Introduction: Intelligence Estates — The Next Evolution of Economic Geography
Every serious student of economic development eventually arrives at the same uncomfortable realization: the map matters as much as the machine. The technologies that define an age are never evenly distributed. They cluster. They concentrate. They cohere into particular places, and those places, in turn, organize the wealth, the talent, and the political power of their era. I first encountered this idea not as an abstraction but as a set of physical drives across Los Angeles. As a graduate student at the University of Southern California in the 1990s, I spent my first years immersed in the literature of regional advantage, and on the days I crossed town — roughly twelve miles from the USC campus near Downtown to the University of California, Los Angeles in Westwood — I was, without quite knowing it, traveling between two intellectual capitals of the same emerging discipline.
At USC, the reading lists circled obsessively around a single, durable insight, captured most famously in Professor AnnaLee Saxenian’s Regional Advantage: Culture and Competition in Silicon Valley and Route 128. [1] Saxenian’s argument — that Silicon Valley out-competed Boston’s Route 128 corridor not because of superior firms but because of a superior regional culture of openness, mobility, and dense informal exchange — reframed competition itself as a property of places rather than companies. At UCLA, where I took a course with the distinguished geographer Allen J. Scott, the lens widened. Scott’s Technopolis: High-Technology Industry and Regional Development in Southern California mapped how aerospace, electronics, and entertainment had woven themselves into the very fabric of Southern California, producing what he called a new kind of industrial space. [2] It was on the UCLA campus that I was urged to read Michael Storper’s The Regional World: Territorial Development in a Global Economy, which argued that regions had become the fundamental units of economic coordination in a globalizing world. [3] And whenever I sat down to write — a seminar paper, a section of a thesis — I returned to the discipline’s commercial cousin, Harvard Business School Professor Michael E. Porter’s Competitive Advantage, whose framework taught a generation to see clusters, value chains, and the deliberate construction of advantage. [4]
Those four books — Saxenian, Scott, Storper, Porter — form the intellectual bedrock of everything that follows. They taught me that the geography of the economy is not a backdrop to innovation; it is the innovation. And they prepared me to notice, decades later, that something fundamentally new was beginning to take shape on the American and global landscape. In my earlier paper, Intelligence Corridors, published on June 2, 2026, I argued that regional ecosystems had begun competing to own a five-layer artificial-intelligence economy, and that this competition would help determine the winners of the twenty-first century. [5] This paper is the sequel and the structural counterpart to that argument. If Intelligence Corridors described the regional flows and competitive dynamics of AI, Intelligence Estates describes the physical estate itself — the purpose-built ecosystem of land, energy, compute, networks, automation, and policy that produces intelligence at industrial scale.
Let me open with a simple but powerful historical observation. Every major technological revolution has reshaped the geography of economic development, and it has done so by inventing a new kind of place to host the dominant mode of production. The progression is unmistakable:
1950s – 1980s │ Industrial Parks
1980s – 2020s │ Technology Parks
2026 and beyond │ Intelligence Estates
During the manufacturing era, nations built Industrial Parks to concentrate factories, transportation, utilities, and labor into efficient production centers. As the digital economy emerged, Industrial Parks gradually gave way to Technology Parks, which brought together software companies, research laboratories, universities, venture capital, and the dense innovation ecosystems that Saxenian and Scott chronicled. Today, the rise of artificial intelligence is beginning a third transformation, and it is a transformation of a different order. Hyperscale AI computing no longer depends primarily on office buildings or conventional business parks. It requires massive quantities of electricity, dedicated fiber networks, water and advanced cooling infrastructure, semiconductor supply chains, AI factories, robotics, and a specialized regulatory and energy-policy environment engineered to support all of it at once.
The next stage of economic evolution is therefore the emergence of what I call Intelligence Estates — purpose-built AI infrastructure ecosystems designed to produce intelligence as an industrial output. I have chosen the word estate deliberately, and for three reasons. First, an estate is a bounded, planned, and self-contained territory — not a scattering of buildings but an integrated whole, master-planned from the electron to the inference. Second, an estate connotes real property and durable value: these are among the most capital-intensive physical assets ever constructed by private enterprise, and like the great industrial estates and landed estates before them, they reorganize the wealth of the regions that host them. Third, an estate implies stewardship and contested ownership — a question of who holds the land, who controls the power, and who captures the value. As this paper will show, that contest is now playing out across American states and across the great powers of the world. The argument, in brief, is that the next generation of economic leadership will be defined not solely by software innovation or by ever-larger AI models, but by the ability to integrate abundant energy, advanced compute, resilient digital networks, automated manufacturing, skilled talent, and forward-looking public policy into cohesive regional ecosystems — Intelligence Estates — in which infrastructure itself becomes the decisive strategic advantage.
Section 1: The Historical Evolution of Economic Geography: From Industrial Parks to Intelligence Estates
To understand why Intelligence Estates represent a genuine break with the past rather than a mere rebranding of the data center, it helps to trace the deep structure of economic geography across three industrial epochs. In each epoch, a dominant general-purpose technology — mechanized production, then digital computation, and now machine intelligence — demanded a characteristic spatial form. Each form concentrated the scarce inputs of its age into a planned territory, and each in turn produced a distinctive economic output. The Industrial Park produced goods; the Technology Park produced knowledge and software; the Intelligence Estate produces intelligence itself. The transitions between them were never instantaneous, and the earlier forms never disappeared entirely. But the center of gravity shifted, and with it shifted the locus of economic power.
1.1 The Industrial Park Era (1950s–1980s)
The classic industrial park was a triumph of mid-century planning logic. Its purpose was to concentrate manufacturing clusters near the arteries of physical distribution — railroads, highways, and ports — and to share the fixed costs of centralized utilities and industrial supply chains across many tenants. The economic geography of this era was organized around the movement of materials: raw inputs flowed in, finished goods flowed out, and the decisive locational variables were transport cost, labor availability, and proximity to markets. The factory town and the planned industrial estate — from the manufacturing belts of the American Midwest to the export-processing zones of Asia — were the spatial signatures of an economy whose fundamental act was making things. Land was valued for its access to logistics; labor was valued for its hands; and the utility grid existed to spin motors and heat furnaces. This was the world Porter’s value-chain analysis was first built to explain, and it remains the substrate on which everything since has been layered.
1.2 The Technology Park Era (1980s–2020s)
As computation became the dominant general-purpose technology, the industrial park gradually evolved into the technology park, and the decisive inputs changed from materials to minds. Silicon Valley between San Jose and San Francisco, the Route 128 corridor outside Boston, and North Carolina’s Research Triangle Park — more than seven thousand acres spread among Raleigh, Durham, and Chapel Hill — became the archetypes of a new spatial form built around universities, research laboratories, venture capital, and the cloud-computing platforms that would later define the knowledge economy. [1] What made these places powerful was not their physical plant but their social density: the rapid recombination of talent, ideas, and capital that Saxenian identified as the true engine of regional advantage. The technology park valued land for its proximity to a great research university and a deep labor pool of engineers; it valued people for their tacit knowledge; and its critical infrastructure was bandwidth and brains. For roughly four decades, this was the model that produced the most valuable companies on earth, and policymakers around the world spent enormous sums trying, with mixed success, to replicate it.
1.3 The Intelligence Estate Era (2026 and Beyond)
The Intelligence Estate marks a third and qualitatively different stage, because the decisive input is no longer materials or minds alone but energy converted into computation. The defining facilities of this era are AI factories: gigawatt-scale compute campuses tied to dedicated energy infrastructure, semiconductor ecosystems, robotics manufacturing, and increasingly orbital and subsea connectivity. The law firm Clifford Chance, surveying the 2026 landscape, captured the shift precisely when it observed that these facilities are being re-imagined as industrial-scale plants that transform raw data and energy into intelligence outputs such as text, code, and video. [56] That sentence is worth dwelling on, because it inverts the entire logic of the technology park. Where the technology park assembled people to produce software, the Intelligence Estate assembles power and silicon to produce cognition, and it does so at a physical scale that has more in common with a steel mill or a nuclear complex than with an office campus. The commercial conversation has shifted accordingly: developers now speak less of square feet and more of time-to-compute and of tokens generated per second under a service-level agreement. [56] Intelligence, in other words, has become a manufactured good — and like every manufactured good before it, it now demands a purpose-built place in which to be made.
Seen through the lens of the three masters of economic geography I read at USC and UCLA, the Intelligence Estate is the synthesis and the supersession of everything that came before. It inherits the industrial park’s obsession with logistics, utilities, and land, except that the critical logistics are now electrons and photons rather than freight. It inherits the technology park’s dependence on universities, capital, and concentrated talent, except that the talent now builds and operates machines that themselves perform cognitive labor. And it raises, with new urgency, the question Storper placed at the center of his work: in a global economy, which territories will succeed in coordinating these new forces, and which will be left to host other people’s infrastructure on other people’s terms? The remainder of this paper takes up that question.
Section 2: Why Artificial Intelligence Requires a New Infrastructure Paradigm
It is tempting to dismiss the AI buildout as simply “more data centers,” a quantitative extension of the cloud-computing infrastructure that has existed for two decades. That view is mistaken, and understanding why it is mistaken is essential to understanding why a new spatial form has become necessary. AI infrastructure in 2026 is not a larger version of the conventional data center; it is a tightly coupled stack of accelerators, high-speed networking, storage, power delivery, and cooling, engineered to run foundation models continuously at scale. [5b] The differences are not marginal. They are differences of kind, and they cascade through every layer of the physical world that supports them — the grid, the water table, the supply chain, the labor market, and the statute book. Taken together, these requirements describe an asset class that simply cannot be accommodated by the office parks and business campuses of the prior era. They describe, instead, the Intelligence Estate.
Gigawatt-scale electricity
The first and most consequential difference is power. A single large-scale AI training facility now requires between one hundred and one thousand megawatts of continuous electricity — the consumption of a small city, and on the order of the output of a large nuclear plant for the biggest campuses. [9b] In 2026, the first wave of true gigawatt-scale AI campuses is moving from announcement to operation: according to research by Epoch AI, five data centers at a scale of one gigawatt or more are expected to come online during the year, each operated by a different hyperscaler, with xAI’s Colossus 2 in Memphis targeting a remarkable twelve-month buildout to reach gigawatt scale. [14] The International Energy Agency reports that global data-center electricity demand grew seventeen percent in 2025, while consumption from AI-focused data centers surged fifty percent — each far outpacing the roughly three percent growth in overall global electricity demand. [8] The IEA projects that data-center electricity consumption will roughly double from about 485 terawatt-hours in 2025 to some 950 terawatt-hours by 2030, accounting for around three percent of global electricity by that date and, in the United States, for nearly half of all the growth in electricity demand. [8] The IEA’s Executive Director Dr. Fatih Birol has reduced the entire dynamic to a single principle:
“there is no AI without energy”
— Dr. Fatih Birol, Executive Director, International Energy Agency [9]
Power, not chips, has become the binding constraint. The North American Electric Reliability Corporation warned in early 2026 that projected summer peak demand had surged largely on the back of new AI and digital-economy load, while analysis presented to the PJM grid operator pointed to a possible forty-nine-gigawatt generation shortfall by 2028. [9b] The defining problem is a timing mismatch: a hyperscale campus can be sited, built, and commissioned in eighteen to thirty-six months, but the transmission infrastructure needed to power it can take five to ten years to plan, permit, and build. [14] This is precisely why the energy question, once an afterthought of site selection, has migrated to the very center of the Intelligence Estate.
Massive GPU clusters and the collapse of conventional density
The second difference is physical density. Training the largest models requires connecting tens of thousands of graphics processing units into tightly coupled clusters in which latency and proximity are decisive. [5b] The result is a staggering escalation of power density at the rack. Where racks once drew thirty to forty kilowatts, designs in 2026 are measured in the hundreds of kilowatts and are approaching the megawatt range, with a single rack of the latest NVIDIA systems drawing up to one hundred and forty kilowatts and next-generation systems expected to exceed two hundred. [7] The IEA has documented an eleven-fold increase in AI-server power density between 2020 and 2025. [7] No conventional building was designed for this. The escalation forces a wholesale redesign of electrical distribution, structural loading, and — above all — heat removal.
Dedicated fiber, water, and advanced cooling
The third cluster of differences concerns the supporting physical envelope. Training clusters must be lashed together by dedicated, high-bandwidth fiber, both within the campus and across the multi-region footprints over which distributed training increasingly runs. At the same time, the heat produced by hundred-kilowatt racks has rendered traditional air cooling obsolete; liquid cooling has shifted from an interesting option to the default starting point for new AI capacity. [8c] Liquid cooling, in turn, raises the question of water. Northern Virginia’s data centers alone consumed close to two billion gallons of water in 2023, a roughly sixty-three percent jump from 2019, and the resulting strain has made water availability a first-order political constraint in markets from Virginia to central Texas. [39] This is why the most sophisticated new estates are being engineered around closed-loop water systems and behind-the-meter generation from the outset, rather than retrofitting them after the fact.
AI factories, robotics, workforce, and state incentives
The remaining differences are organizational and institutional. Hyperscalers no longer optimize individual buildings; they treat entire campuses as integrated products, deployed in large synchronized increments across the supply chain. [5b] The estates increasingly sit alongside, or incorporate, advanced manufacturing — the semiconductor fabs and robotics lines that feed and extend them — and they depend on a specialized workforce of electrical engineers, network operators, data-center technicians, and skilled construction trades that few regions possess in adequate numbers. Finally, and decisively, they depend on the state. The scale of capital, power, land, and permitting involved means that no Intelligence Estate is built without the active collaboration of governors, utilities, regulators, and universities. Where the technology park could be coaxed into existence with a research grant and a tax abatement, the Intelligence Estate requires the coordinated mobilization of an entire regional apparatus. By the time a reader has absorbed the full list of requirements — gigawatt power, massive GPU clusters, dedicated fiber, water, advanced cooling, AI factories, robotics, specialized labor, and bespoke state support — the conclusion writes itself. This is not the data center grown large. It is a new type of infrastructure, and it demands a new type of place.
Section 3: Designing the Intelligence Estate — The Five-Layer Framework
If the Intelligence Estate is a new spatial form, then it requires a design language — a way of decomposing the whole into its essential systems so that planners, governors, investors, and scholars can reason about it coherently. This section introduces the framework that sits at the heart of this paper: the Five-Layer Intelligence Estate Framework. The framework holds that a fully realized Intelligence Estate integrates five stacked layers of infrastructure, each of which must be present, adequately scaled, and tightly coupled to the others. A region that possesses four of the five layers does not possess four-fifths of an Intelligence Estate; it possesses an incomplete one, and incompleteness in any single layer becomes the binding constraint on the whole. This is the central design insight, and it mirrors a lesson Porter taught long ago about value chains: advantage accrues to those who integrate the entire system, not to those who optimize a single link. The five layers, from the ground up, are energy, compute, digital, automation, and economic infrastructure.
Layer 1 — Energy Infrastructure
Energy is the foundation, and in the Intelligence Estate it is foundational in the literal sense that nothing above it can function without it. This layer comprises generation, transmission, on-site and behind-the-meter power, storage, and the cooling systems that manage the thermal consequences of dense compute. The economics here are arresting. The IEA noted that global investment in data centers was expected to reach roughly five hundred and eighty billion dollars in 2025, and Dr. Birol drew the comparison that gives the era its character — that this figure surpassed the roughly five hundred and forty billion dollars being spent on global oil supply, a striking marker of the changing nature of modern economies. [54] The capital expenditure of just five technology companies is now larger than global investment in oil and natural-gas production. [8] Because the grid cannot keep pace, the energy layer increasingly internalizes its own generation: the pipeline of conditional offtake agreements between data-center operators and small modular nuclear projects grew from twenty-five gigawatts at the end of 2024 to forty-five gigawatts by early 2026, even as developers advance a large number of on-site natural-gas projects, particularly in the United States. [8] The estate that controls its electrons controls its destiny; the one that does not is merely a tenant of the grid.
Layer 2 — Compute Infrastructure
The compute layer is the productive core: the accelerators, servers, memory, and the racks, power shelves, and liquid-cooling loops that house them. This is the layer that most directly converts energy into intelligence, and its scale is best read in the results of the company that supplies most of the world’s AI silicon. In its first fiscal quarter of 2027 — the quarter ended April 26, 2026, and the most recent reported as of this writing — NVIDIA posted record revenue of $81.6 billion, up eighty-five percent year over year, of which data-center revenue alone was a record $75.2 billion, up ninety-two percent. [11] Founder and chief executive Jensen Huang framed the moment in language that could serve as the motto of the entire Intelligence Estate thesis, describing the present buildout as
“the largest infrastructure expansion in human history”
— Jensen Huang, Founder and CEO, NVIDIA [10]
The phrase is not marketing hyperbole. It is a reasonable description of a capital cycle in which the four largest American hyperscalers spent roughly $413 billion on capital expenditure in 2025 — an eighty-four percent jump from the prior year — and are collectively guiding toward between six hundred and seven hundred billion dollars in 2026. [14] The compute layer is also where the supply chain bites hardest: high-bandwidth memory is sold out through 2026, lead times are long, and the cost of the input itself is rising, which is one reason hyperscaler capital budgets have climbed even faster than capacity. [28b]
Layer 3 — Digital Infrastructure
The digital layer is the connective tissue: the dedicated fiber, optical networking, subsea and increasingly orbital links, and the storage and data pipelines that move information into and out of the estate and that bind multi-region campuses into single logical machines. As training workloads distribute across geographies and inference workloads demand low-latency responsiveness at the edge, the optical and network fabric becomes a differentiator in its own right. The networking portion of NVIDIA’s data-center business alone reached a record $14.8 billion in the most recent quarter, up nearly two hundred percent year over year — a figure that signals how thoroughly the network has become part of the compute system rather than a mere accessory to it. [11] For nations and regions, the digital layer carries a sovereignty dimension as well: control over where data is stored, processed, and transmitted is what makes domestic law enforceable, and it is increasingly treated, like the power grid, as critical national infrastructure.
Layer 4 — Automation Infrastructure
The fourth layer is what most decisively distinguishes the Intelligence Estate from any prior spatial form: automation. This layer encompasses the robotics, AI-driven manufacturing, and the semiconductor and advanced-packaging facilities that both supply the estate and represent the physical embodiment of what it produces. The estate does not merely consume intelligence; it manufactures the machines that extend it. NVIDIA itself has reorganized its business around this reality, splitting into a Data Center platform and an Edge Computing platform spanning robotics, automotive, and physical AI — the company’s shorthand for intelligence that acts in the world. [11] The automation layer is why the most ambitious Intelligence Estates are being co-located with, or built adjacent to, chip fabrication and robotics manufacturing: the production of intelligence and the production of the bodies that carry it are converging into a single industrial complex. A region that hosts compute without automation hosts a warehouse; a region that integrates both hosts a factory of the future.
Layer 5 — Economic Infrastructure
The fifth and outermost layer is the one most often neglected by engineers and most familiar to economic geographers: the institutional and economic scaffolding that makes the other four layers possible. This layer comprises state incentives and permitting regimes, utility partnerships, university talent pipelines, capital markets, workforce-training systems, and the regulatory frameworks that govern energy cost-allocation, water use, and community impact. It is the layer where Saxenian’s regional culture, Scott’s institutional thickness, and Porter’s deliberate construction of advantage reassert themselves. As this paper’s later sections will show, it is precisely at the economic-infrastructure layer that American states and global powers are now competing most fiercely — because the physical layers, while extraordinarily expensive, are increasingly portable, while the institutional capacity to site, power, permit, staff, and legitimize a gigawatt-scale estate is not. The five layers together form a stack in which value flows upward from energy to economic output, and in which fragility flows downward from any missing piece. This is a highly publishable framework precisely because it is falsifiable: it predicts that the winners of the Intelligence-Estate era will be those regions and nations that achieve integration across all five layers, and that those who excel in only one or two — cheap power without talent, or brilliant research without electrons — will find their ambitions capped by their weakest layer.
Section 4: Building America’s Intelligence Estates
Nowhere is the construction of Intelligence Estates more visible, more contested, or more revealing than in the laboratory of American federalism. Because energy markets, permitting authority, water rights, and economic-development incentives in the United States are largely controlled at the state level, the American buildout has become a fifty-way competition in which governors, utilities, universities, and hyperscalers negotiate the terms of the AI economy region by region. The federal government has set the strategic frame. In July 2025 the White House released Winning the Race: America’s AI Action Plan, organized around three pillars — accelerating innovation, building American AI infrastructure, and leading in international diplomacy and security — and accompanied by executive orders that streamlined federal permitting for data centers and energy infrastructure and that opened federal land to qualifying projects. [24] The administration’s framing was unambiguous; as AI and Crypto Czar David Sacks put it,
“Winning the AI Race is non-negotiable”
— David Sacks, White House AI and Crypto Czar [24]
But it is the states that actually build. What follows is a tour of the regions where America’s Intelligence Estates are taking physical shape, and of the governors whose negotiations are defining the terms of the estate — and increasingly, the terms of the backlash.
Texas: the self-described epicenter
Texas has positioned itself as the leading American Intelligence Estate, and Governor Greg Abbott has not been shy about the claim, describing the state as the “epicenter of AI development” and likening the rush of capital to the California gold rush of 1849. [26] The flagship is the Stargate campus in Abilene, the first site of the $500 billion OpenAI–Oracle–SoftBank venture, a 875-acre, eight-building site engineered for 1.2 gigawatts of power. [25] Crusoe co-founder Cully Cavness, whose firm is building the Abilene project, has said that Texas has been
“one of the best and easiest places to move those projects forward”
— Cully Cavness, Co-Founder, Crusoe [28]
The pipeline behind it is vast: Texas is home to roughly four hundred data centers with some ninety more planned or under construction, and ERCOT, the state grid operator, reports more than 480 large data centers seeking to connect by 2032. [28] In November 2025, alongside Google chief executive Sundar Pichai, Abbott announced a $40 billion Google investment across three AI data-center campuses. [26] By May 2026, Texas had overtaken Northern Virginia as the world’s top primary data-center market, with single-site projects such as CloudBurst’s 1.2-gigawatt campus near San Marcos and Vantage’s $25 billion, 1.4-gigawatt campus in Shackelford County moving into construction. [29] Yet Texas also illustrates the gathering counter-pressure. In June 2025, Abbott signed Senate Bill 6, requiring large-load customers above seventy-five megawatts to contribute to grid-interconnection costs [27]; and in June 2026, in what observers called a striking reversal of his pro-business posture, he directed the Public Utility Commission and ERCOT to ensure that ordinary Texans do not pay for the electrical infrastructure that data centers require, calling additionally for water-efficient cooling mandates, annual consumption reporting, and the repeal of generous sales-tax exemptions. [28] The estate, in Texas, is being built and being fenced at the same time.
California: the regulator of the frontier
California occupies a singular position. It is simultaneously the intellectual birthplace of modern AI — home to thirty-two of the fifty top AI companies worldwide and, per the 2025 Stanford AI Index, to 15.7 percent of all U.S. AI job postings, far ahead of Texas at 8.8 percent [30] — and the most assertive American regulator of the technology. In September 2025, Governor Gavin Newsom signed Senate Bill 53, the Transparency in Frontier Artificial Intelligence Act, which the Brookings Institution described as the first enforceable regulatory framework in the United States for the most advanced AI systems, requiring large frontier developers to publish their safety frameworks. [31] In May 2026, Newsom went further, issuing a first-of-its-kind executive order directing state agencies to prepare workers and communities for AI-driven economic disruption, exploring severance standards, transition support, and updates to the state’s mass-layoff notification law. [32] California’s wager is distinctive: rather than competing primarily to host the largest physical estates, it competes to govern the frontier, betting that the regional advantage of the future lies in setting the rules of trust and talent for an industry that was born within its borders. It is the clearest American example of leadership concentrated in the fifth, economic-and-institutional, layer of the framework.
Virginia: the incumbent capital
Northern Virginia remains the incumbent capital of global data infrastructure. Loudoun and Fairfax counties’ “Data Center Alley” is the densest concentration of data centers on earth, hosting close to fifty million square feet of capacity and more than 4,900 megawatts of commissioned power, with an estimated seventy percent of global internet traffic passing through it. [39] Virginia today hosts roughly six hundred data centers, more than half of them in Data Center Alley, and the fiscal dependence runs deep: Loudoun County collects more than half of its property taxes from data centers. [38] That dependence has bred both ambition and resistance. Governor Glenn Youngkin announced a $35 billion AWS expansion across new campuses, even as the proposed $67 billion Dominion–NextEra utility merger — which would house more than ten percent of all U.S. electric-utility capacity — signaled how thoroughly the region’s energy future has become entangled with its compute future. [40] At the same time, Virginia became an early site of governance experiments, with Loudoun ending “by-right” data-center development in March 2025 so that new applications now require legislative approval. [41] Virginia is the cautionary archetype: a region that mastered the physical layers so completely that it must now wrestle with the social and political limits of the estate it built.
Arizona: the silicon foundation
Arizona is building a different layer of the stack. Where Texas and Virginia host compute, Arizona hosts the automation and semiconductor base on which compute depends. In March 2025, the Taiwan Semiconductor Manufacturing Company announced an additional $100 billion investment, bringing its total Phoenix commitment to $165 billion — the largest single foreign direct investment in U.S. history — encompassing six fabrication plants, two advanced-packaging facilities, and a research-and-development center. [42] The strategic significance is that advanced packaging on American soil completes the domestic AI chip supply chain, with the first fab producing four-nanometer chips for customers including Apple and NVIDIA and later fabs targeting two-nanometer production toward the end of the decade. [42] Arizona demonstrates that an Intelligence Estate need not host the data center to be indispensable to it; a region that controls the fourth layer — the fabs and packaging that make the silicon — holds leverage over every estate downstream.
Pennsylvania: “bring your own energy”
Pennsylvania has emerged as the bellwether of the American debate over who pays for the estate. In June 2025, Governor Josh Shapiro announced a $20 billion Amazon investment to build data-center campuses — the largest private-sector investment in the commonwealth’s history — declaring that Pennsylvania was “all in on AI.” [33] His rationale rested squarely on the energy layer:
“We have the energy resources to support this technology”
— Governor Josh Shapiro, Pennsylvania [33]
The commonwealth’s gas endowment has made it a magnet: at the site of the former Homer City coal plant, developers are building what is slated to be the largest gas-fired power plant in the United States to feed a data-center campus, part of a planned data-center investment that reporting has pinned at roughly one hundred billion dollars statewide. [35] But as household electricity rates rose nearly fourteen percent in a single year and community opposition mounted, Shapiro recalibrated. In his February 2026 budget address and the subsequent GRID standards, he advanced a “bring your own energy” principle requiring developers to build, bring online, or buy the incremental capacity they consume and to pay its full cost, alongside transparency, workforce, and environmental commitments. [34] Pennsylvania crystallizes the central political question of the Intelligence Estate: whether the estate pays its own way, or whether its costs are socialized onto the households around it.
Michigan: the reindustrialization bet
Michigan represents the Intelligence Estate as a vehicle for industrial renewal. Governor Gretchen Whitmer began courting hyperscalers in late 2024 and, by her own account, reached out to OpenAI in February 2025 to pursue a Stargate site. [57] The result, announced in October 2025, was a multi-billion-dollar campus in Saline Township that she called the largest investment in Michigan’s history, expected to create some 2,500 union construction jobs and 450 permanent positions. [36] In May 2026, OpenAI, Oracle, Related Digital, and the utility DTE broke ground on “The Barn,” a one-gigawatt campus, with explicit commitments that local ratepayers would not bear the cost of the supporting infrastructure and that the facility would use closed-loop cooling. [37] Whitmer framed the bet in the language of reindustrialization:
“betting on Michigan”
— Governor Gretchen Whitmer, Michigan [36]
Michigan, like its Midwestern neighbors, offers the combination the estate craves — available land, fresh water, existing transmission, and a skilled construction workforce — and in October 2025 alone the utilities DTE and Consumers Energy announced deals for 6.4 gigawatts of data-center power. [38]
Indiana and Ohio: the Midwestern frontier
Indiana and Ohio round out the Midwestern frontier, and they illustrate both the opportunity and its limits. Ohio now hosts roughly 185 data centers, and Meta’s Prometheus campus in New Albany — the company’s first gigawatt-scale facility, paired with an on-site two-hundred-megawatt natural-gas project — is under construction. [29] Yet the Midwest is also where community resistance has produced outright refusals: in July 2025 the mayor of Michigan City, Indiana, rejected an $800 million data center for lacking meaningful job commitments and community benefits. [38] Across these states, the same pattern recurs. The physical estate can be financed and engineered almost anywhere with land, water, and power; what determines whether it is welcomed is the fifth layer — the negotiated bargain among developer, governor, utility, and community over who captures the value and who absorbs the cost. The American experience, taken as a whole, demonstrates that the Intelligence Estate is not merely an engineering project but a political settlement, renegotiated in every region it touches.
Section 5: The Global Race for Intelligence Estates
The competition that plays out among American states is replicated, at higher stakes, among the great powers. As 2026 has progressed, a striking consensus has formed among analysts: the primary bottleneck in artificial intelligence is no longer algorithmic but physical — a matter of hyperscale data centers, electrical grids, cooling, and silicon — and “AI sovereignty” has migrated from a technology buzzword to a question of national security. [44] Worldwide AI spending is forecast to reach roughly $2.5 trillion in 2026, and sovereign-AI initiatives proliferated across the EU, Canada, the Gulf, and Asia in the first quarter alone, each following a recognizable playbook of a national fund, a cloud or telecommunications partner, a GPU allocation, and a data-sovereignty clause. [45] The Center for a New American Security’s Sovereign AI Index found that infrastructure projects — data centers, supercomputers, GPU clusters — now constitute fifty-nine percent of all tracked sovereign-AI projects, but that investment is heavily concentrated: the Middle East and East Asia together account for more than eighty percent of disclosed sovereign-AI spending, and the United Arab Emirates and Japan alone for more than two-thirds. [48] The sections below survey the principal contenders.
5.1 United States
The United States enters this race from a position of structural dominance. It hosts the world’s leading hyperscalers, the dominant chip designer, and the frontier model labs, and roughly three-quarters of the more than twenty-three gigawatts of data-center capacity under construction globally at the end of September 2025 was rising on American soil. [6] That lead is sustained by an unmatched capital engine: according to first-quarter 2026 earnings compiled by the Financial Times, Google, Amazon, Microsoft, and Meta collectively plan to spend $725 billion on capital expenditure in 2026, up seventy-seven percent from 2025’s record $410 billion. [12] Microsoft alone set 2026 capital expenditure at roughly $190 billion, Amazon near $200 billion, Alphabet at $180–190 billion, and Meta at $125–145 billion. [13] The defining American asset is the conversion of that spending into revenue: Google Cloud grew sixty-three percent year over year to $20 billion with a backlog approaching $462 billion, which Alphabet’s chief financial officer Anat Ashkenazi attributed to “unprecedented internal and external demand for AI compute resources.” [13] Jefferies analyst Brent Thill captured the prevailing view of those who believe the spending is justified by the revenue beneath it:
“The bear thesis is garbage”
— Brent Thill, Analyst, Jefferies [12]
Anchoring the American effort is Stargate, the $500 billion, ten-gigawatt joint venture among OpenAI, SoftBank, Oracle, and MGX announced at the White House in January 2025, which by late 2025 had reached nearly seven gigawatts of planned capacity and over $400 billion of committed investment across sites in Texas, New Mexico, Ohio, Michigan, and beyond. [15] OpenAI’s Sam Altman framed the venture’s logic in terms that echo the thesis of this paper:
“AI can only fulfill its promise if we build the compute to power it”
— Sam Altman, CEO, OpenAI [15]
5.2 China
China is pursuing the most fully sovereign strategy of any nation, by necessity as much as by design. Subjected to severe export restrictions on advanced chips, Beijing has reorganized its AI supply chain around domestic hardware — Huawei’s Ascend accelerators and other local silicon now power state-oriented server farms — and around state-coordinated vertical integration rather than the corporate competition that characterizes the American ecosystem. [44] Its flagship spatial program, “Eastern Data, Western Computing,” engineers mega-clusters in the country’s western provinces, where land and renewable energy are abundant, and pipes the compute eastward to the densely populated coast — a continental-scale infrastructure project that is, in effect, a national Intelligence Estate. [44] Chinese cloud and platform giants are matching the capital intensity of their Western rivals: Alibaba announced it would invest at least $52 billion in cloud and AI infrastructure over three years, and ByteDance may spend roughly $30 billion on AI infrastructure in 2026 alone. [51] The Stanford AI Index notes that while the United States retains a decisive lead in private investment, the release of powerful Chinese models means American advantages cannot be taken for granted.
5.3 Europe
Europe’s position is paradoxical: it leads the world in the legal architecture of AI while suffering a chronic deficit in hard infrastructure. [44] The European Union has moved to close that gap through its AI Continent Action Plan, which envisions setting up at least nineteen AI Factories across the continent and establishing up to five AI Gigafactories — large-scale facilities each designed to house around one hundred thousand advanced processors, roughly four times as powerful as the earlier AI Factories — backed by an InvestAI facility intended to mobilize twenty billion euros. [46] In January 2026 the Council formally amended the EuroHPC regulation to authorize the Gigafactories, with phase-one construction set to begin in the third quarter of 2026. [45] France has gone furthest among member states: President Emmanuel Macron announced some 109 billion euros in total AI investment, framing the effort explicitly as a contest for autonomy:
“our fight for sovereignty, for strategic autonomy”
— President Emmanuel Macron, France [53]
Europe’s distinctive bet is the publicly anchored, sovereignty-first estate — open to researchers, startups, and small firms rather than reserved for hyperscale enterprise customers — a model that reflects the continent’s preference for treating compute as a public good even as it scrambles to assemble enough electrons to power it.
5.4 The Middle East
The Gulf states have made the most aggressive bid to convert hydrocarbon wealth into compute. The United Arab Emirates is developing what it describes as the largest AI campus outside the United States — a five-gigawatt facility in Abu Dhabi — while Microsoft has committed $15.2 billion to the UAE through its partnership with G42, and Saudi Arabia’s state-backed HUMAIN has announced a $77 billion strategy targeting 1.9 gigawatts of capacity by 2030. [47] The structural advantages are formidable: sovereign wealth funds can deploy capital without the fundraising cycles that delay private developers; electricity costs of roughly five to six cents per kilowatt-hour undercut the American average; and the region’s position between Europe, Africa, and Asia, with subsea connectivity to all three, gives it genuine geographic value. [47] Regional data-center capacity is projected to more than triple from about one gigawatt in 2025 to over three gigawatts by 2030. [47] The Gulf exemplifies a nation deliberately exporting compute capacity where it once exported oil — the purest case of the Intelligence Estate as an instrument of economic transformation.
5.5 Japan
Japan is, by the CNAS measure, one of the two largest sovereign-AI investors in the world. [48] Its strategy combines a national sovereign supercomputer — the AI Bridging Cloud Infrastructure, ABCI 3.0 — with a government commitment of roughly one trillion yen annually for AI and semiconductor development. [50] Japan’s energy constraint is acute; the IEA projects that data centers could account for more than half of the country’s electricity-demand growth, which makes the energy layer the decisive variable in Japanese planning. [7] Japan illustrates how an advanced economy with limited domestic energy headroom must treat the first layer of the framework as its binding constraint, and must compensate through efficiency and partnership.
5.6 South Korea
South Korea has mounted one of the most comprehensive national programs, anchored by a sweeping sovereign-AI initiative and a 2026 national AI budget of 9.9 trillion won — roughly $6.7 billion — with nearly half directed to infrastructure. [50] Its AI Basic Act took effect in January 2026, and the government elevated AI to cabinet-level priority by appointing the country’s first senior presidential secretary dedicated to AI, who has championed a vision of “sovereign AI that has learned from Korea’s culture and history.” [49] Korea’s plan envisions building toward dozens of data centers and a multi-gigawatt power expansion, leveraging the same industrial coordination that once gave it a commanding share of the global semiconductor market. [49]
5.7 India
India is pursuing what analysts describe as the democratization of AI compute, combining a national mission with vast private buildouts. The Adani Group has announced a roughly $100 billion push into AI and energy infrastructure, and the world’s largest single data center under construction is rising in Jamnagar, with NVIDIA deeply involved — a facility that could ultimately draw electricity on the scale of the ten million people living in its surrounding region. [52] The Asia-Pacific region as a whole is shifting from a spillover market into a demand engine in its own right; McKinsey estimates that keeping pace with demand could require some $6.7 trillion of cumulative capital across the global data-center value chain between 2025 and 2030, and projects that Asia-Pacific could account for roughly thirty-four percent of global data-center demand by 2030. [51]
A sobering thread runs through every national story. The CNAS Sovereign AI Index found that NVIDIA alone supplies the GPUs for fifty-two percent of all tracked sovereign-infrastructure projects, that roughly seventy percent of those projects involve at least one foreign partner, and that four-fifths of those partnerships involve an American company. [48] In practice, then, “sovereignty” at the data-center layer rarely extends to the silicon layer beneath it. The global race for Intelligence Estates is real, but it is being run, for now, largely on American rails — a dependency that is itself becoming a central axis of twenty-first-century geopolitics, and that explains why the United States has organized an entire pillar of its AI Action Plan around exporting the full American technology stack to allied nations.
Section 6. The Strategic Pillars of the Intelligence Estate
Having traced the historical lineage of the Intelligence Estate, the technical reasons for its distinctiveness, its five-layer architecture, and its construction across American states and rival nations, this section distills the analysis into a set of strategic pillars. Where the Five-Layer Framework of Section 3 describes the anatomy of an Intelligence Estate — what it is made of — these pillars describe its strategy — what a region or nation must get right to build and sustain one. I have expanded the original five into seven, because the experience of 2025 and 2026 has revealed two additional pillars, talent and social license, without which the others cannot stand.
Pillar 1 — Purpose-Built Infrastructure
The first pillar is the recognition that the Intelligence Estate must be designed as an integrated whole, not assembled piecemeal from repurposed parts. The industry has crossed the threshold the engineers at Data Center World 2026 described, in which hyperscalers treat the entire campus as a product that must balance flexibility, scale, and rapid deployment across multiple hardware generations. [5b] Purpose-built means designing for the megawatt rack, the closed-loop cooling loop, and the behind-the-meter generator from the first drawing — because retrofitting these into a conventional facility is, in most cases, impossible. The estate is engineered backward from the electron to the inference, and the regions that internalize this principle build faster and cheaper than those that learn it the hard way.
Pillar 2 — Integrated Intelligence Ecosystems
The second pillar holds that value accrues to integration across all five layers, not to excellence in any one. This is the Porterian lesson restated for the AI age: a region with cheap power but no talent, or brilliant universities but no electrons, will find its weakest layer governing its outcome. The most successful estates — Abilene’s Stargate flagship, the Saline campus, the TSMC–compute complex forming across the American Southwest — succeed because they bind energy, compute, networking, automation, and institutional support into a single coordinated system. The implication for policymakers is that fragmented incentives produce fragmented estates; integration must be designed, not hoped for.
Pillar 3 — Energy as the Foundation of Intelligence
The third pillar elevates energy from an input to the foundation of competitive advantage. The IEA’s data make the case unanswerable: data-center investment surpassed global oil-supply investment in 2025, and the binding constraint on the entire sector has shifted from capital to power. [54] As Dr. Birol has observed, AI is becoming not only an energy taker but an energy maker, driving the commercialization of next-generation nuclear reactors, flexible data centers, and long-duration storage; the tech sector accounted for roughly forty percent of all corporate renewable power-purchase agreements signed in 2025. [8] The strategic conclusion follows directly: the nation or region that secures secure, affordable, and rapid access to electricity will, in Birol’s words, be one step ahead — and the one that does not will find its ambitions stranded behind an interconnection queue.
Pillar 4 — Regional and National Competitiveness
The fourth pillar recognizes that Intelligence Estates have become the principal arena of regional and national economic competition. The fifty-state American contest and the multi-continental sovereign-AI race are two expressions of the same dynamic: jurisdictions are competing to capture the jobs, tax base, and strategic capability that the estates confer. But the competition is double-edged. The same Stanford AI Index that documents America’s investment lead also records that countries with the highest AI investment, including the United States, express more public skepticism about AI than countries spending far less [18] — a reminder that competitiveness is not only about attracting capital but about sustaining the public consent on which capital depends.
Pillar 5 — Infrastructure-Driven Economic Leadership
The fifth pillar is the thesis of this paper in its most compressed form: that economic leadership in the AI era will be infrastructure-driven. For a generation, the commanding heights of the economy were occupied by those who owned the software and the models. Increasingly, they will be occupied by those who own the physical means of producing intelligence — the estates, the energy, the silicon. This is why the world’s most valuable company is now a chipmaker, why oil-rich states are racing to become compute-rich states, and why governors stake their political futures on landing a single campus. Infrastructure, long treated as the unglamorous substrate of the economy, has become its commanding height.
Pillar 6 — Talent and Workforce as Critical Infrastructure
The sixth pillar, which the events of 2025 and 2026 forced onto the agenda, is talent. The estate cannot be built or run without electrical engineers, network operators, data-center technicians, and skilled construction trades, and the regions winning the largest projects — Michigan with its union building trades, the EU with its planned vocational academies training forty-five thousand technicians by 2028 — are those that treat workforce development as infrastructure rather than as an afterthought. [45] The labor dimension is also where the estate connects to the broader anxieties of the age. The IMF’s managing director, Kristalina Georgieva, has warned that AI will affect some sixty percent of jobs in advanced economies and forty percent globally, and her injunction to the world’s leaders was blunt:
“AI is for real and it is transforming our world”
— Kristalina Georgieva, Managing Director, IMF [21]
The estates that endure will be those that build their workforce pillar deliberately, ensuring that the communities helping to construct the infrastructure of intelligence also share in the opportunities it creates.
Pillar 7 — Governance and Social License
The seventh and final pillar is the one most often missing from engineering accounts and most decisive in practice: the social license to operate. A Gallup poll released in May 2026 found that more than seventy percent of Americans oppose the construction of an AI data center in their local area, citing electricity costs and water consumption. [26] The backlash is bipartisan and global, expressed in moratoriums, permitting delays, and outright rejections from Indiana to Ireland. The governance question — who pays for the grid, who controls the water, who captures the value — has become the gating factor on the entire buildout, which is why Texas, Pennsylvania, and California have each moved to impose cost-allocation rules, “bring your own energy” mandates, and transparency requirements. An Intelligence Estate without social license is a stranded asset waiting to happen; the pillar of governance is what converts a contested project into a durable one.
Conclusion: Infrastructure as the Commanding Height of the AI Economy
Return, for a moment, to the historical progression with which this paper began. Industrial Parks transformed manufacturing by concentrating the inputs of the machine age — materials, logistics, and labor — into planned territories of production. Technology Parks accelerated the digital economy by concentrating the inputs of the knowledge age — talent, capital, and research — into the regional ecosystems that Saxenian, Scott, and Storper taught us to see. The Intelligence Estate now concentrates the inputs of the machine-intelligence age — energy, silicon, networks, and automation — into purpose-built ecosystems that produce intelligence at industrial scale. This is why I have named the framework as I have. The estate is the physical foundation of the artificial-intelligence economy, and like the estates of every prior era, it reorganizes wealth, power, and geography around itself.
The economic significance of this shift should not be overstated, and the framework would be weaker if it ignored the dissenting voices. MIT’s Daron Acemoglu has argued, in the most influential skeptical analysis of the period, that the aggregate productivity gains from AI over the coming decade are likely to be “nontrivial but modest” — on the order of well under one percent of total factor productivity [19] — a sharp corrective to the more exuberant forecasts. Whether one sides with Acemoglu’s caution or with the more optimistic estimates, the infrastructure thesis holds either way: even a modest productivity revolution requires an immense physical apparatus to deliver it, and the regions that own that apparatus will capture a disproportionate share of whatever value it creates. The estate is the toll road of the AI economy, and toll roads earn whether traffic grows quickly or slowly.
The deeper lesson is the one that economic geography has taught since its inception, and that this paper has tried to carry forward into a new age. The next generation of economic leadership will not be defined solely by software innovation or by larger AI models, but by the ability to integrate abundant energy, advanced compute, resilient digital networks, automated manufacturing, skilled talent, and forward-looking public policy into cohesive regional ecosystems. Intelligence Estates represent this new model of development, in which infrastructure itself becomes a strategic advantage and the production of intelligence emerges as a defining engine of economic growth. The contest to build them — among American states and among the great powers — is, at bottom, a contest over the geography of the twenty-first century. The map, as ever, will matter as much as the machine. And the regions that learn to build, power, staff, and legitimize their Intelligence Estates will be the ones that own the century that these estates are now bringing into being.
Endnotes and Sources:
[1] AnnaLee Saxenian, Regional Advantage: Culture and Competition in Silicon Valley and Route 128 (Cambridge, MA: Harvard University Press, 1994). https://www.hup.harvard.edu/books/9780674753400
[2] Allen J. Scott, Technopolis: High-Technology Industry and Regional Development in Southern California (Berkeley: University of California Press, 1993). https://www.ucpress.edu/books
[3] Michael Storper, The Regional World: Territorial Development in a Global Economy (New York: Guilford Press, 1997). https://www.guilford.com
[4] Michael E. Porter, Competitive Advantage: Creating and Sustaining Superior Performance (New York: Free Press, 1985). https://www.simonandschuster.com
[5] Stefanus.AI, “Intelligence Corridors: How Regional Ecosystems Are Competing to Own the Five-Layer AI Economy,” June 2, 2026. https://stefanus.ai/intelligence-corridors-how-regional-ecosystems-are-competing-to-own-the-five-layer-ai-economy-and-why-the-new-geography-of-artificial-intelligence-will-determine-the-winners-of-the-twenty-f/
[5b] Data Center Knowledge, “Data Center World 2026: AI Pushes Infrastructure to New Limits” (rack density; campus-as-product; training vs. inference), May 2026. https://www.datacenterknowledge.com/build-design/data-center-world-2026-ai-pushes-infrastructure-to-new-limits
[6] BloombergNEF, “AI Data Center Build Advances at Full Speed: Five Things to Know,” March 2026. https://about.bnef.com/insights/commodities/ai-data-center-build-advances-at-full-speed-five-things-to-know/
[7] Harian Basis / Epoch AI, “Five Gigawatt-Scale AI Data Centers Set to Launch in 2026” (NVIDIA rack density; IEA 11-fold figure), May 2026. https://www.harianbasis.co/en/gigawatt-scale-ai-data-centers-2026
[8] International Energy Agency, “Key Questions on Energy and AI — Executive Summary,” April 2026 (Fatih Birol). https://www.iea.org/reports/key-questions-on-energy-and-ai/executive-summary
[8c] TechArena, “8 Ways AI Will Rewrite Data Center Infrastructure in 2026” (liquid cooling as default), December 2025. https://techarena.ai/content/8-ways-ai-will-rewrite-data-center-infrastructure-in-2026
[9] International Energy Agency, “Data centre electricity use surged in 2025” (Fatih Birol, “no AI without energy”), April 2026. https://www.iea.org/news/data-centre-electricity-use-surged-in-2025-even-with-tightening-bottlenecks-driving-a-scramble-for-solutions
[9b] Tech-Insider, “AI Data Centers: 1,000 TWh by 2026” (NERC; PJM shortfall; per-facility power), April 2026. https://tech-insider.org/ai-data-center-power-crisis-2026/
[10] NVIDIA, “Financial Results for First Quarter Fiscal 2027” (Jensen Huang), May 20, 2026. https://nvidianews.nvidia.com/news/nvidia-announces-financial-results-for-first-quarter-fiscal-2027
[11] NVIDIA Corp., Form 8-K, CFO Commentary, First Quarter Fiscal 2027 (Data Center revenue $75.2B), May 20, 2026. https://www.sec.gov/Archives/edgar/data/0001045810/000104581026000051/q1fy27cfocommentary.htm
[12] Tom’s Hardware (citing Financial Times), “Big Tech’s AI Spending Plans Reach $725 Billion” (Brent Thill, Jefferies), April 2026. https://www.tomshardware.com/tech-industry/big-tech/big-techs-ai-spending-plans-reach-725-billion
[13] Fortune, “Microsoft, Meta, Google AI Capex Spending” (Anat Ashkenazi, Alphabet CFO), April 29, 2026. https://fortune.com/2026/04/29/microsoft-meta-google-ai-capex-spending-billions/
[14] Quartz, “AI Data Centers Pass 1 Gigawatt and Strain the U.S. Power Grid” (Epoch AI; Zuckerberg; McKinsey $5.2T), May 14, 2026. https://qz.com/ai-data-centers-gigawatt-power-grid-strain-051126
[15] OpenAI, “OpenAI, Oracle, and SoftBank Expand Stargate with Five New AI Data Center Sites” (Sam Altman), September 2025. https://openai.com/index/five-new-stargate-sites/
[16] OpenAI, “Announcing The Stargate Project,” January 2025. https://openai.com/index/announcing-the-stargate-project/
[17] Stanford HAI, 2025 AI Index Report — Economy chapter. https://hai.stanford.edu/ai-index/2025-ai-index-report/economy
[18] Stanford HAI, 2026 AI Index Report. https://hai.stanford.edu/ai-index/2026-ai-index-report
[19] Daron Acemoglu, “The Simple Macroeconomics of AI,” NBER Working Paper No. 32487 / Economic Policy 40(121), 2025. https://www.nber.org/papers/w32487
[20] MIT Sloan, “A New Look at the Economics of AI” (Daron Acemoglu), January 2026. https://mitsloan.mit.edu/ideas-made-to-matter/a-new-look-economics-ai
[21] Fortune, “AI Productivity, Economic Spillover, and Low-Wage Workers” (Kristalina Georgieva, IMF, Davos), January 24, 2026. https://fortune.com/2026/01/24/ai-productivity-economic-spillover-low-wage-workers-imf-chief-kristalina-georgieva/
[22] International Monetary Fund, “AI Adoption and Inequality,” IMF Working Paper 2025/068, April 2025. https://www.imf.org/en/publications/wp/issues/2025/04/04/ai-adoption-and-inequality-565729
[23] The White House, “Winning the Race: America’s AI Action Plan,” July 2025. https://www.whitehouse.gov/wp-content/uploads/2025/07/Americas-AI-Action-Plan.pdf
[24] The White House, “White House Unveils America’s AI Action Plan” (David Sacks; Michael Kratsios), July 23, 2025. https://www.whitehouse.gov/releases/2025/07/white-house-unveils-americas-ai-action-plan/
[25] InnovationMap, “Gov. Abbott Says Texas AI Projects Could Be Bigger Than Project Stargate,” September 2025. https://houston.innovationmap.com/abbott-ai-boom-stargate-2674045926.html
[26] E&E News by POLITICO, “Texas Governor Talks Tough on Data Centers, Calls for Clampdown” (Abbott; Pichai $40B; Gallup poll), June 2026. https://www.eenews.net/articles/texas-governor-talks-tough-on-data-centers-calls-for-clampdown/
[27] DatacenterDynamics, “Texas Governor Approves Bill Mandating Data Centers to Contribute to Interconnection Costs” (SB 6), June 2025. https://www.datacenterdynamics.com/en/news/texas-governor-approves-bill-mandating-data-centers-to-contribute-to-costs-of-interconnection/
[28] The Texas Tribune, “A Data Center Boom Is Coming to Texas” (JLL; ERCOT queue; Cully Cavness, Crusoe), June 8, 2026. https://www.texastribune.org/2026/06/08/texas-regulation-data-centers-electricity-power-water/
[28b] UncoverAlpha, “Amazon, Google, Microsoft, Meta Q1 Earnings: AI Profits Are Here” (HBM sold out; Meta capex), April 2026. https://www.uncoveralpha.com/p/amazon-google-microsoft-meta-q1-earnings
[29] Westside Construction Group, “The Data Center Power Crisis” (Texas overtakes Northern Virginia; Meta Prometheus, Ohio), June 2026. https://www.buildwcg.com/blog-posts/data-center-power-grid-construction-ai-infrastructure-2026
[30] Office of the Governor of California, “Governor Newsom Signs SB 53” (AI job-posting share), September 29, 2025. https://www.gov.ca.gov/2025/09/29/governor-newsom-signs-sb-53-advancing-californias-world-leading-artificial-intelligence-industry/
[31] Brookings Institution, “What Is California’s AI Safety Law?” (SB 53 analysis), April 2026. https://www.brookings.edu/articles/what-is-californias-ai-safety-law/
[32] Office of the Governor of California, “Governor Newsom Signs First-of-Its-Kind Executive Order to Prepare Workers for AI Disruption,” May 21, 2026. https://www.gov.ca.gov/2026/05/21/governor-newsom-signs-first-of-its-kind-executive-order-to-prepare-workers-and-businesses-for-potential-ai-disruption/
[33] Pennsylvania DCED, “Governor Josh Shapiro Announces Amazon Plans to Invest $20 Billion” (“all in on AI”), June 12, 2025. https://dced.pa.gov/newsroom/governor-josh-shapiro-announces-amazon-plans-to-invest-20-billion-in-pennsylvania-for-ai-infrastructure-in-largest-capital-investment-in-commonwealth-history/
[34] Inside Climate News, “Pennsylvania’s Governor Has a Plan to Make Data Centers Bring Their Own Energy” (GRID standards), May 2026. https://insideclimatenews.org/news/28052026/pennsylvania-gov-shapiro-ai-data-center-plan/
[35] Pennsylvania Capital-Star, “Pa.’s Approach to Joining the AI Race Must Put People First” (Homer City; rate increases), February 2026. https://penncapital-star.com/campaigns-elections/pa-s-approach-to-joining-the-ai-race-must-put-people-first-shapiro-said-in-budget-address/
[36] Michigan Advance, “Whitmer: Multi-Billion-Dollar Saline Township Data Center ‘Largest Investment in Michigan History’,” October 30, 2025. https://michiganadvance.com/2025/10/30/whitmer-multi-billion-dollar-saline-township-data-center-largest-investment-in-michigan-history/
[37] OpenAI, “Building the Infrastructure for the Intelligence Age in Michigan” (The Barn, 1 GW), 2026. https://openai.com/index/stargate-michigan-data-center/
[38] Planet Detroit, “Data Centers in Michigan: What You Need to Know” (DTE/Consumers 6.4 GW; Loudoun property taxes; Michigan City rejection), November 2025. https://planetdetroit.org/2025/11/data-centers-water-energy-impact/
[39] Lightyear, “Data Center Alley: Ashburn, VA’s Data Center Hub (2026 Guide).” https://lightyear.ai/blogs/ashburn-colocation-data-center-alley
[40] Governing, “The Data Center Capital of the World Is in Virginia” (Youngkin AWS deal). https://www.governing.com/infrastructure/the-data-center-capital-of-the-world-is-in-virginia
[41] Technical.ly, “$67B Dominion-NextEra Deal Could Reshape Northern Virginia’s Data Center Economy” (603 data centers), May 2026. https://technical.ly/civics/dominion-nextera-deal-data-center-power/
[42] TSMC, “TSMC Intends to Expand Its Investment in the United States to US$165 Billion to Power the Future of AI,” March 4, 2025. https://pr.tsmc.com/english/news/3210
[44] Viqus, “The Global AI Infrastructure Race in 2026: Who Is Winning the Battle for Compute?” (China; Gulf), February 2026. https://viqus.ai/blog/global-ai-infrastructure-race-2026
[45] The AI Forest, “The Rise of Sovereign AI Data Centers” (EU Gigafactories; sovereign-AI spending; EU vocational academies), May 2026. https://theaiforest.com/sovereign-ai-data-centers-whats-driving-the-2026-boom/
[46] European Commission, “AI Continent Action Plan” (AI Factories; AI Gigafactories; InvestAI). https://commission.europa.eu/topics/competitiveness/ai-continent_en
[47] Introl, “The Middle East’s Trillion-Dollar Bet on AI Infrastructure” (UAE 5 GW; Microsoft/G42; HUMAIN), April 2026. https://introl.com/blog/middle-east-uae-saudi-arabia-ai-data-center-boom-2025
[48] Center for a New American Security, “Sovereign AI Index,” 2026 (NVIDIA share; UAE/Japan concentration). https://interactives.cnas.org/reports/sovereign-ai-index/
[49] Introl, “South Korea’s $735B Sovereign AI Initiative,” April 2026. https://introl.com/blog/south-korea-735b-sovereign-ai-initiative-infrastructure-requirements-opportunities
[50] Futurum, “AI Capex 2026: The $690B Infrastructure Sprint” (Japan ¥1T; South Korea budget; EU €200B), February 2026. https://futurumgroup.com/insights/ai-capex-2026-the-690b-infrastructure-sprint/
[51] McKinsey & Company, “Beyond the Spillover: Asia-Pacific, the Next Engine of Data Center Demand” ($6.7T; Alibaba; ByteDance), June 2026. https://www.mckinsey.com/featured-insights/future-of-asia/countries-and-regions/southeast-asia/southeast-asia-perspectives/beyond-the-spillover-asia-pacific-the-next-engine-of-data-center-demand
[52] Global Data Center Hub, “Q1 2026: The Quarter AI Infrastructure Became Energy-Constrained” (Adani; sovereign buildouts). https://www.globaldatacenterhub.com/p/q1-2026-the-quarter-ai-infrastructure
[53] RAISE Summit, “Sovereign AI: Why Nations Are Treating Compute as Critical Infrastructure” (Macron; France €109B). https://www.raisesummit.com/post/sovereign-ai-compute-critical-infrastructure
[54] Data Center Knowledge, “World Energy Outlook 2025: Data Center Energy Drain” (Fatih Birol, $580B; “data is the new oil”), November 2025. https://www.datacenterknowledge.com/energy-power-supply/world-energy-outlook-2025-skyrocketing-data-center-demand-outpaces-grid
[56] Clifford Chance, “Data Centres & AI Compute Infrastructure Insights 2026” (industrial-scale plants transforming data and energy into intelligence). https://www.cliffordchance.com/insights/thought_leadership/trends/2026/data-centres-and-ai-compute-infrastructure-insights-2026.html
[57] Fortune, “A Michigan Farm Town Voted Down Plans for a Giant OpenAI-Oracle Data Center” (Whitmer outreach; July 2025 EO; $5.2T), May 6, 2026. https://fortune.com/2026/05/06/ai-data-center-michigan-saline-politics-farmland/
