract manufacturer that actually builds the vast majority of advanced AI chips. Its results show how valuable that role is. In Q2 2025, TSMC reported $30.07 billion in revenue with 58.6% gross margin and 49.6% operating margin. Most of its sales came from leading-edge processes which it has a virtual monopoly over allowing it to charge premium prices.

The bottom line is that semiconductor revenues and profits vastly outstrip those of AI applications. While labs generate billions in revenue, Nvidia and TSMC generate tens of billions in profit. The imbalance confirms that the bottleneck at the top of the stack is where the surplus flows.

Private equity and asset managers finance the build-out

Private equity and large asset managers are central to this build-out, supplying development equity, project finance, and private credit in exchange for fees, interest, and upside tied to long leases and power contracts. On the credit side, the new GPU-backed lending structures and headline facilities noted above are quite illustrative. CoreWeave’s $7.5 billion debt-financed expansion was led by Blackstone and Magnetar. These investors effectively convert future compute and lease commitments into cash today—speeding construction while shifting utilization and operating risk onto the independent clouds and tenants that must keep the hardware busy.

On the equity and development side, global asset managers are co-funding campus-scale data centers through joint ventures with REITs and operators as well. Equinix has struck multi-billion-dollar joint ventures (over $15 billion) with sovereign and pension capital—GIC and CPP Investments among them—to expand its xScale hyperscale platform in the U.S., while also creating project-specific joint ventures such as a $600 million partnership with PGIM Real Estate in Silicon Valley. Meanwhile, large buyout firms are tying power and data-center supply together. Blackstone reports more than $70 billion of data-center assets and another $100 billion in its pipeline, and has publicly outlined plans to invest $25 billion in Pennsylvania across data centers and generation to meet AI loads.

Sovereign and public investors in Asia are also channeling capital into this stack. Temasek joined the AI Infrastructure Partnership alongside Microsoft, BlackRock, and MGX, which targets $30 billion+ initially and aims to mobilize up to $100 billion for AI-driven data centers and energy systems—an explicit example of sovereign-backed capital buttressing GPU-centric build-outs. The net effect is that private equity and asset managers convert future compute and lease commitments into cash today, accelerating construction while shifting performance and utilization risk onto operators and tenants, and locking in upstream revenues for chipmakers and power/land owners.

The AI value chain funnels surplus upward, but each layer has a distinct role.

We can certainly widen the scope of actors to include communities, politicians, business coalitions, upstream energy firms, and more. But for now, let’s set those wider entanglements aside and simply recap the map so far:

AI labs and indie GPU clouds (downstream): These are the teams building and running AI products. They often grow revenue fast but spend even faster on computing time and equipment. Because they sign multi-year contracts for capacity, they can end up owing a lot even if usage or sales fall short.

Hyperscale clouds (midstream): Big platforms bundle chips, power, networks, and software and rent them to everyone else. They’re profitable at scale, but only if their giant data centers stay busy. Their main worries are overbuilding, power costs, and slower-than-expected demand.

Utilities and data center REITs (physical enablers): Power companies and data-center landlords control the basics—electricity, grid connections, cooling, fiber routes, and suitable land. They earn steady returns through regulated rates or long leases, and they’re less exposed to whether an AI app succeeds. Their risks are political and permitting delays, not product cycles.

Chip designers and foundries (upstream): The firms that design and manufacture the most advanced chips sit at a bottleneck. Because their products are scarce and essential, they capture the richest profits. If AMD or custom silicon designs from hyperscalers gain traction, Nvidia’s margins could compress. Similarly, if new foundries emerge or geopolitical tensions disrupt Taiwan, TSMC’s pricing power could change.

Private equity, asset managers, and sovereign funds (the capital layer): These investors finance the build-out across all the layers. They turn future rents and power contracts into cash today through joint ventures, project finance, and loans (including ones backed by GPU hardware). They collect fees and interest while most day-to-day operating risk stays with the companies using the assets.

You’ll notice I’ve set aside end-user demand (the usual focus of AI “use case” reporting). That’s because, right now, speculative capital—especially venture capital—underwrites the buildout on the strength of its own demand imaginaries. In this frame, AI labs—the “miners” in the analogy—function as the effective customers, lured by dreams of striking it rich. The main point is that most of the profit ends up where scarcity and long contracts live—chips, power, and prime data-center sites. The capital layer speeds everything up by supplying money, but it also spreads risk downward to the operators that must keep the machines full and the bills paid.

The current AI boom is a rent-seeker’s paradise.

The Gold Rush remade not only fortunes but also landscapes and communities. Indigenous nations faced catastrophic demographic collapse and forced labor regimes; Mexican communities saw land and political standing eroded through lawfare, taxation, and violence. In both cases, scarcity rents from gold (and later land and water) accrued upward, while risk and dispossession were pushed onto subordinated peoples. Mining devastated rivers, forests, and soils, while boom towns sprang up and collapsed with the busts. Many migrants and miners left with little to show but debts, while a handful of merchants, landlords, and financiers accumulated enduring wealth.

What happens after the rush is just as important. The flood of bullion swelled the money supply, loosened credit, and subtly re-priced trade and labor across the Atlantic world—effects that outlasted the rush itself. The AI boom may not scar mountainsides or poison streams (for the most part), but it carries its own forms of extraction—of electricity, land, and labor—that is similarly repricing inputs, rewiring infrastructure, and shifting bargaining power among regions, firms, and labor classes long after the speculative period has passed.

In sum, the AI boom is a layered economy in which profits and risks are distributed unevenly. Money flows into flashy applications and speculative bets, but much of it ends up as revenue for hyperscalers, power providers, and chipmakers. The sustainability of this structure depends on whether AI adoption grows fast enough to justify the “cloud debt” being accumulated today. If it does, downstream firms may eventually capture more value. If it does not, many of them will bear the cost of an infrastructure build-out that enriched their suppliers more than themselves.

AI labs (and venture capitalists), of course, price failure into the model; power-law returns mean a few wins can cover many losses. But that doesn’t mean they absorb all the downside.

The most consequential outcomes won’t show up in valuation charts so much as in rate cases, power contracts, and water rights—where costs are socialized and baked into everyday life. One important area of value extraction and transfer is the public. Communities are already bearing the costs as value flows up the chain. Reports in mainstream outlets are circulating charts like the one below on the rising energy prices, with a substantial driver being data center buildouts.

Average Price of Electricity per Kilowatt-Hour in U.S. Cities, 1976-2025

Following new federal support for AI industrial policy, states are beginning to chart their own paths to accelerate AI development and expand the energy infrastructure needed to sustain it. In the past year, states like Rhode Island, Virginia, and Washington have created task forces to explore AI’s role in government, while California launched a sweeping initiative to deploy generative AI in transportation and tax services. Mississippi ordered agencies to name AI coordinators and set standards for fairness and transparency, and Ohio is requiring every public school to adopt AI policies by 2026. Other states are still sorting out their approach—Colorado, for example, recently pushed back implementation of its AI Act until mid-2026.

I’m working on a broader review of these state-level initiatives, but a recent executive order in North Carolina offers an early glimpse of how these strategies are beginning taking shape. For further background, you can read my two-part series on federal AI industrial policy here:

Community resistance to data centers is colliding with state-level efforts to fast-track AI infrastructure.

In Tarboro, a small town of about 11,000 residents east of Raleigh, a five-hour public hearing earlier this month ended in an unusual decision. The town council voted 6–1 to block a proposed multibillion-dollar data center from regional developer Energy Storage Solutions. The project promised 500 jobs and $11 million in property taxes each year, but it also required enormous energy and water use—500,000 gallons of water a day, according to the developer—and would have transformed 52 acres into a hyperscale campus.

The opposition was vocal. Residents described the project as a threat to Tarboro’s character, environment, and household utility bills. More than 700 people signed a petition saying the facility would “dramatically alter” the community. Some council members admitted they didn’t have enough information to judge the proposal, while the mayor recused himself after acknowledging he couldn’t ignore the views of constituents.

Tarboro isn’t alone. Residents of Apex are fighting a rezoning request for 190 acres near the Shearon Harris Nuclear Plant. A 100MW natural-gas-powered campus was just rejected in another town. And in Mooresville, a $30 billion data center proposal was pulled after officials signaled they would not support it. Communities are repeatedly raising concerns about noise, appearance, energy strain, and modest job creation. Developers counter that data centers bring tax revenue, high-wage tech jobs, and a place in the digital economy.

While residents push back, industry groups and lawmakers are working to limit local control. In several states, legislatures have advanced preemption laws that curb municipal authority over zoning and permitting for data centers and energy projects. These measures shift power upward, weakening the ability of communities to shape decisions about land and resource use in their own backyards. In North Carolina, debate is underway on HB 1002, a bill that would prevent utilities from recovering the costs of grid upgrades for data centers through general ratepayer charges—that households could be left to subsidize the infrastructure needed for massive private facilities. This is something other states have also passed or are considering.

Against this backdrop of local resistance and legislative maneuvering, Governor Josh Stein issued Executive Order 24. The order recognizes that the rapid buildout of AI and data centers will require new substations, expanded water systems, and potentially higher costs for households if those expenses are spread through general rates. It situates AI governance not just in the realm of ethics and innovation but firmly within the contested politics of infrastructure and community control.

North Carolina is positioning itself as both builder and regulator of the AI economy.

On September 2, 2025, Governor Josh Stein signed Executive Order 24. The order creates three new structures meant to put North Carolina at the forefront of artificial intelligence.

First, there is the AI Leadership Council, a 25-member group that brings together state officials, lawmakers, company executives from firms like SAS, IBM, and Pryon, along with university researchers from UNC, Duke, and North Carolina Central. The council will advise on AI strategy, ethics, and workforce planning.

Second, the order sets up an AI Accelerator inside the Department of Information Technology. This office will define standards, set rules for data and intellectual property, and review pilot projects across state agencies.

Third, every cabinet agency now has to form its own AI Oversight Team and submit at least three potential AI use cases within 180 days. That means proposals for things like using AI to speed up licensing, screen eligibility for programs, or build digital assistants to guide residents through state services.

The order also ties AI planning directly to North Carolina’s Energy Policy Task Force. That’s because AI growth depends on expanding physical infrastructure. As I’ve covered extensively in Techno-Statecraft, Data centers and AI campuses use large amounts of electricity and water. Amazon has already announced a $10 billion AI campus in Richmond County, and Duke Energy has pointed to AI data centers as one of the biggest new demands on the grid.

The executive order does more than set ethical guidelines. In fact, its core aims are further afield from ethics. It positions the state as both promoter and regulator of AI infrastructure. From a critical political economy perspective, three dynamics stand out.

Market making by the state. By requiring every cabinet agency to submit pilot projects, the state manufactures demand for AI services inside government. This guarantees vendors proof-of-concept contracts and a steady pipeline of projects. With SAS, IBM, and Pryon executives seated on the Leadership Council, the same firms likely to benefit from these contracts are helping define the rules of engagement. This resembles earlier state-backed technology pushes, where public procurement created markets for private providers.

Utilities and cost allocation. Duke Energy and other utilities are central players in this story. Adding data centers means adding massive new electric loads. Utilities can either require companies like Amazon to finance their own infrastructure or socialize those costs across all ratepayers. Recent North Carolina rate cases and press coverage show watchdog groups already raising alarms about special tariffs for “big-user” loads. The executive order acknowledges the tension but routes responsibility to the Energy Policy Task Force, creating a separate forum where decisions about who pays will play out.

Universities and research pipelines. The Triangle’s universities gain formal roles in shaping standards and piloting projects. UNC has recently launched a School of Data Science and Society, Duke hosts AI and science-and-society initiatives, and North Carolina Central has created new AI institutes. The order positions them as intermediaries, producing research and talent to feed both public projects and private industry. Yet intellectual property and data ownership rules are centralized in the AI Accelerator, which could limit how much remains in public hands.

Workforce and communities. The council is tasked with tracking AI’s labor impacts and recommending retraining strategies. But the representation for labor, consumers, and local communities is pretty thin. There’s only one consumer seat, one workforce board seat, and one local government seat. Without bargaining mechanisms or enforceable obligations, workers are mainly asked to adapt through literacy programs rather than share in productivity gains. The push for corporate AI adoption in schooling is mirrored at the federal level.

Data center disputes expose the divide between corporate benefits and community risks.

The Tarboro fight shows what is at stake in NC and other parts of the U.S. While residents successfully mobilized to stop a $6.4 billion project they feared would disrupt their community, the company has already signaled it will sue to overturn the decision. In these hearings, councils will act more like courts, weighing expert testimony rather than public sentiment. That means local opposition often has limited legal standing.

At the same time, when the state positions itself as both promoter and regulator, it tilts the balance toward rapid adoption. That may help attract companies like Amazon and SAS, but it leaves communities like Tarboro to potentially absorb the consequences. The order’s references to energy and water impacts show that policymakers are aware of the stakes. Whether those concerns lead to real safeguards will depend on how the Energy Policy Task Force acts, how the Accelerator defines “high-risk” uses, and whether transparency mechanisms are enforced. But these things will likely depend on the balance of forces at play—who stands to benefit and who doesn’t, and more importantly, who has power and who doesn’t.

For big firms, the benefits are pretty clear. Amazon secures the infrastructure for its new campus. Duke Energy gains profitable new customers and justification for grid investments. SAS, IBM, and Pryon help shape the standards state agencies must adopt. Universities strengthen their research and training roles.

For residents, the picture is murky. Energy bills could rise if costs are shifted to ratepayers. There’s little sign of HB 1002 moving forward. Water and land use may also strain local resources and jobs may not match the scale of the promises. In these towns, people are learning that the legal system gives them few tools to stop projects once they are in motion.

The question is not only how artificial intelligence will be governed, but how the physical infrastructure that supports it will be built, financed, and distributed across the state. Executive Order 24 sets the rules for the AI economy. The community battles over data centers show how those rules will be tested on the ground.

From Cold War Control Rooms to the Cloud

If today’s data centers and satellite communication hubs embody national security and surveillance—like the NSA’s massive data center complex—then blinking consoles and sprawling control rooms captured the Cold War imagination of computing. In the Cold War United States, this took its first form in the Semi-Automatic Ground Environment (SAGE), a 1950s defense project linking dozens of IBM machines into a continent-wide radar and missile coordination grid—the first truly networked computer system, radiating centralized command and control from the heart of the military-industrial complex.

Across the Iron Curtain, Soviet planners pursued their own cybernetic dream: the National Automated System for Computation and Information Processing (OGAS). Advocates promised a single automated brain capable of “continuous optimal planning and control” across the entire socialist economy. To American observers, OGAS appeared less like a tool of rational coordination than the blueprint for a cybernetic Leviathan. CIA analysts warned of “closed-loop, feedback control” that could allow Moscow to outpace American capitalism. Even Arthur Schlesinger Jr., Kennedy’s liberal consigliere, fretted that Moscow’s machine planning might outpace American capitalism. One Air Force commander put it in starker terms—such a system could be “imposed upon us from an authoritarian, centralized, cybernated, world-powerful command and control center in Moscow.”¹

The irony, of course, is that the nightmare Schlesinger feared was not imposed from Moscow at all. It was built in California. Neoliberal America, not Soviet socialism, produced the global digital infrastructure that now governs daily life and state power alike. What we call the “cloud”—a planetary-scale lattice of semiconductors, networks, data centers, and software platforms—is precisely the kind of totalizing system Americans were taught to fear. It coordinates production and consumption, organizes supply chains, captures and processes vast streams of behavioral data, and increasingly integrates advanced artificial intelligence. In short, it is a distributed apparatus of planning and control—but one linked to corporate profit and national security.

Today, there is no command post or blinking console to dramatize power; control has been diffused into global networks and condensed into the devices we carry in our pockets. Yet the Cold War visions of SAGE and OGAS remind us that computation was never just about calculation—it was about command. By the 1980s, that logic had hardened into economic policy. The question was no longer whether technologies might spill across civilian and military lines, but how national survival itself would hinge on securing the commercial base of electronics, chips, and the minerals that made them possible.

Techno-Nationalism and the Material Base

The phrase “dual-use”—popularized in U.S. Cold War policy to describe technologies with both civilian and military applications—suggests a neat division between the two domains. In practice, much of modern industry has always carried war within it. Railroads were designed to move troops as readily as they carried commuters. Fertilizers such as ammonium nitrate and other nitrogen compounds fed crops but also provided the chemical base for explosives and munitions. From the start, industrial capitalism blurred the line between commerce and conflict. The roots of Silicon Valley, and of the broader computer industry, were likewise seeded by military contracting, beginning with Navy funding for radio research around the Spanish-American War. Missiles, telecommunications, and surveillance systems grew from this foundation and were continuously sustained by defense budgets throughout the twentieth century.

By the late twentieth century, the relationship between military and commercial technology shifted but remained deeply institutionalized. Facing ballooning costs, U.S. defense agencies in the 1980s and 1990s began adopting the “commercial off-the-shelf” (COTS) policy, sourcing widely available electronics for weapons systems rather than commissioning bespoke designs. The logic was pragmatic: Silicon Valley was already producing advanced components at scale, so the military could leverage private innovation rather than duplicate it. But the consequences ran deeper. National security now depended not only on traditional defense contractors but also on the stability and competitiveness of the commercial technology sector itself.

This shift collided with a new anxiety. In the 1980s, as corporate America was looking to undercut its own domestic manufacturing base in a push for more overseas branch plants and outsourcing, Japanese firms were overtaking U.S. firms in semiconductors. Suddenly, what had been treated as consumer goods—memory chips, microprocessors—were recast as lifelines of national industrial survival. A 1984 National Research Council report titled Race for the New Frontier warned that “the U.S. advanced technology enterprise… must be held as one of the country’s most valued objectives.”² The Defense Science Board went further, declaring that U.S. forces “depended heavily on technological superiority in electronics; that semiconductors were the key to such leadership; [and] that this depended on strength in commercial markets.”³

This techno-nationalist turn⁴ also extends to raw materials—the foundation of any nation’s ‘industrial base.’ Due to material shortfalls and restrictions leading up to and during World War II, Cold War planners feared bottlenecks in materials including cobalt, tungsten, and rare earths. Well into the Reagan era, entire programs were devoted to monitoring and stockpiling what were deemed “strategic” minerals.⁵ Decades later, these concerns returned with force. In 2017, the Trump administration issued an executive order naming 35 “critical minerals” essential to both economic growth and military capacity. The United States Geological Survey identified 50, while the Pentagon tracked more than 250, “strategic and critical” materials defined as “those that support military and essential civilian industry” (with a special focus on semiconductor production in Taiwan and Korea), identifying many concentrated in Chinese supply chains.

Seen this way, dual-use is not just about chips or drones. It is also about the material foundations of modern industry—foundations governments treat as matters of national security. Conventional warfare required enormous inputs of steel, fuel, and minerals. But by the late Cold War, defense strategists began to argue that the decisive raw material was shifting. The future of war, it was claimed, would hinge not only on industrial stockpiles but on the ability to capture, transmit, and process information.

In the 1990s, this logic crystallized in what U.S. defense thinkers called a “Revolution in Military Affairs.” Admiral William Owens predicted that networks of sensors, satellites, and algorithms could dissolve the ‘fog of war’ altogether, making victory flow not from sheer firepower but from information itself.⁶ As asymmetric warfare and counterinsurgency replaced conventional battlefields, the raw material of intelligence became data—and the infrastructures for harvesting and analyzing it were increasingly concentrated in Silicon Valley.

Silicon Valley Goes to War (Again)

The techno-nationalist anxieties of the 1980s thus found their proving ground in the 2000s. The dot-com crash had left Silicon Valley adrift—trillions in market value evaporated, startups shuttered, and engineers scattered. Then came 9/11. Federal budgets for intelligence, surveillance, and homeland security exploded, with the Department of Homeland Security alone requesting more than $40 billion by 2005. For firms desperate for cash flow, empire offered a new business model.

Commentators Mark Mills and Peter Huber captured the moment bluntly in City Journal, celebrating the “trillion-dollar infrastructure of semiconductor and software industries, with deep roots as defense contractors.” To them, dual-use surveillance was the embodiment of American exceptionalism. At home, these “technologies of freedom” would guarantee security and liberty; abroad, they would “destroy privacy everywhere we need to destroy it.” Their conclusion was striking in its candor: “it will end up as their sons against our silicon. Our silicon will win.” U.S. empire now marched openly under the banner of Moore’s Law.

The War on Terror didn’t rescue Silicon Valley wholesale—Google’s advertising engine and Apple’s iPhone did that—but it gave a crucial subset of firms and engineers a steady lifeline. In 1999, the CIA created In-Q-Tel, a venture arm designed to channel private innovation into the intelligence community. One of its first investments was the geospatial startup Keyhole, whose EarthViewer platform was later deployed in Iraq and ultimately acquired by Google to become Google Earth. In 2003, Palantir emerged directly from counterterrorism demand, with U.S. intelligence agencies funding and piloting its data-fusion platforms before they became indispensable in Iraq and Afghanistan and later in commercial markets. A year later, DARPA’s Grand Challenge poured federal resources into autonomous vehicle experiments, cultivating teams and technologies that would ultimately seed Google’s self-driving program and Uber’s foray into automation. Together, these initiatives provided legitimacy, stability, and cash during the lean years between the dot-com bust and the rise of Web 2.0 consumer platforms, anchoring a defense-driven pipeline of talent and technologies that would later spill into the commercial mainstream.

What began as stopgap lifelines in the wake of the dot-com crash hardened into structural dependence, and by the 2010s, these relationships had been institutionalized. The Defense Innovation Unit, headquartered in Mountain View, funneled billions into startups working on drones, AI, and cyber defense. Congress codified the fusion with the 2018 CLOUD Act, compelling U.S. cloud providers to share data stored abroad. In 2022, the Pentagon signed a $9 billion Joint Warfighting Cloud Capability contract with Google, Amazon, Microsoft, and Oracle, effectively deputizing them as military infrastructure. Eric Schmidt, Google’s former CEO, captured the reality with unusual candor: “Because of the way the system works, I am now a computer scientist, businessman, and an arms dealer.”

The Chip Wars

If the militarization of the cloud revealed how digital networks were woven into the architecture of war, semiconductors show the same condition at a deeper level. Chips are not merely components of everyday devices; they are the substrate of twenty-first-century power. Every missile system, drone swarm, and AI model is only as capable as the processors inside it. Silicon Valley had already ceased to be a civilian sector occasionally tapped for defense. By the 2010s, it had become a critical extension of the national security state—and semiconductors stood at the center of this fusion.

First, indispensability. Chips underpin every frontier of military technology. The CHIPS Act of 2021 and the CHIPS and Science Act of 2022 together committed more than $50 billion to domestic capacity, with Intel receiving billions specifically for “secure enclave” chips designed for Pentagon use—an example of how industrial subsidies also function as defense procurement. This logic is also behind the U.S. government’s recent $8.9 billion stake in Intel to support its floundering foundry program. By contrast, Taiwan’s TSMC has long dominated the foundry model that underwrites much of the U.S. military-industrial complex. Yet Taipei remains cautious about overcommitting TSMC to defense production, wary of eroding its “silicon shield”—the geopolitical leverage it derives from being indispensable to global chip supply.

Second, concentration. For decades, U.S. dominance rested on this sort of a hybrid model—corporate leadership at home (Intel, Qualcomm, Micron) paired with offshore fabrication in Japan, South Korea, and above all Taiwan. This distributed supply chain maximized efficiency but created acute vulnerabilities. China’s Made in China 2025 program underscored the fragility, as Beijing poured subsidies into fabs and design houses to close the technological gap. Although China’s pursuit has been bumpy, the prospect of it catching up in advanced nodes set off alarms in Washington, where the lesson of the War on Terror was simply that national security could not be disentangled from corporate innovation.

Third, alliances. While the U.S. response to Chinese advances (and COVID-19 bottlenecks) was nothing short of mobilization, allies were also pulled into orbit while looking to secure their own industrial base. Japan lured TSMC into Kumamoto with billions in subsidies; South Korea rolled out tax relief for Samsung and SK Hynix; Europe launched its own Chips Act; and India pledged $10 billion to jump-start domestic production. These initiatives also coalesced into the “Chip 4 Alliance,” a geopolitical bloc aimed at fencing off China from the commanding heights of semiconductor production.

Arizona offers a rather compelling case on these points. TSMC’s $165 billion Phoenix megafab is not just about supply-chain “resilience.” NVIDIA’s GPUs, built on TSMC manufacturing processes, are increasingly embedded in military AI workloads, from Navy procurement of H100 accelerators to autonomous systems developed by defense startups like Anduril. Xilinx’s FPGAs—long fabricated by TSMC—show up in satellites and radios used by Raytheon and Northrop Grumman. Geographically, Arizona sits between the bomb-making complexes of New Mexico and the aerospace hubs of Southern California. Its desert is becoming a new frontier for a fusion of commercial electronics and military power.

Between 2021 and 2025, announced semiconductor investments exceeded $630 billion in the U.S. alone. The global scale of investment is also staggering. Samsung’s new Pyeongtaek campus nears $100 billion—second to TSMC’s campus in Arizona. Taiwan also continues to invest heavily, with TSMC planning $100 billion in new projects. These are not just factories; they are massive territorial projects reshaping entire regions, from the deserts of Arizona to the Korean peninsula.

Map of current and announced semiconductor investments 2021 to 2025

Data sources: SEMI 2023, GMFUS 2023, SIA 2025

The Condition of Modern Techno-Statecraft

Charles Wilson, the former president of General Motors and later Secretary of Defense, famously captured mid-century corporate nationalism with his remark: “What’s good for General Motors is good for America.” Decades later, amid intensifying U.S.–China tensions, this logic resurfaced in a new register when private equity founder and current U.S. ambassador Thomas Barrack declared in 2017: “What’s good for America is good for the world.” Both statements reflect a persistent conflation of corporate power, national interest, and global order. As Tony Smith observed, the dominance of American techno-capital has produced a global market structure in which “a handful of giant First-World oligopolies operating at or near the frontier of scientific-technical knowledge” dictate terms to “small-scale Third-World producers far from that frontier.”⁸

Seen in this light, the empire Silicon Valley has built makes the Cold War imagination of Soviet planning appear almost quaint. Semiconductors—and the cloud infrastructure they underpin—are no longer just another industry; they form the architecture of twenty-first-century power. Control over chip supply chains now equates to control over commerce, surveillance, and war alike. What began as ad-hoc procurement strategies and techno-nationalist subsidies has since hardened into a regime of authoritarian techno-statecraft, where defense budgets, corporate R&D, and industrial policy converge into one seamless system of power projection.

Lewis Mumford foresaw this trajectory in 1967 when he described the rise of “authoritarian megatechnics,” warning that a “dominant minority will create a uniform, all-enveloping, super-planetary structure, designed for automatic operation…” that would subordinate individuals to vast systems of control.⁹ That warning no longer reads as dystopian speculation but as a description of our present. The cloud that stores family photos is the same one that processes drone surveillance. The chips powering video games also drive missile guidance. Convenience and coercion now run on the same circuits.

The war in Gaza shows these dynamics with brutal clarity. Google and Amazon’s $1.2 billion Project Nimbus contract arms Israel’s Ministry of Defense with cloud and AI capacity, while Microsoft, IBM, HP, and Palantir provide the infrastructure for military data, surveillance, and population control. Worker resistance has flared—Google employees forced the company to back away from Project Maven, and Amazon and Google staff protested Nimbus—but these efforts have barely dented the machinery of militarized computing.

And the fusion is not limited to foreign battlefields. The very tools of empire are increasingly turned inward. Palantir’s $30 million ImmigrationOS contract with ICE is designed to consolidate sensitive federal data to streamline deportations, raising alarms about civil liberty violations and mass surveillance. Lawmakers have also pressed Palantir over its role in building a government-wide “mega-database” linking IRS and agency records—“a surveillance nightmare” in the words of its critics. Meanwhile, Google, Amazon, and Microsoft quietly supply cloud services to ICE and CBP, facilitating border surveillance even as their own employees protest.

As consumer and enterprise markets plateau, Big Tech’s profitability hinges on public subsidies, defense contracts, and a steady pipeline of militarized demand. The new industrial strategies around AI and semiconductors make this dependence explicit—fast-tracking energy- and land-intensive buildouts, channeling public funds into cloud and chip monopolies, and using export controls and subsidies to project U.S. corporate power abroad. This dependence tightens the knot between corporations and the state, ensuring that the military-industrial complex is not just a customer but a guarantor of profitability. And the deeper this entanglement grows, the harder it becomes to imagine a technological future driven by public needs rather than by the imperatives of empire.

References

1. Gerovitch 2010, cited in Vincent Mosco, To the Cloud: Big Data in a Turbulent World (Routledge, 2015).

2. National Research Council, The Race for the New Frontier: International Competition in Advanced Technology (New York: Simon & Schuster, 1984): 6.

3. National Research Council, U.S.-Japan Strategic Alliances in the Semiconductor Industry: Technology Transfer, Competition, and Public Policy (National Academies Press, 1992): 85.

4. OECD, Strategic Industries in a Global Economy: Policy Issues for the 1990s (OECD, 1991).

5. See reports of the President’s International Materials Policy (Paley) Commission: Materials Policy Commission, Resources for Freedom Volume 1: Foundations for Growth and Security, 1952. In addition to recommendations for foreign and industrial policy, the final reports also advocate for continued public investment in technology as it claims that “the strongest and most versatile single resource in the fight against scarcities of materials is technology.” (132)

6. William A. Owens, “System of Systems,” Armed Forces Journal, January 1996.: 47; On Owens’ views on RMA, see William A. Owens, High Seas: The Naval Passage to an Uncharted World (Naval Institute Press, 1995). Owens would go on to serve as president, COO, and vice chairman of the defense contractor Science Applications International Corporation (now Leidos) as well as various positions in other telecommunications and technology companies.

7. Schmidt, Eric. “The Age of AI.” presented at the Stanford ECON295/CS323 Course, August 15, 2024. A recording can be found here.

8. Tony Smith, Globalisation: A Systematic Marxian Account (Brill, 2005): 173.

9. Lewis Mumford, The Myth of the Machine: Technics and Human Development (Harcourt, Brace & World, 1967): 3.

In the late 1990s, the dot-com boom was defined by euphoria. Stock tickers surged as day traders chased internet IPOs, venture capitalists bankrolled websites selling everything from books to pet food, and Wired magazine cast the era as the dawn of a limitless digital frontier. But behind the exuberance lay something more durable than speculative startups—the laying of a physical skeleton for the digital age. While Wall Street inflated valuations, crews from Qwest Communications threaded hair-thin strands of glass into conduits along the Burlington Northern Santa Fe railway outside Denver, joining similar projects that stitched hundreds of thousands of miles of fiber across the continent. Each strand was capable of carrying millions of simultaneous conversations, embedding the future of connectivity into the industrial corridors of an older America.

When most people recall the dot-com bubble, they think of speculative websites, IPO frenzies, and start-ups with no profits but sky-high valuations. Less remembered is that it was also built into the ground, a frenzied effort to rewire the nation’s infrastructure with glass. At the time, Wired captured the paradox of this overbuilding with a sharp observation:

The silicon economy obeys the law that supply creates demand. Too bad it’s not true for fiber. It’s hard to remember, as the tech world drowns in a sea of glass—the megamiles of excess fiber-optic cable dragging down the telecom sector—that glut was supposed to be a good thing. ‘Supply creates demand’ was the rallying cry of the semiconductor economy: Carpet the world with cheap technology, and clever hands will put it to work in a thousand ways never before imagined, giving rise to new markets and new demand. Want to carry your entire music collection in your pocket? Sure! Now how about all the music ever recorded? You bet. Moore’s law boiled down to one word: more. The more you have, the more you use. While traditional economics are driven by scarcity, the world created by the microchip is one of abundance.

That tension between speculative optimism and infrastructural overcapacity defined the bubble’s legacy. And it lingers today, as some analysts warn that the current AI data center boom may echo the 1990s “fiber glut”—with massive overbuilding, uncertain demand, and bubble risks despite the prevailing faith in long-term growth.

Deregulating the Backbone

The “internet economy” was, in practice, a massive archaeological project—turning the buried infrastructure of industrial capitalism into the nervous system of digital capitalism. The bubble had a geography, mapped onto rights-of-way assembled generations before anyone had dreamed of the internet.

The real wave of excavation began with the Telecommunications Act of 1996, legislation that promised to dismantle entrenched monopolies in local and long-distance telephone service while opening the way for commercialization of NSFNET and the wider Internet backbone. Backed by a Republican congressional majority and a corporate sector eager for new markets, the Act drew ideological support from the mid-1990s rhetoric of the “knowledge age.” A 1994 manifesto titled Cyberspace and the American Dream: A Magna Carta for the Knowledge Age captured the mood, declaring: “The central event of the 20th century is the overthrow of matter. … The powers of mind are everywhere ascendant over the brute force of things.”¹

Two years later, on the very day the Act passed, John Perry Barlow issued his now-famous Declaration of the Independence of Cyberspace at the World Economic Forum in Davos.² Barlow’s declaration echoed the frontier language of the moment, envisioning cyberspace as a realm beyond government regulation and the legal frameworks “based on matter,” a space that claimed to transcend the material world altogether.

Beneath the rhetoric, industry insiders were more cleareyed on the “matter.” An article in Network World explained the reality of the situation at the time, “[u]ltimately, the Internet will boil down to a few big providers with bilateral agreements on how to interoperate, service and support one another’s networks.”³ Yet, for many others immersed in the zeitgeist of cyberspace, it seemed that anyone with enough capital could become a carrier.

Wall Street responded to the Act by financing dozens of new telecommunications companies, each racing to string fiber across the landscape before their rivals could claim the choicest routes. The logic seemed unassailable. If every website, email, and eventually video would rely on moving bits of light through glass, then whoever controlled the most glass would win. Equipment vendors like Lucent and Nortel sweetened the stakes with generous vendor financing—essentially lending money to carriers so they could buy more gear. Bond markets and equity investors poured in billions, convinced that internet traffic was set to explode exponentially.

What emerged was a peculiar marriage of old and new America. Qwest, which had grown out of a railroad construction company, raced to string fiber along rail corridors stretching from Denver to Seattle. Williams Communications, born from a natural gas pipeline firm, launched WilTel to thread cables through pipeline rights-of-way across the South and Midwest. Even the Southern Pacific Railroad’s telecommunications unit evolved into Sprint (Southern Pacific Communications Company), transforming steel rails into information superhighways.

Already in 1995, Bay Area Rapid Transit struck an agreement with MFS WorldCom to install fiber through BART's tunnels beneath San Francisco Bay—valued around $40 million. The negotiations, conducted in fluorescent-lit conference rooms overlooking the Bay, would have seemed surreal to the nineteenth-century engineers who first imagined moving people through underwater tubes. Today, BART still licenses strands and touts connections to 40+ data centers, demonstrating how an “information superhighway” trope hardened into a durable revenue line—and a regional connectivity scaffold.

Overlapping Fiber-optic Cable Network and Railway Network in 2025

The Mathematics of Delusion

This continental rewiring was also justified by another powerful myth—that internet traffic was doubling every 90 days. The claim spread through analyst reports, earnings calls, and investor presentations like a particularly virulent meme. If true, it meant that demand was growing exponentially, far outpacing any conceivable supply, and that every new trench of fiber would soon pay for itself many times over.

But the mathematics were fiction. Network researchers like Andrew Odlyzko (at AT&T), looking at actual traffic data, found that U.S. backbone traffic was doubling roughly once a year—rapid growth, certainly, but nowhere near the purported 90-day cycle. Meanwhile, advances in fiber technology were making each strand exponentially more powerful. Dense wavelength-division multiplexing allowed dozens of signals to travel simultaneously down the same line at different wavelengths of light, like multiple conversations happening in different colors.

While demand doubled annually, supply expanded tenfold or more. Carriers buried the discrepancy under layers of creative accounting that would have impressed medieval alchemists. They sold “indefeasible rights of use”—essentially decades-long leases on fiber capacity—and booked the entire value immediately as revenue. They engaged in elaborate “capacity swaps,” trading bandwidth with competitors and treating each exchange as a sale, manufacturing revenue from thin air.

Companies like Global Crossing built themselves into financial giants through such sleight of hand, reporting explosive growth quarters while laying ever more fiber through an already saturated landscape. The executives knew they were building far ahead of demand, but the stock market rewarded growth above all else, and growth meant more fiber, more routes, more light racing through more glass.

The Unraveling

When the illusion finally collapsed, thousands of miles of fiber remained “dark,” unused. Global Crossing declared bankruptcy in 2002, leaving behind $12.4 billion in debt and thousands of miles of dark fiber. WorldCom followed in what became the largest accounting fraud in U.S. history, its executives having inflated revenues through the same capacity-swap schemes that had fooled Wall Street for years. By 2004, wholesale long-haul prices had fallen roughly 55% year-over-year in the U.S., with analysts estimating only about one-tenth of installed fiber was actually “lit.”

The human cost was immediate. Hundreds of thousands of jobs vanished, pension funds evaporated, and entire communities built around telecom hubs found themselves stranded. But the fiber itself remained buried, waiting in the dark like some technological archaeological deposit. What emerged from the wreckage was not emptiness but a new geography of connectivity. The crash didn’t erase the networks; it reorganized them, concentrating power in unexpected places while creating opportunities elsewhere.

Legal battles also reshaped the map. After the bubble burst, landowners filed class-action suits against carriers who had strung fiber along railroad rights-of-way, claiming the installations exceeded the scope of century-old easements. Courtrooms from Montana to Georgia heard testimony about the original intent of nineteenth-century railroad grants, with lawyers arguing over whether "telegraph and telephone" rights covered fiber-optic cables. By the mid-2000s, companies like Sprint, Qwest, and Level 3 were paying out hundreds of millions in settlements, but the legal victories effectively legitimized thousands of miles of routes, cementing the corridors of national internet backbones.

Meanwhile, a few urban buildings became irreplaceable nerve centers. One Wilshire Boulevard in Los Angeles, once a generic downtown office tower, transformed into the primary connection point for trans-Pacific internet traffic. Walking through its floors today feels like entering the circulatory system of the global internet—endless racks of equipment, cables snaking between floors, the constant hum of cooling systems maintaining the temperature for machines that never sleep. In 2013, the building sold for $437 million, its value derived not from square footage but from its position at the center of digital trade routes.

Similar transformations occurred across the continent. New York's 60 Hudson Street, a 1930s Art Deco monument to the telegraph age, reinvented itself as a global internet hub. Google acquired 111 Eighth Avenue—once a freight terminal for the city's garment district—for $1.8 billion in 2010, converting it into a fortress of servers and switches.

From Glut to Cloud

It’s tempting to moralize the bubble as pure waste. Yet the spatial outcome complicates that view. A chunk of today’s digital economy sits on infrastructure financed by yesterday’s exuberance, regulated by yesterday’s scandals, and optimized by today’s platforms. The bubble “mispriced risk,” but it also front-loaded an intercity network whose replacement cost would be politically and fiscally daunting now. In practical terms, the crash was a transfer mechanism—from speculative carriers to durable holders (carriers, clouds, and public networks) that could squeeze more utility from the same glass.

The fiber glut also created unexpected opportunities for public institutions. Universities and state networks, suddenly able to acquire dark fiber at fire-sale prices, built research backbones that would have been unimaginable during the bubble years. Ohio's OARnet purchased fiber in 2001 and 2002, lighting a statewide network by 2004 that connected rural hospitals to urban medical centers and linked small colleges to supercomputing resources. Internet2's FiberCo signed deals with Level 3 in 2003 for thousands of route-miles of dark fiber, enabling a new generation of research collaborations spanning continents.

Perhaps the most profound consequence was how cheap long-haul capacity enabled an entirely new model of computing. With bandwidth costs in freefall, technology companies could locate massive data centers wherever land and power were cheapest, as long as they could connect to the fiber backbone.

Google's decision to build one of its first hyperscale data centers in The Dalles, Oregon exemplified this new geography. The town of 15,000, situated along the Columbia River, combined abundant hydroelectric power with access to multiple fiber routes threading through the Columbia River Gorge. What had once been aluminum smelting territory—industries that relocated to chase cheap electricity—became the foundation of cloud computing. See my article on this here:

Northern Virginia's “Data Center Alley” grew even more dramatically, transforming former tobacco fields in Ashburn and Sterling into the world’s largest concentration of server farms. The region’s advantage wasn't natural resources but historical accident with some of the earliest internet exchange points and a dense concentration of long-haul cables converging on the nation’s capital.

As predicted, the survivors of the telecom crash had consolidated into a few large networks by the 2010s, eliminating redundant routes and rationalizing the backbone. Meanwhile, hyperscale companies like Google, Microsoft, and Facebook began building their own long-haul and subsea cables, taking advantage of surplus capacity while bypassing traditional carriers entirely. The immense wealth and power of these companies built on the backbone of a prior industrial era.

The Maps that Infrastructure Bubbles Make

The dot-com bubble is often remembered as a cautionary tale about irrational exuberance and failed startups, but its most enduring legacy lies buried under railroad beds and highway shoulders across the continent. Three patterns continue to shape digital geography and the new infrastructural order. First, the corridors of industrial capitalism became the arteries of digital capitalism. Railroad rights-of-way and pipeline easements offered the fastest paths to lay fiber, and even after bankruptcies and lawsuits, those routes remain the backbone of continental networks. Second, a handful of urban buildings became irreplaceable switching points where networks connect and traffic flows. The crash eliminated many carriers but reinforced the critical importance of neutral interconnection facilities. Third, the dramatic oversupply created lasting opportunities for public institutions and eventually hyperscale companies to acquire infrastructure at prices that would never again be available.

The bubble didn't just misprice internet stocks—it restructured territory itself. Capital evaporated, but glass remained in the ground. That glass, threaded through the landscape during a few years of technological optimism and financial excess, continues to channel the uneven and unequal flows of our digital age. As the AI boom accelerates, we should look not only at valuations but at the infrastructures being laid down—data centers, energy systems, water pipelines—because these will sketch the next territorial outlines long after the hype has passed. Bubbles burst, but the infrastructures they leave behind endure, shaping power across and over many landscapes. What sort of map will be left behind in the next few years?

References

1. This manifesto was authored by prominent technologists and neoliberals: Esther Dyson, former Wall Street analyst and member of Clinton's National Information Infrastructure Advisory Council; George Gilder, supply-side economics advocate and author of Wealth and Poverty; George Keyworth, Reagan’s Science Adviser; and Alvin Toffler, futurist and author of The Third Wave, which promoted the post-industrial society thesis. See Dyson, Esther, George Gilder, George Keyworth, and Alvin Toffler. 1994. “Cyberspace and the American Dream: A Magna Carta for the Knowledge Age.” Future Insight. http://www.pff.org/issues-pubs/futureinsights/fi1.2magnacarta.html.

2. Barlow, John Perry. 1996. “A Declaration of the Independence of Cyberspace.” https://www.eff.org/cyberspace-independence.

3. Cooney, Michael, Adam Gaffin, and Ellen Messmer. 1995. “Internet Surge Strains Already Shaky Infrastructure: Who Will Manage the ’Net’s Commercialization?” Network World 12 (14): 1; 67.

For more reading on these events and their enduring legacy, I recommend:

• McChesney, Robert W. 1999. Rich Media, Poor Democracy: Communication Politics in Dubious Times. Baltimore, MD: University of Illinois Press.

• Aufderheide, Patricia. 1999. Communications Policy and the Public Interest: The Telecommunications Act of 1996. Guilford Press.

• Kesan, J., and Rajiv C. Shah. 2001. “Fool Us Once Shame on You - Fool Us Twice Shame on Us: What We Can Learn from the Privatizations of the Internet Backbone Network and the Domain Name System.” Cyberspace Law eJournal, February.

• Turner, Fred. 2010. From Counterculture to Cyberculture: Stewart Brand, the Whole Earth Network, and the Rise of Digital Utopianism. University of Chicago Press.

• Schiller, Dan. 2014. Digital Depression : Information Technology and Economic Crisis. Geopolitics of Information. Baltimore, MD: University of Illinois Press.
  
  

Silicon Valley has always run on power, but the question of who delivers it is suddenly reshaping the map. Just a few miles apart, San José and Santa Clara are charting starkly different paths. Santa Clara’s municipal utility, once the quiet backbone of the Valley’s server farms, has hit a wall. Its grid is saturated, its rates rising, its options limited. San José, meanwhile, has gone in the opposite direction—turning to PG&E, the state’s embattled investor-owned utility, and striking a landmark deal to guarantee gigawatts of new capacity for AI facilities.

These diverging strategies tell us something larger about the politics of infrastructure today. What looks like local planning disputes is in fact the embedding of cities and utilities—sometimes reluctantly—into the industrial strategy behind the AI boom. Where resilience once meant decentralization, it now means securing gigawatt guarantees from investor-owned utilities. This new infrastructural order fuses public authority and private growth imperatives in the name of competitiveness. But the same old “growth machine” dynamics are also at play.¹

In 2019, Mayor Sam Liccardo floated the idea of creating a municipal utility after PG&E’s wildfire-prevention shutoffs left 60,000 residents without power, arguing the city needed its own lines, microgrids, and storage to ensure reliability. At the time, PG&E was bankrupt, widely blamed for wildfires, and politically toxic. Six years later, the city has not only abandoned that push for independence but entered a landmark agreement with PG&E to guarantee power delivery for a wave of new data centers. San José now markets itself as the “West Coast’s premier data center hub.” The reversal reflects less a change in sentiment than a change in scale. The rise of energy-intensive AI infrastructure has reordered local priorities, turning utilities once seen as liabilities into indispensable partners in urban development.

The political economy of Silicon Valley’s electricity regime explains this pivot. For decades, Santa Clara—San Jose’s neighbor—dominated data center development. Its municipal utility, Silicon Valley Power (SVP), offered some of the lowest industrial rates in California, undercutting PG&E by as much as 40 percent. Coupled with a dense fiber network, the pricing advantage drew more than 55 data centers into a seven-square-mile cluster north of El Camino Real. Today, those facilities consume 60 percent of the city’s power. The fiscal returns are substantial—roughly $41 million annually in tax and utility payments—but SVP has reached its limits. By late 2024, it announced it could no longer accommodate new data centers until transmission upgrades finish in 2029. At precisely this moment, San Jose entered with PG&E agreement, promising “speed and certainty” backed by 2 gigawatts of new transmission capacity. The juxtaposition highlights two competing utility models. One is Santa Clara’s city-owned grid constrained by scale. The second is PG&E’s investor-owned network, which profits from expansion and has every incentive to chase large new loads.

The new projects illustrate how infrastructure and land are being reorganized around data centers. In April 2025, San Jose approved Microsoft’s 99MW campus in Alviso, requiring removal of nearly 80 trees, demolition of an existing water district visitors’ center, construction of 224 natural gas backup generators, and the displacement of an unhoused encampment. The facility sits adjacent to the Los Esteros Energy Center, an industrial zone now effectively repurposed as a computing district. Terra Ventures, working with Arcadis, has proposed a natural-gas-powered facility nearby that integrates Bloom Energy fuel cells, absorption chillers, and even a greenhouse marketed as a community amenity—on land once pitched for an entertainment complex. Meanwhile, PG&E itself, in partnership with Westbank, plans a 200MW downtown campus linked to thousands of housing units, with a district energy system that will use waste server heat to supply hot water to apartments. Each project combines private capital with public infrastructure narratives—jobs, community amenities, or clean energy branding. But their core function is simply to provide uninterrupted computing capacity for AI workloads. In April 2025, San Jose’s City Council unanimously approved the first two projects under Mayor Matt Mahan’s new Innovative Project Pathway Program that aims to attract development through tax and impact fee abatement.

The alignment of municipal governments, utilities, and technology firms is straightforward. San Jose officials estimate each data center could bring $3 to $7 million annually in property and utility tax revenue. PG&E has pledged $2.6 billion in infrastructure investment in the South Bay between 2026 and 2035, tied explicitly to data center growth. For companies like Microsoft, Meta, and Google, the ability to secure megawatt-scale commitments from PG&E determines whether their facilities can be built on time. For utilities, these projects guarantee a stable baseload customer whose growth justifies transmission upgrades. PG&E executives now present data centers as “good news” for everyone, claiming that each gigawatt of new load could reduce average household bills by one to two percent. Infrastructure expansion is sold as universally beneficial, even as costs are, in reality, unevenly distributed and profits accrue most directly to the utility and its largest industrial customers.²

Santa Clara provides a clear example of this inequality. Its municipal utility’s rate structure discounts bulk users while charging households more as consumption rises. When SVP built out new substations and lines to accommodate data center demand, residential bills rose—8 percent in early 2023, 10 percent in 2024, and then 5 percent in 2025 to cover higher construction costs and fund new infrastructure for data center expansion. Planning commissioners raised concerns about land commitments, diesel generator emissions, and water use for cooling. Yet the fiscal contribution made resistance politically difficult. By May 2025, SVP admitted it had capped out. Requests to add 500MW of new load could not be met until at least 2029. In the process, the utility demonstrated the hard limits of the municipal model when confronted with exponential AI-driven demand. More scale is needed!

The technological fixes proposed in San Jose point to the way constraints are managed within the system rather than avoided. Microsoft’s Alviso project will be supported by an Enchanted Rock microgrid fueled with natural gas sourced from food waste. PG&E and Westbank’s downtown development promotes district heating as an ecological benefit of concentrated computing. Terra Ventures markets direct-current fuel cell integration as an efficiency gain. These measures do not alter the fundamental trajectory—massive expansion of electricity and water consumption—but provide the language by which projects are justified in regulatory, political, and financial terms. They are “bridge solutions” that maintain momentum while longer-term investments, such as new transmission lines, are built.

The implications extend beyond Silicon Valley. Across California, PG&E has reported 26 new data center applications requiring 3.5GW of baseload power—equivalent to three nuclear reactors. To stabilize supply, the state has extended the Diablo Canyon nuclear plant and delayed gas plant retirements. In Imperial County, near the Salton Sea, developers propose 500MW data centers drawing on geothermal and solar resources, framed as catalysts for “Lithium Valley.” Yet these facilities still draw heavily on contested water supplies and face local opposition over health risks and ecological strain. What happens in Santa Clara and San Jose is not contained within the Bay Area—it is the front line of a regional and nationwide restructuring of grids, land, and finance around AI.

Map of Santa Clara and San José Digital and Electric Infrastructure

What distinguishes the current AI bubble is its integration of public utilities into the speculative dynamics of digital capitalism—which I covered in a previous post. In past booms, from the railroads of the 19th century to the dot-com surge of the 1990s, overbuilding was fueled by speculative finance, with the public left to absorb the fallout after collapse. Like today’s cycle, both of those earlier crises were rooted here in Silicon Valley’s backyard. In today’s AI expansion, utilities are direct participants, structuring deals with municipalities and developers to guarantee delivery, investing billions in new infrastructure, and positioning themselves as partners in technological progress. For PG&E, once discredited by wildfires and bankruptcy, the AI boom provides not only new revenue but renewed legitimacy as the indispensable enabler of Silicon Valley’s next phase.

San Jose’s evolution from PG&E adversary to PG&E partner is symptomatic of this new infrastructural order. Where once resilience alluded to decentralization and municipal control, it now means aligning with the investor-owned utility to secure megawatt guarantees for private data centers. Santa Clara’s exhaustion of capacity and San Jose’s embrace of PG&E illustrate the dual pressures. Local utilities face hard ceilings, while larger utilities see “limitless” opportunity. Cities and utilities organize land, water, and energy around the demands of AI infrastructure, justified through fiscal returns and technical fixes, but ultimately driven by the imperatives of speculative growth. The result is a new kind of techno-statecraft—not just via national industrial policy—but emerging through negotiation of utilities, land-use authorities, and corporate demand. This is how the infrastructural core of AI is being built—through deals that fuse public authority to private growth imperatives.

References

1. Molotch, Harvey L. 1976. “The City as a Growth Machine.” The American Journal of Sociology 82 (2): 309–32.

2. Peskoe, Ari, and Eliza Martin. 2025. “Extracting Profits from the Public: How Utility Ratepayers Are Paying for Big Tech’s Power.” Harvard Electricity Law Initiative. https://eelp.law.harvard.edu/extracting-profits-from-the-public-how-utility-ratepayers-are-paying-for-big-techs-power/.
   

Mainstream industrial policy analysts often mistake infrastructure booms for evidence of economic renewal. In doing so, they overlook the deeper speculative dynamics that underpin these cycles of expansion. Recent coverage in the Wall Street Journal describes the AI infrastructure surge as a new “age of infrastructure,” lauding the record $102.5 billion in capital expenditures by Google, Meta, Microsoft, and Amazon as a sign of industrial strength. Going further, Axios characterizes AI-related investments as an “AI super-stimulant” propping up the U.S. economy amid broader weakness. Bank of America has projected $700 billion in AI capex through 2026, framing this as a bulwark against stagflation (although the reality may be a little different). These narratives treat investment volume as inherently productive, even as analysts are beginning to warn of overbuild risk, speculative finance, and fragile debt structures.

We’d do well to look past the glossy narratives and see AI “innovation” economy for what it really is—capitalism with faster processors.

Along with the acceleration, the underlying dynamics of fast capitalism haven’t changed all that much—speculative hype inflates asset values, insiders extract returns early, and risks are ultimately offloaded onto the public or “dumb money.” We can see that the financial press has begun to reflect this growing concern about an AI bubble:

• Fortune (July 22, 2025): “The AI Boom Is Now Bigger than the ’90s Dotcom Bubble.”

• Financial Times (July 30, 2025): “What’ll Happen If We Spend Nearly $3tn on Data Centres No One Needs?”

• Wall Street Journal (August 3, 2025): “The AI Boom’s Hidden Risk to the Economy.”

Like the railroad and dot-com booms before it, the AI bubble is not a deviation from normal economic cycles—it is a structural feature of capitalism itself. Far from being an accidental byproduct, it is deliberately constructed—extraction is repackaged as innovation, hype replaces substance, and inflated promises attract speculative capital. This process is actively supported by an industrial policy and regulatory apparatus oriented toward private gain. The bubble is not a flaw in the system—it is how the system works.

The AI sector has emerged as the primary engine of economic optimism in an otherwise fragile U.S. economy.

Driven by enormous capital expenditures in data centers, chips, and energy infrastructure, firms like Microsoft and Meta are now allocating over a third of their total revenue to AI infrastructure alone. These investments have contributed more to GDP growth over the past two quarters than all consumer spending combined.

This rapid expansion is not powered by organic revenue, but by a speculative financial engine. Venture-backed firms like CoreWeave—originally a crypto miner—have reached valuations as high as $19 billion with minimal revenue, raising $650 million in secondary sales and targeting a $1.5 billion IPO. Microsoft’s finance leases, mostly tied to infrastructure, have tripled since 2023 to $46 billion, while Meta is seeking $30 billion in private credit for its data center expansion. These are not isolated cases—they point to a broader shift in the structure of financing behind AI growth.

What we are witnessing is not a conventional investment cycle but a high-leverage system built on opaque instruments and limited oversight. Early dismissals of systemic risk of the AI bubble—such as those made in 2019, when its expansion was still thought to be largely equity-financed—have become outdated. Since the release of ChatGPT in 2022, the sector’s capital demands have surged, outpacing equity investment and pushing firms more heavily into debt markets.

Morgan Stanley projects that AI infrastructure spending could reach $2.9 trillion by 2028, leaving a $1.5 trillion financing gap. To close it, firms are turning to complex forms of credit: multi-billion-dollar loans backed by GPUs, drawing on venture debt, and leveraging structured credit based on projected revenues. The private credit market for data infrastructure—virtually nonexistent in 2018—now exceeds $50 billion and is largely funded by insurance companies, private equity firms, and traditional banks. According to the Boston Fed, U.S. banks now lend 14% of their commercial portfolios to non-bank financial institutions like private credit funds.

If AI revenue projections fail to materialize, the fallout will extend far beyond Silicon Valley. A downturn would ripple through the broader financial system, revealing just how much institutional risk has been taken on to underwrite the illusion of limitless AI-driven growth.

Historical precedents abound.

During the late 19th and early 20th centuries, the rise of the railroads gave way to a series of speculative bubbles built on lofty promises and shaky financial foundations. The Crédit Mobilier scandal of the 1870s saw Union Pacific executives siphon off $44 million in government contracts through a fake construction company, exposing the extent to which public infrastructure projects could be used for private gain. The ensuing panics of 1873 and 1893 were triggered in part by the overbuilding of railroads fueled by debt, subsidies, and inflated valuations. In the 1920s, the Van Sweringen brothers built a 28,000-mile rail empire through a pyramid of holding companies financed largely through debt and public securities, despite having little direct equity. Revenues declined with the rise of the automobile, but asset values remained artificially high until the 1929 crash revealed the empire’s hollow core—leading to widespread bankruptcies and community collapse during the Great Depression.

Just as the internet was built (literally) within the rail system’s rights-of-way, the dot-com bubble of the late 1990s followed a similar speculative path. Startups with no viable revenue models attracted massive investment on the promise of revolutionary innovation, often underwritten by investment banks eager to feed the hype. IPOs were deliberately underpriced to generate dramatic first-day surges—VA Linux soared 698% on debut—while analysts issued glowing ratings to maintain deal flow. Profitability was replaced by speculative metrics like “eyeballs,” and when lock-up periods expired, insiders dumped their shares. As in the railroad era, the crash was swift and brutal: between 2000 and 2002, the NASDAQ lost 78% of its value, erasing $5 trillion in market capitalization and leaving pension funds, retail investors, and employees to absorb the damage.

Both eras were marked by exuberant narratives of transformative infrastructure, underwritten by financial institutions that prioritized hype over fundamentals. In each case, insiders and early investors cashed out before the bust, leaving everyone else to absorb the fallout. The physical networks—rail lines and fiber optic cables—persisted, but the financial structures collapsed under their own weight.

The current AI cycle is more complex but no less extractive.

AI infrastructure spending now exceeds telecom investment during the dot‑com bubble. U.S. tech giants are on track to spend between $320 and $350 billion on AI-related infrastructure in 2025 alone. These investments are also reshaping land, energy, and water systems in major ways that tether these resources and their governance to speculative buildouts across many states. Northern Virginia, Georgia, and parts of Texas have become epicenters of AI infrastructure. State and municipal governments are offering enormous tax breaks—often exceeding $2 million per permanent job created. Permitting is being streamlined through federal executive orders and state preemption laws. Many data centers are being colocated with legacy fossil fuel infrastructure or built on repurposed industrial land, including former steel mills and nuclear power sites.

The environmental and social costs are mounting. Data centers already consume over 4% of U.S. electricity, and that share is expected to triple by 2030. Their water demands reach into the millions of gallons per day in some regions, intensifying pressure on already strained systems. These facilities are increasingly powered by coal- and gas-intensive grids. The costs—energy procurement, transmission upgrades, water delivery—are being passed on to ratepayers through long-term utility contracts. Meanwhile, grid reliability is deteriorating, and local communities face infrastructure stress without commensurate public benefits.

Discontent is growing. In education, the American Association of University Professors (AAUP) has pushed back against AI edtech mandates, citing degraded student learning conditions and loss of academic freedom. Students describe AI-augmented education as alienating and ineffective. Artists and journalists have begun to organize against what they see as the expropriation of their work by generative AI tools. Public sentiment toward AI remains mixed at best—widely viewed as disruptive, extractive, and imposed without consent.

These tensions are structural, not incidental.

The AI boom, like previous speculative cycles, is propelled by inflated promises and strategic narratives of national competitiveness—but behind the scenes, it operates as a mechanism for transferring risk from capital to the public. The architecture of this risk has long been in place, papered over by the language of innovation and progress.

Historically, each phase of speculative infrastructure development has left behind overbuilt systems, concentrated wealth, and long-term costs borne by society at large. The railroads enabled telegraph and telecom networks from which the dot-com bubble laid the digital backbone for cloud computing. Today’s AI expansion follows the same pathways—quite literally, in the case of fiber laid along railroad rights-of-way—drawing on deregulated energy markets, subsidized land deals, and even Cold War-era industrial zones once left behind by corporate giants in search of new frontiers. It repurposes the material residue of past cycles while reproducing their core logic—private enrichment through public exposure.

Whether this current boom ends in a sharp collapse or a slow deflation remains uncertain. The more urgent question, then, is what follows the bust. Will there be public reckoning and structural reform, or a deeper consolidation of corporate power as crisis is leveraged to entrench control?

Further reading:

• Baran, Paul A., and Paul Marlor Sweezy. 1966. Monopoly Capital: An Essay on the American Economic and Social Order. NYU Press.

• Starr, Paul. 2002. “The Great Telecom Implosion.” The American Prospect, September 8, 2002. https://www.princeton.edu/~starr/articles/articles02/Starr-TelecomImplosion-9-02.htm.
  

We are in the midst of a new infrastructural order where the biggest tech firms—led by Meta, Alphabet, Microsoft, and Amazon—are spending record sums on physical assets like data centers, factories, and energy, with their combined quarterly capital expenditures recently surpassing $100 billion. This wave of investment parallels the era of the industrial titans, as today's giants deepen their competitive moats through vertical integration and monopolization of resources, talent, and strategic sites—including former steel mills repurposed for data centers.

This corporate buildout is not occurring in a vacuum—it is being actively enabled and accelerated by state policy. On Wednesday, July 23, 2025, the Trump administration released its AI Action Plan and companion website, ai.gov. Billed as a national “renaissance,” the plan casts artificial intelligence as the decisive frontier in a new era of strategic competition—where technological supremacy is not only a pathway to prosperity, but a condition for security, sovereignty, and survival. Its messaging is unequivocal:

“Win the AI race, and America secures its future; lose, and it risks irrelevance.”

This framing echoes across both official policy and corporate rhetoric. In a Substack post on launch day, OpenAI described its 4.5 GW Stargate expansion as a “democratic” countermodel to China’s “state-led authoritarian” approach, warning that “the window to shape this future is closing fast.” But behind the language of urgency and empowerment lies a sweeping consolidation of state authority to accelerate infrastructure, deregulate oversight, and channel public resources toward fossil-powered compute buildouts—slyly recasting environmental law, land use policy, and democratic review as impediments to national competitiveness.

The plan codifies a direction already set in motion. Since early 2025, the administration has issued a flurry of executive orders dismantling Biden-era rules on climate, equity, and AI governance while invoking a national energy emergency to fast-track fossil and nuclear development. What began as discrete policy shifts has now been woven into a unified strategy: compress permitting timelines, roll back oversight, mobilize federal procurement, and open public lands to hyperscale infrastructure. In effect, the state is retooled to serve as a growth engine for private cloud and compute capital.

Major technology firms—Alphabet, Meta, Microsoft, Nvidia—have lobbied aggressively for these outcomes, winning regulatory relief, procurement preferences, and public subsidies. But while industry celebrates, over a hundred labor, civil rights, and environmental groups have condemned the plan as a corporate land grab disguised as national renewal. Even centrist policy experts caution that the rollback of antitrust enforcement, defunding of research, and restrictions on immigration threaten long-term innovation in favor of short-term consolidation.

At its core, the AI Action Plan is not about ethics or algorithms—it is about infrastructure. The governing logic is clear: accelerate the deployment of fixed capital—data centers, semiconductor fabs, energy systems, and transmission corridors—and clear the regulatory, legal, and territorial pathways to make it possible. Public governance is subordinated to private investment imperatives. The apparatus of the state, once tasked with balancing public interests, is now marshaled to eliminate “friction” for capital.

This analysis builds on Part I of AI Industrial Policy Under Trump 2.0, which outlined the administration’s early moves to redefine AI through a national security lens. Those tactical shifts are now formalized into a systemic program of infrastructural acceleration and techno-industrial enclosure.

Share

Barriers—Clearing the Way for AI Industrialization

The Trump administration’s AI strategy is not a conventional policy roadmap; it is a systematic dismantling of regulatory guardrails in the service of digital capitalism. From its earliest days, a cascade of executive orders reoriented the federal bureaucracy around one goal: to eliminate legal and institutional constraints on AI-driven infrastructure expansion. A January 20 order repealed Biden-era mandates on climate, equity, and algorithmic safety. The same day, a national energy emergency was declared, framing energy-intensive AI development as a matter of security. A subsequent January 23 order instructed agencies to eliminate “barriers” to innovation and prioritize competitiveness and national security over accountability. Then, on January 31, a sweeping deregulatory order required agencies to repeal ten existing rules for every new one and maintain negative net regulatory cost. These moves, overseen by an empowered Office of Management and Budget (OMB), recast regulatory authority as an engine for capital formation rather than public protection.

The strategy centers on energy, rebranding fossil fuels and nuclear power as essential to AI development. An April 8 executive order revives coal under a national security rationale, dismantling environmental protections, expanding federal leasing, and linking coal-fired generation to data center growth. Three subsequent nuclear orders advance this agenda by reinvigorating the industrial base, deploying advanced reactors, and restructuring the Nuclear Regulatory Commission. Together, they accelerate licensing, expand uranium enrichment and fuel recycling, and authorize Defense Production Act powers to guarantee offtake agreements and financing. Deregulation is paired with subsidies, loan guarantees, and export promotion, embedding high-cost, high-risk fossil and nuclear infrastructure into the federally supported foundation of AI-driven industrial growth.

Procurement reforms further entrench this industrial realignment. The April 15 order, “Restoring Common Sense to Federal Procurement,” strips discretionary compliance provisions from the Federal Acquisition Regulation, enabling agencies to bypass environmental, labor, and equity requirements. OMB memoranda M-25-21 and M-25-22 empower Chief AI Officers to prioritize domestic suppliers and accelerate delivery timelines. Procurement becomes a tool for accelerating infrastructure deployment, embedding private platforms in core state functions while shifting operational and financial risk onto public agencies. “Innovation” is the alibi for eroding oversight.

The plan also reaches into public education. An April 23 order establishes a national K–12 AI curriculum backed by over sixty corporations, including Amazon, Google, and Microsoft. These firms serve not just as advisors, but as platform providers—shaping tools, content, and pedagogy. Marketed as competitiveness policy, the initiative effectively transforms public education into a workforce pipeline, aligning civic institutions with private sector needs from an early age.

Legislatively, the strategy culminates in HR 1—the so-called “One Big Beautiful Bill”—signed into law on July 4. It codifies AI as a national security imperative, directing billions in subsidies to cloud infrastructure, defense AI applications, and microelectronics research. Military contractors are guaranteed demand pipelines, while the Department of Energy is tasked with building an “American Science Cloud” to centralize federal data and computing power. Under the idea of “technological sovereignty,” public resources are redeployed as profit engines for private firms—absent meaningful democratic or environmental oversight.

These executive and legislative maneuvers converge in the AI Action Plan’s generic-sounding first pillar—“Accelerate AI Innovation.” Yet, federal funding is weaponized to discipline state and local policy. Jurisdictions imposing “burdensome AI regulations” risk losing access to grants and contracting opportunities. Algorithmic accountability, once a governance priority, is reframed as an obstacle to market dynamism. Behind the rhetoric of “consumer choice” lies a deeper restructuring of regulatory federalism—subordinating subnational autonomy to the imperatives of infrastructural acceleration.

Bypasses—The Energy–AI Nexus and Grid Politics

At the heart of the AI Action Plan lies a central assumption: no AI future without abundant, uninterrupted power. This is a central underlying feature of Pillar II, “Build American AI Infrastructure.” Energy, once framed as a public utility or climate concern, is recast as a strategic weapon in a geopolitical arms race. The plan warns that U.S. capacity has stagnated since the 1970s and must now be radically expanded—not for decarbonization, but to serve compute-intensive growth—echoing the drive to restrict renewables, retrench fossil fuels, and expand nuclear energy.

Projected demand figures are staggering. Data centers already consume about 4% of U.S. electricity, and that share could triple by 2030. Deloitte projects load could increase thirtyfold, from roughly 4 GW today to 123 GW by 2035. Yet instead of pursuing equitable or climate-conscious grid modernization, the Action Plan entrenches reliance on fossil fuels and nuclear power—reversing coal retirements, subsidizing gas-fired plants, and accelerating reactor approvals. Transmission and storage investments are nominally included, but the underlying financial structure favors capital-intensive, centralized infrastructure, locking in the dominance of investor-owned utilities.

This is not technical modernization—it is strategic entrenchment. Utilities like NextEra, Duke, and Dominion, guaranteed profits through regulated rates of return, stand to extract significant rents from speculative overbuilds marketed as “AI readiness.” Ratepayers, however, are offered no protections against the costs. Meanwhile, tech giants increasingly pursue co-location deals, securing direct energy supply from adjacent generation sites. Though the Federal Energy Regulatory Commission (FERC) blocked Amazon’s 2024 attempt to link a data center to Talen Energy’s Susquehanna nuclear plant, the Action Plan now aims to normalize such arrangements—particularly under emergency provisions designed to bypass procedural safeguards.

FERC itself is undergoing transformation. With Project 2025 advisor David LeCerte appointed to a leadership role, the Commission has aligned more closely with administration goals. In February, FERC launched a “show-cause” proceeding against PJM Interconnection—the country’s largest grid operator—pressuring it to revise cost-sharing rules and accelerate siting for AI-linked infrastructure. These actions build on earlier Orders 1920 and 2023, which require long-term transmission planning and faster, centralized approval processes. FERC is no longer just regulating the grid; it is actively reshaping it to serve digital infrastructure development.

The deregulatory push extends beyond electricity. The administration proposes eliminating rehearing-period protections for natural gas pipelines, allowing construction to begin immediately after initial approval. Justified under the national energy emergency, this rollback exemplifies a broader strategy: fast-track fossil infrastructure, displace environmental and property rights protections, and create a streamlined siting regime for compute-linked development.

In this vision, the grid is not a public good but an industrial platform. “Modernization” is stripped of its ecological content and redefined as utility-led overbuild. Environmental laws are reframed as bottlenecks; permitting becomes an accelerant for territorial expansion. What emerges is a political economy in which the foundational infrastructure of AI is not only energy-intensive, but fossil-fueled, investor-driven, and federally guaranteed—a structure designed to concentrate power while externalizing social and environmental risk.

Buildouts—Land, Permitting, and the Speculative Frontier

The other dimension of the Action Plan’s Pillar II focuses on the transformation of land use and environmental permitting into strategic instruments of industrial acceleration, dismantling long-standing legal safeguards to clear space—literally—for hyperscale expansion. A new executive order expands categorical exclusions under the National Environmental Policy Act (NEPA), streamlines Clean Water and Clean Air Act reviews, and applies the FAST-41 regime to compress permitting timelines from years to months. Blanket approvals are coupled with subsidies, tax credits, and guaranteed offtake agreements to insulate projects from litigation and de-risk investment. Environmental lawyers warn that these blanket exemptions risk ignoring site-specific harms, particularly to wetlands, waterways, and endangered species.

Nowhere is this more visible than on federal land, where the state plays both landlord and booster. Agencies are directed to identify “priority sites” for AI development, targeting deserts in the Southwest, deindustrialized regions in the Midwest, and contaminated zones like Brownfields and Superfund sites. Sixteen federally owned properties—many legacies of the Cold War nuclear and defense complex—are being rebranded as “AI-ready” zones. These include vast sites like the Idaho National Laboratory (890 sq mi), Savannah River Site (310 sq mi), and former uranium enrichment facilities in Paducah and Portsmouth. Urban-adjacent labs like Argonne (Chicago) and Brookhaven (Long Island) are favored for their access to labor markets, cooling water, and network infrastructure.

This territorial pivot is not accidental—it has been incubated by national security think tanks and AI-focused advocacy groups. The Special Competitive Studies Project and the Institute for Progress, among others, have pushed proposals for “Special Compute Zones” with relaxed regulation, direct subsidies, and streamlined permitting. These ideas now appear almost verbatim in HR 1, which authorizes land sales, grants infrastructure siting rights, and expands eminent domain authority to support AI-related development.

On the ground, this strategy is already in motion. Stranded coal plants are reimagined as energy anchors for data centers; old transmission corridors are retooled to serve compute campuses; remediated industrial sites are rebranded as innovation hubs. What emerges is not just a buildout—it is a spatial reordering of the country, where techno-industrial expansion exploits the fiscal and physical residue of past state-building projects.

While cloaked in the language of modernization and competitiveness, these developments enact a deeper logic of enclosure. Public lands, public subsidies, and public infrastructure are increasingly reserved for private cloud platforms and energy monopolies. Environmental protections are sidelined, local governance is preempted, and the risks—pollution, resource strain, community displacement—are externalized onto already-vulnerable populations. The frontier, once a metaphor, becomes a regulatory and territorial condition for AI-led growth.

Geopolitics — Compute Power as Strategic Doctrine

The AI Action Plan is not only domestic industrial policy—it is a geopolitical blueprint for organizing global power through infrastructure. At its center is a civilizational narrative. AI supremacy is framed as a zero-sum contest with China, where victory ensures U.S. prosperity and security, and loss signals decline. This framing extends the frontier logic of deregulation and enclosure outward, justifying the transformation of digital infrastructure into the hard assets of state power.

To operationalize this vision, another executive order creates the American AI Exports Program, which bundles hardware, software, models, and cloud infrastructure into a subsidized national offering for foreign markets. Backed by federal financing, this program reorients diplomacy and trade policy toward platform expansion, channeling public funds into the global ambitions of Amazon, Microsoft, Google, and Nvidia. Here, export strategy becomes market capture; diplomacy becomes corporate growth.

Domestically, the adversarial framing justifies sweeping institutional exceptions. What was once environmental governance or public interest planning is reclassified as a national defense concern. Under the Plan, agencies may override local review processes, invoke emergency powers, and deploy Defense Production Act authority to expedite infrastructure deemed critical to AI capacity. These moves not only concentrate state power but also delegitimize dissent: labor, environmental, or community resistance can now be cast as threats to national survival.

Meanwhile, the CHIPS Act’s “guardrails” are sharpened. Export controls on advanced semiconductors are tightened, and a new Technology Diplomacy Strategic Plan is launched to draw allies into alignment with U.S. supply chain policy. The geography of geopolitical influence shifts accordingly: fabs, server farms, and transmission corridors become the new bases of power projection. Compute capacity—once an economic asset—is reframed as a defense asset to be subsidized, secured, and geopolitically deployed.

In this landscape, corporate actors are elevated to national champions. Their infrastructure buildouts are presented not as private ventures but as patriotic imperatives. This fusion of state and cloud capital erodes the boundary between public and private, diplomacy and commerce, defense and development. What emerges is a techno-industrial bloc with state power reconfigured to defend and extend capital-intensive digital infrastructure.

The implications are profound. Internationally, this strategy risks intensifying technological militarization, triggering retaliation, and fueling a global infrastructure arms race. Domestically, it enshrines a political order in which militarized growth, territorial enclosure, and corporate consolidation are pursued under the banner of innovation. As the Council on Foreign Relations has warned, this approach may deliver short-term dominance but at the cost of long-term instability—environmental, diplomatic, and democratic.

At its core, what’s proposed is a geopolitical economy built on extraction—of resources, of territory, and of institutional legitimacy. Laws are rewritten, permitting weaponized, and risks offloaded onto communities and ecosystems. The spoils accrue to hyperscalers and utilities; the costs are socialized. The AI Action Plan thus marks not just a shift in industrial strategy, but the consolidation of a state-backed regime of techno-capital accumulation—premised on urgency, structured through exception, and sustained by the erosion of democratic constraint.

Conclusion—A New Infrastructural Order

Microsoft’s $4 trillion valuation—reached after a 4.5% stock jump—marks a historic surge driven largely by AI infrastructure demand, particularly through its Azure cloud services and deep ties to OpenAI. This rise parallels Nvidia’s similar valuation milestone and reflects a broader capital expenditure boom in AI, which, according to analysts, contributed more to recent U.S. economic growth than all consumer spending. Yet underlying this investor euphoria are growing concerns as much of the revenue is unprofitable or opaque, AI tools face public skepticism, and critics warn the boom increasingly resembles a speculative bubble rather than sustainable technological transformation.

On the one hand, Trump 2.0’s AI Action Plan is a political project that consolidates executive power, corporate interest, and territorial governance into a new infrastructure regime. By weaponizing permitting, subsidizing fossil-heavy energy systems, and overriding local authority, it remakes the state as an instrument of computational expansion. What was once environmental review, democratic planning, or educational policy is now repurposed to secure market share and geopolitical standing.

On the other hand, this is not a story of industrial revival. It is a highly speculative reordering of land, law, and legitimacy, driven by a fusion of state and cloud capital. The costs—environmental degradation, community displacement, regulatory erosion—are diffused across space and time. The rewards—access to compute, energy monopolies, infrastructure rents—are concentrated in the hands of a few. Yet, if the boom turns bust, expect the bulk of consequences to be “democratically” distributed.

Stay tuned for Part III, which examines how state-level legislation and preemption further entrench this regime against resistance—and Part IV, which explores the geopolitics of computing power between the U.S. and China.

Also, check out the People’s AI Action Plan—a broad coalition-backed initiative calling for democratic oversight and public-interest safeguards in AI development, aiming to counter corporate-driven agendas and ensure technology serves workers, communities, and the environment.

Executive Orders Referenced:

• “Accelerating Federal Permitting of Data Center Infrastructure.” The White House. July 23, 2025. https://www.whitehouse.gov/presidential-actions/2025/07/accelerating-federal-permitting-of-data-center-infrastructure/.

• “Initial Rescissions Of Harmful Executive Orders And Actions.” The White House. January 20, 2025. https://www.whitehouse.gov/presidential-actions/2025/01/initial-rescissions-of-harmful-executive-orders-and-actions/.

• “Declaring a National Energy Emergency.” The White House. January 21, 2025. https://www.whitehouse.gov/presidential-actions/2025/01/declaring-a-national-energy-emergency/.

• “Removing Barriers to American Leadership in Artificial Intelligence.” The White House. January 23, 2025. https://www.whitehouse.gov/presidential-actions/2025/01/removing-barriers-to-american-leadership-in-artificial-intelligence/.

• “Unleashing Prosperity Through Deregulation.” The White House. January 31, 2025. https://www.whitehouse.gov/presidential-actions/2025/01/unleashing-prosperity-through-deregulation/.

• “Restoring Common Sense to Federal Procurement.” The White House. April 15, 2025. https://www.whitehouse.gov/presidential-actions/2025/04/restoring-common-sense-to-federal-procurement/.

• “Advancing Artificial Intelligence Education for American Youth.” The White House. April 23, 2025. https://www.whitehouse.gov/presidential-actions/2025/04/advancing-artificial-intelligence-education-for-american-youth/.

• “Accelerating Federal Permitting of Data Center Infrastructure.” The White House. July 23, 2025. https://www.whitehouse.gov/presidential-actions/2025/07/accelerating-federal-permitting-of-data-center-infrastructure/.

• “Promoting The Export of the American AI Technology Stack.” The White House. July 23, 2025. https://www.whitehouse.gov/presidential-actions/2025/07/promoting-the-export-of-the-american-ai-technology-stack/.

Today, over 90% of the world’s most advanced semiconductors are manufactured in Taiwan, with Taiwan Semiconductor Manufacturing Company (TSMC) alone responsible for nearly two-thirds of global foundry revenue. *It’s also important to note that nearly all advanced AI chips globally are currently produced by TSMC fabs in Taiwan. These chips power not just consumer electronics but also underpin critical infrastructures—from AI and cloud computing to weapons systems and military logistics. The sheer concentration of this production capacity—both geographically and corporately—has triggered alarm among U.S. policymakers, who increasingly cast it as a national security threat and a supply chain “vulnerability,” especially in light of escalating tensions between the U.S. and China and across the Taiwan Strait.

But this dependence is not simply the byproduct of globalization or foreign competition. China didn’t take manufacturing away from the U.S. It is the result of a deliberate, decades-long trajectory of offshoring and disinvestment by U.S. firms, shaped by political choices and the logic of financialized capitalism. The current “crisis” of semiconductor supply is not external to U.S. power—it is a crisis made by it.

Semiconductor dominance in the U.S. was never the outcome of market forces alone. It was built through deep state involvement: Cold War military procurement, federal R&D, and university-based innovation systems. Initiatives like DARPA’s MOSIS program and the public-private SEMATECH consortium were instrumental in constructing a domestic semiconductor ecosystem. Yet by the 1980s and 1990s, U.S. firms—having benefited from these investments—began to dismantle vertically integrated production models, outsourcing fabrication to East Asia to lower costs and satisfy shareholder demands. This shift was not inevitable; it mirrored the broader discipline imposed by financial markets, where long-term industrial capacity was sacrificed for quarterly returns.

Today, attempts to rebuild domestic semiconductor production—most visibly through the CHIPS and Science Act’s $52 billion subsidy package—unfold against a backdrop of systematic disinvestment and financialized retreat. The current push for “reshoring” may be less of a return to industrial strength and more of a state-financed response to a crisis engineered by decades of corporate offshoring and policy complicity. Nowhere is this more evident than in the U.S. dependence on Taiwan and TSMC—not a geopolitical accident, but the culmination of U.S. firms' decisions to externalize fabrication while preserving intellectual property and capital accumulation at home.

TSMC’s rise was enabled by the very firms and policies that dismantled U.S. manufacturing, and it now occupies a position of structural indispensability in the global digital economy. Far from simply a supply chain risk, Taiwan represents the infrastructural expression of a techno-industrial model that displaces the costs of production while consolidating profits through design, platforms, and control. In this article, I trace how U.S. public investment built the original semiconductor ecosystem, how private capital hollowed it out in pursuit of returns, and how today’s techno-nationalist industrial policy risks deepening monopoly and dependency—locking the future of digital infrastructure into a global architecture of asymmetry, extraction, and strategic entanglement.

The Rise and Fall of U.S. Technological Power

The rise of U.S. semiconductor dominance was inseparable from state intervention, including public investment, military coordination, and government regulatory action. Far from being a triumph of market forces or entrepreneurial genius, the industry emerged from a dense infrastructure of public support and oversight. Following World War II, government agencies, military laboratories, and university research centers such as Lincoln Laboratory and Project Whirlwind laid the technological groundwork for semiconductor innovation. The 1956 antitrust consent decree against AT&T, which forced the company to license patents and share research developed at Bell Labs—including foundational work on the transistor—further ensured that critical semiconductor technologies were disseminated beyond a single corporate monopoly. This combination of government funding, procurement, and regulatory policy constituted what Fred Block described as a “hidden developmental state,” making possible the entire industrial ecosystem later claimed as a private sector achievement.¹

Throughout the Cold War, the federal government actively nurtured the industry. DARPA’s MOSIS project in the 1980s funded a fabrication facility at the University of Southern California, allowing startups and researchers to prototype chip designs without prohibitive costs. But before China was considered a geopolitical and economic security threat, Japan was a major worry for U.S. politicians and firms. SEMATECH, launched in 1987 with Defense Department backing, coordinated research among fourteen U.S. semiconductor companies to counter Japanese competition.² Federal support helped stabilize collaboration, despite firms’ reluctance to share proprietary knowledge or assign their top talent. By 1997, SEMATECH had become financially independent, but its success was premised on years of public subsidy and state-led coordination.

These investments built an innovation ecosystem that wove together academic research, industrial application, and public funding. As sociologist Manuel Castells noted, the strength of technological clusters depended not just on markets but on the co-location of research institutions, skilled labor pipelines, and government-backed resources.³ Yet as this ecosystem matured and consumer products overtook military procurement for semiconductors and related technologies, U.S. industrial policy increasingly celebrated market primacy while downplaying the state’s foundational role. In practice, public money had laid the groundwork for private accumulation.

By the 1980s and 1990s, corporate strategies began decoupling design from manufacturing. The rise of the “fabless” model—exemplified by new firms like Qualcomm (1985) and later Nvidia (1993)—allowed firms to outsource production to foundries like TSMC while avoiding the soaring capital costs of fabrication plants. This was not just an operational shift but a financial one. Capital markets rewarded asset-light strategies, outsourcing, and short-term returns over the maintenance of domestic production. The foundry model’s expansion paralleled growing shareholder demands for cost-cutting and global supply chain rationalization. This paralleled domestic policy shifts that enabled geographic shifts in manufacturing more broadly.

In this environment, offshoring fabrication to Taiwan and South Korea became a rational strategy aligned with financialized imperatives. Between 2000 and 2020, U.S. semiconductor and microelectronics employment (NAICS 3344) shrank by more than 370,000 jobs, reflecting a systemic disinvestment from domestic manufacturing. This offshoring was not inevitable; it was facilitated by policy frameworks and corporate decisions that privileged financial engineering over industrial resilience—a policy event that I’ll cover in more depth in a separate article. In effect, the public funding that once nurtured the industry was subordinated to a political economy in which shareholder value overrode considerations of national industrial capacity. As U.S. firms offshored fabrication, they locked themselves into reliance on an abundance of external suppliers—setting the stage for today’s structural dependence and geopolitical vulnerability. *Patrick McGee’s Apple in China gives an excellent overview of the role Apple played in building Chinese supply chains.

Change in Semiconductor and Microelectronics Industry Employment (NAICS 3344) in U.S. by CBSA, 2000 to 2020

Distribution of Global Semiconductor Production Capacity in 2023

The Rise of TSMC and U.S. Tech Dependence

The outsourcing of semiconductor manufacturing from the United States to East Asia reached its apex in the rise of Taiwan Semiconductor Manufacturing Company (TSMC). Founded in 1987 as a venture initially aimed at attracting U.S. clients, TSMC pioneered the pure-play foundry model, specializing exclusively in contract manufacturing for firms that designed but did not fabricate their own chips. This model enabled companies like Qualcomm, Nvidia, and Broadcom to shed the capital-intensive burdens of chip fabrication while retaining control over design and intellectual property. This also involved considerable U.S. support as U.S. firms and government were primarily worried about Japanese competition at the time.

Over the next three decades, TSMC consolidated its position at the heart of global semiconductor production. By 2024, it controlled nearly 64% of global foundry revenue, a dominance that extended even further at advanced process nodes like 7-nanometer and 5-nanometer technologies, where TSMC accounted for nearly all available global capacity. This supremacy wasn’t only a function of scale, but a reflection of sustained investments in research, process innovation, and a reputation for neutrality in serving multiple competing customers without direct involvement in chip design—a reputation firms like Samsung and Intel have had difficulty maintaining.

For U.S. tech giants, TSMC became indispensable. The infrastructure of U.S. artificial intelligence, cloud computing, and military systems increasingly rested on chips fabricated in Taiwan. Apple depended on TSMC’s cutting-edge manufacturing for its A-series and M-series chips powering iPhones and Macs. Amazon’s AWS relied on TSMC to produce its custom server chips, including the Trainium and Graviton processors that underpin its cloud and machine learning services. Google used TSMC’s fabrication for its Tensor Processing Units, essential for neural network training and AI services. Microsoft’s Azure Maia 100 AI chip, used to train large language models, was fabricated by TSMC at 5-nanometer nodes. Nvidia’s AI accelerators and AMD’s server processors likewise depended on TSMC’s advanced technologies. Even direct competitors like Intel turned to TSMC to access leading-edge fabrication capacity unavailable domestically. Looking at the chart below, we can see just how dominant TSMC-manufactured GPUs have become in recent years.

Moore’s Law in GPU Production by Manufacturer 1995 to 2025

Rebuilding domestic fabrication is no simple task. TSMC’s dominance also reflected cumulative investments in specialized equipment, materials, technical expertise, and supplier networks developed over decades. Establishing competitive alternatives required not just capital—individual TSMC fabs cost tens of billions of dollars—but an ecosystem of knowledge and infrastructure not easily replicated. At the most advanced nodes, dual sourcing by downstream firms became impractical. Technological complexity, yield optimization (i.e. production of viable chips), and sunk costs meant firms could not easily maintain parallel suppliers. In sum, U.S. firms, and by extension, the U.S. defense and high-tech industrial base, found itself structurally dependent on TSMC—a dependence not incidental but the product of decisions that prioritized short-term returns over long-term industrial resilience.

High-Technology Industrial Policy and Hegemony

By the 2020s, the dependence of U.S. firms grew more apparent to investors, military planners, and government officials alike as investment in data centers and AI infrastructure began to grow. Heightened competition with China and Covid-related supply chain bottlenecks laid bare the price of just-in-time production and hyper-consolidated supply chains. To get a grip on advanced semiconductor production, U.S. congress passed the CHIPS and Science Act in 2022, among other legislation and executive actions by the Biden administration. The Act authorized tens of billions of dollars in federal funding to rebuild domestic semiconductor production, framed as a strategic response to economic insecurity, supply chain risks, and the growing technological rivalry with China. Yet, the CHIPS and Science Act must be understood not simply as an industrial subsidy but as part of a broader trajectory of techno-nationalist industrial policy.

This policy response did not emerge from a single problem, but a confluence of issues. During the Cold War, U.S. technological dominance in semiconductors was built through massive public investment, from Pentagon contracts to DARPA-funded research to university and corporate R&D collaborations. As Henry Kissinger noted, technological power and managerial expertise enabled the U.S. to shape international systems and development trajectories abroad. Yet by the post-Cold War period, U.S. firms had offshored fabrication to East Asia to cut costs and satisfy capital market demands, retaining intellectual property while hollowing out domestic manufacturing. Dependency on external suppliers like TSMC was the result.

China’s semiconductor ambitions, embodied in the Made in China 2025 initiative, reframed this dependency as a national security threat. Although China remained a minor producer of high-end chips, its goal of achieving self-sufficiency by 2030 alarmed U.S. policymakers concerned about losing technological and military advantage. The CHIPS Act emerged as a state-backed attempt to reverse decades of disinvestment by underwriting new domestic manufacturing capacity, research centers, and supply chain coordination.

Yet the Act’s structure reflects continuity with longstanding state-corporate alliances. Of the $52.7 billion authorized, $39 billion was earmarked for manufacturing subsidies, $11 billion for research and development, and $2 billion for defense-specific programs. Oversight was delegated to the CHIPS Program Office within the Department of Commerce and the National Semiconductor Technology Center (NSTC), a public-private consortium coordinating research and workforce development. An Industrial Advisory Committee, comprising leaders from major semiconductor firms, academic institutions, and federal labs, was tasked with guiding implementation. Corporate representation was embedded directly into policy governance.

It would not be inaccurate to say this was an emblematic case of corporate-led policymaking in the name of “public-private partnership.” Parallel advisory bodies, including the President’s Council of Advisors on Science and Technology, institutionalized private sector influence over national technology policy. Figures like Eric Schmidt, former Google CEO, exemplify these entanglements. Through Schmidt Futures, his foundation financed staffing within the Office of Science and Technology Policy and throughout government and policy ecosystems, more illustrative of the pervasiveness of private interests shaping public research agendas than a one-off influence scheme.

The CHIPS Act also aligned semiconductor industrial policy with military procurement objectives. The Department of Defense’s Trusted Foundry and Trusted and Assured Microelectronics programs secured defense-critical chip supply chains. In 2023, $3.5 billion in CHIPS Act appropriations was redirected from research and development toward Intel’s Secure Enclave program, reinforcing the role of domestic incumbents in military supply chains. Additional defense-funded projects, including the Enterprise Parts Management System and the Printed Circuit Board Market Catalyst project, further embedded defense-industrial priorities into semiconductor policy.

These dynamics were not unique to the United States. Similar techno-nationalist industrial strategies were pursued by many others. Europe’s European Chips Act, Japan’s multi-billion-dollar subsidies, South Korea’s incentives to localize materials sourcing, and Taiwan’s ongoing infrastructure investments all aimed to secure positions in global semiconductor value chains. These initiatives reflected a broader geopolitical competition for technological sovereignty, often reinforcing existing hierarchies of dependency and industrial power.

What emerged was a geopolitical industrial boom, marked by state-backed investments of unprecedented scale: Intel’s $100 billion expansion in Ohio, TSMC’s ~$165 billion in Arizona, Samsung’s $98 billion in South Korea, and Micron’s $100 billion in New York. Yet this mobilization of public funds did not disrupt the distribution of power in the industry. Instead, it consolidated the position of incumbent oligopolies while embedding new territorial inequalities—reproducing a global infrastructure of high-tech development structured around monopoly control, geopolitical dependency, and state-capital alliances under the banner of economic security.

Map of Announced Semiconductor Investments, 2021 to 2024

Locked In? Dependency and Consolidation

The semiconductor “crisis” facing the United States is often framed by policymakers as a supply chain failure or a geopolitical risk, a vulnerability exposed by global disruptions or foreign competition. But the real crisis lies deeper: it is the consequence of decades of offshoring, financialization, and political decisions that hollowed out domestic manufacturing in favor of short-term profits and shareholder returns. U.S. dependence on Taiwan’s TSMC was partly geopolitical, partly promoted by large U.S. firms that led the retreat from domestic industrial investment—facilitated by the public policy they promoted that enabled and rewarded disinvestment.

The CHIPS and Science Act is likely poorly constructed corrective to this trajectory, but we can also read it as an extension of it. By funneling over $50 billion in public subsidies to incumbent firms while embedding corporate executives into advisory and governance bodies, the Act reinforces the very structural issues that created the problem. Its alignment with military-industrial priorities and its integration of defense procurement objectives make clear that industrial policy today is not about democratizing production or securing technological sovereignty for the public good. It is about shoring up monopoly control and state-corporate entanglement under the banner of economic security—even more now “national security.”

Globally, the U.S. is not alone. Similar techno-nationalist strategies have emerged across Europe and East Asia, as governments race to secure positions in semiconductor value chains. But these parallel efforts mirror—not challenge—the existing hierarchies of technological accumulation. They entrench a global infrastructure in which a handful of firms and states monopolize advanced manufacturing capacity while peripheral regions remain locked into dependency.

The new semiconductor—and AI—boom promises to transform landscapes, labor markets, and supply chains. But it does so within a political economy structured to prioritize monopoly profits, geopolitical rivalry, and elite governance over public accountability, labor empowerment, or equitable development. Far from heralding a new era of industrial renewal, the CHIPS Act (and whatever comes after) will deepen the very dependencies and inequalities they claim to resolve. If anything has been “secured” in this moment, it is the future of state-backed monopoly power itself—the beating heart of imperial rivalry and competition that led many nations into global war in the early to mid 20th century.

References

1. Block, “Swimming Against the Current: The Rise of a Hidden Developmental State in the United States”; Block and Keller, State of Innovation: The U.S. Government’s Role in Technology Development. The American developmental state traces its roots to the Smithsonian Institution in 1846, evolving over the 19th and 20th centuries to significantly advance science and technology. Key developments included the Department of Agriculture in 1862, the Agricultural Research Service, and the National Institutes of Health in 1887. Labs under the War Department, later transferred to the Department of Energy, played crucial roles in the Manhattan Project during WWII. Federal labs have driven technological breakthroughs like ENIAC, touchscreen technology, VR, GPS, and genomics, including the Human Genome Project. In 2016, over 300 federal labs accounted for 38% of federal R&D funding. Eleven agencies, including the DOD, DOE, HHS, and NASA, manage significant lab operations with various missions. Key legislation, such as the Stevenson-Wydler Technology Innovation Act, the Bayh-Dole Act, and the Federal Technology Transfer Act, has facilitated technology transfer to the private sector, enhancing the impact of federally funded research.

2. Wade, “The Paradox of US Industrial Policy: The Developmental State in Disguise.” Wade also critiques the “varieties of capitalism” literature which some claimed the absence of U.S. industrial policy was due to separation of powers. Wade suggests that the decentralized nature of U.S. industrial policy aligns well with the U.S.’s networked production structure and governance system, potentially enhancing industrial policy effectiveness. See Mann, “Has Globalization Ended the Rise and Rise of the Nation-State?”

3. Castells, The Informational City: Information Technology, Economic Restructuring, and the Urban Regional Process.: 82–88.