Introduction: When the Electricity Customer Became Something More
On the final day of August 2026, an unusual financial detail emerged from the rapidly expanding artificial-intelligence infrastructure economy, and although it arrived in the understated language of a wire-service dispatch, it deserves to be read as a signal of a much deeper structural transformation. Reuters reported, citing the Wall Street Journal and draft initial-public-offering documents reviewed by the newspaper, that OpenAI had been issued warrants in SB Energy—the SoftBank-controlled power and datacenter developer—valued at approximately $5.5 billion. Reuters noted that it could not immediately verify the report, and that neither SB Energy nor OpenAI responded to requests for comment outside regular business hours, so the precise contractual terms of the instruments remained undisclosed at the time of writing. [1]
Yet the significance of the disclosure extends far beyond the valuation itself. Subsequent reporting on the draft IPO documents added texture that transforms the story from a curiosity into a case study. According to the Journal’s account, the warrants were issued in January 2026 at an estimated value of roughly $3.6 billion, and by the end of June 2026 their estimated value had risen to approximately $5.5 billion, an appreciation of more than fifty percent in under six months, driven by the rising implied valuation of the infrastructure company itself. The warrants were reportedly granted as an inducement to secure OpenAI as a data-center tenant, they are expected to vest in stages after SB Energy’s initial public offering as the company reaches specified valuation milestones, and OpenAI—which also invested $500 million of cash into SB Energy earlier in the year—is expected to hold a single-digit ownership stake following the listing. SB Energy, majority-owned by SoftBank Group, is preparing an IPO that bankers expect could raise between $5 billion and $7 billion as soon as the following month, at a valuation that earlier reporting suggested could exceed $50 billion. [2]
If the reported arrangement is understood correctly, one of the world’s most important frontier artificial-intelligence laboratories was no longer behaving merely as a corporation purchasing electricity, leasing computing capacity, or signing a long-term agreement with an infrastructure provider. It had acquired a potential financial claim on the future value of the company helping to build the physical infrastructure required to run its models. The customer had become, simultaneously, an investor, a strategic partner, a software vendor, and a holder of equity-linked upside in its own landlord.
The relationship had already been deepening for months before the warrant disclosure. In January 2026, OpenAI and SoftBank Group each agreed to invest $500 million in SB Energy as part of the Stargate infrastructure program, and OpenAI simultaneously selected SB Energy to build and operate its previously announced 1.2-gigawatt data-center site in Milam County, Texas. [5] Greg Brockman, OpenAI’s co-founder and president, framed the logic of the partnership in explicitly industrial terms, describing the combination of SB Energy’s infrastructure and energy-development capabilities with OpenAI’s data-center engineering expertise as producing, in his words,
“a fast, reliable way to scale compute through large, highly optimized AI data centers.” — Greg Brockman, President of OpenAI [6]
SB Energy, in turn, agreed to become an OpenAI customer, adopting the company’s APIs and deploying ChatGPT internally for its employees. [7] The arrangement therefore contained capital moving in one direction, software and services moving in the other direction, and physical infrastructure being built between them to support a much larger future compute economy. It was already an unusually entangled commercial relationship. Then the scale changed dramatically.
On August 17, 2026, Nvidia disclosed that it had secured land, power and shell capacity through a partnership with SB Energy at the PORTS-Pike Technology Campus in Pike County, Ohio—a site adjacent to, and partly upon, the remediated federal land of the former Portsmouth Gaseous Diffusion Plant, one of the great uranium-enrichment complexes of the Cold War. OpenAI would be the customer for approximately eight gigawatts of IT capacity, while SB Energy would build, own and operate the datacenter infrastructure under twenty-year leases. Nvidia agreed to provide credit support on the land, power and shell buildout to secure an initial 4.25 gigawatts of IT capacity, with an option to support approximately 3.75 to 3.8 additional gigawatts as the site scales, and announced a direct $1.5 billion equity investment in SB Energy. [8] Nvidia’s subsequent SEC disclosure placed the aggregate cap on its initial guarantee obligations at $105 billion, becoming effective in phases as data centers reach ready-for-service milestones beginning in its fiscal 2029, with exposure declining as OpenAI fulfills its lease payments. The same filing offered an extraordinary glimpse of the commercial stakes: Nvidia estimated that each generation of its infrastructure deployed at PORTS-Pike could represent approximately 1.5 million GPUs, or roughly $150 billion to $200 billion in Nvidia revenue, with the site expected to support multiple upgrade cycles over its twenty-year life. [9]
The architecture that emerges from these disclosures is remarkable, and it is worth stating plainly. OpenAI provides the future demand. SB Energy develops the land, the electricity, the transmission relationships and the datacenter shells. Nvidia provides the computing platform and, through its guarantees and investment, helps make portions of the infrastructure financing possible at terms the developer could not obtain alone. OpenAI receives the computing capacity necessary to train and operate increasingly capable models. Nvidia secures a flagship location designed exclusively around its own AI infrastructure. SB Energy receives long-duration contracted customers, stronger financing credibility, and a credible path to the public markets. And, according to the August 31 report, OpenAI may also hold warrants whose value rises if the infrastructure company itself becomes more valuable—which is to say, warrants whose value rises substantially because of OpenAI’s own demand. [2]
What appears on the surface to be an electricity-and-real-estate transaction therefore begins to resemble something much larger: a self-reinforcing financial ecosystem connecting energy, chips, datacenters, models and applications, in which the boundaries between customer, supplier, investor, guarantor and owner are deliberately blurred. That is precisely where the analytical framework of this paper—the Five-Layer AI Economy—becomes useful.
In the traditional technology economy, a software company purchased electricity almost incidentally. The electric utility was several layers removed from the company’s competitive advantage, and a successful search engine, social network or enterprise-software vendor might consume enormous amounts of power while treating electricity as nothing more than an operating expense to be minimized. Frontier artificial intelligence is reversing that relationship. When the next generation of models requires clusters measured in hundreds of thousands or millions of accelerators; when datacenter campuses are measured in gigawatts rather than megawatts; when grid interconnection can require years of study and construction; when generation plants, substations, transmission corridors, cooling systems and land must be secured years before the compute is available; and when financing institutions require credible long-term customers before releasing tens of billions of dollars, electricity stops being merely an input. Electricity becomes a strategic right. The International Energy Agency, in its landmark Energy and AI special report, projected that global electricity consumption by data centres will more than double to around 945 terawatt-hours by 2030—slightly more than the entire present-day electricity consumption of Japan—with AI as the most important driver of that growth, and with data centres accounting for nearly half of all electricity demand growth in the United States between now and 2030. [14]
And once access to electricity becomes a strategic right, the companies creating the demand may increasingly seek something beyond electricity itself. They may seek participation in the economics of the infrastructure created because of that demand. That possibility gives rise to the concept developed throughout this paper.
Power Warrants describes the emerging financial architecture in which frontier AI companies obtain financial participation, ownership optionality, preferential economic rights, or other forms of upside in the energy and infrastructure businesses whose assets are being built to satisfy the AI companies’ own extraordinary demand for compute.
The term does not imply that every future arrangement will literally involve a stock warrant. A financial warrant, in its conventional definition, is a contract that gives the holder the right to purchase from the issuer a certain number of additional shares in the future at a specified price, and it is precisely this optionality—value that appears only if the issuer succeeds—that makes it such a revealing instrument in the present context. [13] In this paper, the OpenAI–SB Energy disclosure supplies the literal starting point, while Power Warrants becomes a broader analytical concept encompassing warrants, equity interests, preferred securities, project-development rights, credit guarantees, capacity options and similar arrangements that economically connect the future growth of artificial intelligence with the future valuation of the infrastructure underneath it.
The fundamental question of the AI infrastructure era is therefore no longer simply: Who will supply enough electricity for artificial intelligence? The deeper question, and the one this paper attempts to answer, is: Who will own the financial claims on the power infrastructure that artificial intelligence causes the world to build?
Why I Chose the Title “Power Warrants”
I chose Power Warrants because the phrase captures a structural transition that ordinary terms such as power procurement, datacenter financing, energy partnership or AI infrastructure investment do not fully explain. The August 31 OpenAI–SB Energy report provides an unusually literal example of the phenomenon: a frontier AI laboratory reportedly received warrants—an instrument whose entire value derives from the future appreciation of its issuer—in the company responsible for supplying part of the infrastructure upon which its own future compute expansion depends. Instead of describing electricity solely as a purchased commodity, Power Warrants asks whether AI demand itself can become a source of bargaining power capable of generating equity participation, financial optionality and long-term infrastructure upside for the companies creating that demand. The tenant, in this framing, is no longer merely paying rent; the tenant is being paid, in equity-linked instruments, for the economic gravity its demand exerts on the landlord’s valuation.
I also chose the title because it connects naturally with the Five-Layer AI Economy framework that organizes my broader research program. Layer One is energy; Layer Two is semiconductors and computing hardware; Layer Three is datacenters and physical infrastructure; Layer Four is frontier models and the laboratories that build them; and Layer Five is applications and the revenue that ultimately sustains everything beneath it. The Power Warrants phenomenon demonstrates that Layer Four—the model laboratory—can now influence and partly own the economics of Layer One energy and Layer Three datacenters, while Layer Two chip companies such as Nvidia can provide the capital, guarantees and technical architecture linking all of them together. The term therefore captures more than a financial instrument. It describes a possible redistribution of economic power across the entire AI stack.
There is a final reason for the title, and it is deliberately double-edged. The word power refers simultaneously to electricity and to economic leverage, and the word warrant refers simultaneously to a security and to a justification. The frontier laboratory of the late 2020s may no longer simply rent GPUs from a hyperscaler and pay an electricity bill indirectly through its cloud invoice. It may increasingly participate in selecting the geography, financing the developer, underwriting the infrastructure, shaping the generation portfolio—and receiving financial upside if that infrastructure becomes more valuable. Whether the world should welcome that development, regulate it, or merely understand it is a question the title is intended to keep permanently open. What warrants this power, and what powers these warrants, are the twin questions running beneath every section that follows.
Section 1: From Power Purchasing to Power Participation
1.1 The Old Technology Model: Electricity as an Operating Expense
To understand what is changing, it is necessary to first establish the historical baseline with some care, because the transformation described in this paper is visible only against that background. For roughly the first seventy years of commercial computing, technology companies treated electricity largely as a cost of running datacenters—an important cost, occasionally a site-selection criterion, but fundamentally an operating expense rather than a strategic asset. Even the hyperscale cloud companies of the 2010s, which developed genuinely sophisticated renewable-energy procurement programs and pioneered the corporate power-purchase agreement as a mainstream instrument, were focused on a relatively narrow set of objectives: securing adequate supply, stabilizing long-run costs, meeting public sustainability commitments, and ensuring the reliability of their services. The relationship could be simplified into a linear chain in which the utility generated and delivered electricity, the datacenter consumed it, the computing infrastructure transformed it into digital services, and the software company sold those services to the world. Electricity was essential, but the software company rarely needed to participate directly in the capitalization of the energy ecosystem supporting any particular application, and it almost never received equity in the companies that built its power supply.
The economics of that era explain the arrangement. A large search or social-media datacenter of the early 2010s might draw twenty to fifty megawatts—a meaningful industrial load, but one that existing utility systems, with modest reinforcement, could typically absorb. Power represented perhaps a fifth of a datacenter’s operating cost and a far smaller fraction of the parent company’s income statement. The grid was, from the software company’s perspective, ambient infrastructure: something one connected to, complained about occasionally, and otherwise ignored. Nothing about that posture was irrational. It was the correct response to a world in which computing demand grew smoothly and electricity was abundant relative to the requests being made of it.
1.2 The AI Transition: Electricity as a Capacity Constraint
Frontier artificial intelligence has broken every assumption in that model, because the scale, density, concentration and timing of AI power demand are radically different from anything the digital economy previously produced. A frontier training cluster is not a twenty-megawatt facility; the PORTS-Pike campus alone is designed around approximately eight gigawatts of IT capacity supported by at least ten gigawatts of new generation—several hundred times the load of a classic cloud datacenter, concentrated at a single rural site. [8] The IEA calculates that data-centre electricity consumption is growing at roughly fifteen percent per year through 2030, more than four times faster than total electricity consumption from all other sectors combined, and that by the end of the decade the United States will consume more electricity for data centres than for the production of aluminium, steel, cement, chemicals and all other energy-intensive goods combined. [14] Fatih Birol, the IEA’s Executive Director, captured the institutional astonishment succinctly:
“AI is one of the biggest stories in the energy world today.” — Fatih Birol, Executive Director, International Energy Agency [15]
The Brookings Institution, surveying the regulatory landscape in mid-2026, noted that U.S. data-center electricity demand is projected to increase by roughly 130 percent by 2030 relative to 2024, and that data centers already consumed approximately 4.4 percent of total U.S. electricity as of the most recent federal assessment, with credible projections ranging from 6.7 to 12 percent of national consumption by 2028. [16]
Under these conditions, a frontier AI laboratory can no longer assume that sufficient powered datacenter capacity will simply appear on the market when it is needed. Model development increasingly requires simultaneous access to land, grid interconnection, generation, substations, high-voltage transmission, backup systems, cooling, datacenter shells, networking, accelerators, financing and long-duration contractual commitments—and every one of those elements sits on a different timeline, controlled by a different set of institutions, with different regulatory gatekeepers. A GPU can be manufactured in months; a 765-kilovolt transmission line takes years; a natural-gas combined-cycle plant takes years; grid-interconnection queues in some U.S. regions stretch beyond half a decade. The scarce product is therefore no longer merely a GPU. It is a powered GPU at the correct location, connected to the correct network, at the required scale, on the required date. Whoever controls the assembly of those conditions controls the rate at which frontier AI can advance, and that realization—more than any single transaction—is the origin of the Power Warrants phenomenon.
1.3 The Warrant as the Symbol of a New Relationship
It is here that the financial mechanics deserve careful introduction, because the warrant is not an arbitrary instrument; it is arguably the perfect instrument for this moment, and understanding why illuminates the entire structure. A traditional warrant gives its holder the right, but not the obligation, to purchase securities from the issuer under predetermined conditions—typically a fixed price and a fixed period. [13] Its value at issuance may be modest; its value at maturity depends entirely on whether the issuing company succeeds. A warrant is therefore a mechanism for transferring future upside without transferring present cash, and for binding two parties’ fortunes together without an immediate change of control.
The OpenAI–SB Energy report makes the warrant especially important conceptually because the recipient of the warrants is also expected to be the infrastructure company’s most important customer. The draft IPO documents reportedly show that SB Energy—whose existing renewable-energy operations generated only about $140 million of revenue in the first half of 2026, and which has yet to bring a single data center online—nevertheless discloses nearly nine gigawatts of contracted computing capacity and a revenue backlog exceeding $400 billion, the overwhelming majority of it attributable to OpenAI’s seventeen leases at the Ohio campus. [4] The same draft filing reportedly flags SB Energy’s heavy reliance on OpenAI as a key risk factor, warns that any weakening of OpenAI’s finances could have adverse consequences for SB Energy, and discloses that the developer’s capacity to fund the Ohio campus is meaningfully contingent on Nvidia’s residual-value guarantee; it also shows SB Energy’s net loss widening to $3.2 billion in the first half of 2026, from roughly $250 million a year earlier, as the buildout accelerated. [3]
This creates a potentially reflexive arrangement whose logic can be traced step by step. OpenAI’s demand produces the infrastructure contract. The infrastructure contract produces the backlog. The backlog produces the developer’s valuation. The valuation produces the appreciation of the warrants. And the appreciation of the warrants flows back to OpenAI, the customer whose demand initiated the sequence. The customer’s own growth improves the economics of the developer; the developer’s growth creates the capacity necessary for the customer’s expansion; and the financial instrument sitting between them converts that mutual dependence into a measurable asset on the customer’s balance sheet. The distinction between customer, investor and strategic partner does not merely blur—it is engineered away, deliberately, because the blurring is the point. Each party is buying insurance against the other’s hesitation.
Nor is the SB Energy warrant an isolated invention. In October 2025, AMD issued OpenAI a warrant for up to 160 million shares of AMD common stock—approximately ten percent of the company—at an exercise price of one cent per share, with tranches vesting as OpenAI scales its deployment of AMD GPUs from one gigawatt toward six gigawatts and as AMD’s own share price reaches specified targets, exercisable through October 2030. [20] AMD’s own securities filings describe the structure in the dry language of a product purchase agreement paired with a milestone-vested warrant, [21] but the commercial meaning was unmistakable, and markets understood it instantly: AMD’s shares surged more than twenty percent on the announcement as investors priced in tens of billions of dollars of expected revenue, while analysts observed that OpenAI could become one of AMD’s largest shareholders without ever writing a meaningful check for the stock. [22] The chip supplier paid for volume with equity; the customer converted its purchasing power into ownership. The SB Energy arrangement extends the identical logic downward from Layer Two into Layers Three and One—from silicon into buildings, transmission and generation—and that extension is exactly the transition this paper names.
1.4 From Anchor Tenant to Infrastructure-Forming Customer
Commercial real estate has relied for a century on the concept of the anchor tenant: the department store whose presence justified the shopping mall, the investment-grade corporate lessee whose signature made an office tower financeable. Project-financed infrastructure similarly depends on long-term contracted customers—the airline underpinning the terminal, the utility offtaker underpinning the merchant power plant. The Power Warrants economy extends this familiar model, but it extends it so far that the extension becomes a difference in kind rather than degree, because a frontier laboratory can potentially provide, simultaneously, at least seven distinct forms of value to an infrastructure developer: contractual demand measured in decades and hundreds of billions of dollars; direct equity investment; technological significance that attracts other partners; financing credibility that lowers the developer’s cost of capital across its entire portfolio; political visibility that accelerates permits and public support; an ecosystem of allied suppliers, led by the chip vendor, who bring their own balance sheets; and future expansion commitments that convert a single project into a multi-decade pipeline.
A tenant that provides all seven is no longer usefully described as a tenant. It is what this paper calls an infrastructure-forming customer: an entity whose demand does not merely fill capacity but causes capacity—and the companies that build it—to exist. And an infrastructure-forming customer, aware of its own gravitational role, will predictably begin to ask why it should generate billions of dollars of enterprise value for its counterparties while capturing none of it. The warrant is the answer to that question. It is the price the developer pays for the privilege of being formed.
1.5 Defining the Power Warrants Framework: Five Stages
The progression from the old model to the new one can be formalized as a five-stage framework, which will organize much of the analysis in the remainder of this paper.
| Stage | Name | Description | Illustrative Example |
| One | Power Procurement | The AI company purchases or indirectly consumes electricity as an ordinary operating expense. | Classic cloud-era datacenter contracts; electricity embedded in cloud invoices. |
| Two | Capacity Reservation | The company signs long-term contracts for powered datacenter capacity, becoming an anchor tenant. | Twenty-year hyperscaler PPAs; Microsoft’s 835 MW Three Mile Island restart agreement. [42] |
| Three | Infrastructure Participation | The company invests directly in the developer or project, holding equity alongside its lease. | OpenAI’s and SoftBank’s $500 million investments in SB Energy, January 2026. [5] |
| Four | Financial Optionality | Warrants, options, preferred rights or milestone instruments create additional, contingent upside. | The reported $5.5 billion SB Energy warrant package; AMD’s 160-million-share warrant. [2][20] |
| Five | Infrastructure Integration | Power development, compute procurement, credit support, financial ownership and model expansion become strategically coordinated across parties. | The OpenAI–SB Energy–Nvidia PORTS-Pike architecture, with $105 billion of conditional guarantees. [9] |
Power Warrants, as a phenomenon, is the movement of the frontier AI economy from Stage One toward Stages Four and Five. Stages One and Two are procurement; Stages Three through Five are participation. The historical novelty of 2026 is not that AI companies buy a great deal of electricity—it is that the largest of them are climbing this ladder faster than regulators, investors and the public have developed the vocabulary to describe, and the purpose of this framework is to supply that vocabulary before the structure hardens into permanence.
1.6 What the Framework Excludes, and Why That Matters
Intellectual honesty requires acknowledging what the Power Warrants framework does not claim. It does not claim that every AI company will hold warrants in its power suppliers; most will not, because most lack the demand gravity to command them. It does not claim that the arrangements are secretly nefarious; every element of the SB Energy structure described above was disclosed in press releases, securities filings or reported IPO documents, and disclosure is itself a form of legitimacy. And it does not claim that the structure is unprecedented in industrial history—the railroads, the early electricity trusts, the postwar aircraft consortia and the semiconductor equipment ecosystem all featured deep financial entanglement between customers and suppliers. What the framework does claim is narrower and, I would argue, more defensible: that a specific, identifiable, and rapidly generalizing financial architecture is forming around frontier AI infrastructure; that this architecture systematically converts compute demand into ownership claims on the physical economy; and that its consequences—for competition, for financial stability, for ratepayers, and for the geography of American industry—deserve analysis before, rather than after, they become irreversible. The remainder of this paper undertakes that analysis layer by layer.
Section 2: The New Capital Architecture of Artificial Intelligence
2.1 Demand Becomes Collateral
Gigawatt-scale AI infrastructure requires enormous upfront investment, and the sequencing of that investment is the key to understanding why the capital architecture of AI has become so unusual. Before lenders will finance transmission lines, generation plants, buildings and equipment—assets that may take four to six years to complete and twenty to forty years to amortize—they need confidence that customers will eventually occupy the capacity and pay for it across decades. In conventional project finance, that confidence comes from an offtake agreement with an investment-grade counterparty. In frontier AI, the counterparties are young, unprofitable, privately held laboratories whose revenue, however spectacular its growth, remains small relative to their commitments: OpenAI’s annualized revenue passed roughly $25 billion by early 2026 and, by CNBC’s August 2026 reporting, exceeded a $40 billion annualized pace, [44] yet the company has simultaneously accumulated infrastructure commitments across its vendor ecosystem that independent tallies place above one trillion dollars over the coming decade. [41]
The resolution of this paradox is that AI demand itself has begun to function as collateral—not in the strict legal sense, but in the economic sense that a long-duration OpenAI lease is now the foundational document upon which billions of dollars of construction debt is raised. A twenty-year, multi-gigawatt lease from the operator of the world’s most widely used AI product is more than future revenue for the developer; it is the instrument that converts a field in Appalachian Ohio into a financeable asset. The draft SB Energy IPO documents make this dependency explicit: a company with about $140 million of first-half revenue presents investors with a contracted backlog above $400 billion, [4] and candidly warns that its fortunes rise and fall with the finances of a single tenant. [3] When demand becomes collateral, the demander acquires leverage over the entire capital stack—and Power Warrants is what that leverage looks like when it is exercised.
2.2 OpenAI, SoftBank and SB Energy: The First Case Study
The January 2026 transaction deserves development as the opening case study, because it contained, in embryonic form, every element that would later scale up in Ohio. On January 9, 2026, SB Energy announced a strategic partnership with OpenAI under the Stargate program—the $500 billion multi-year AI-infrastructure initiative announced at the White House in January 2025 with SoftBank, Oracle and other investors—in which OpenAI and SoftBank Group each invested $500 million into SB Energy, and OpenAI selected SB Energy to build and operate its previously announced 1.2-gigawatt data-center site in Milam County, Texas. [5] The announcement emphasized that the equity funding would support SB Energy’s growth as a development and execution partner for data-center campuses, and that the collaboration would combine OpenAI’s first-party data-center design with SB Energy’s expertise in speed, cost discipline and integrated energy. [6]
What made the transaction structurally novel was its deliberate circularity of roles. OpenAI became, at once, an investor in SB Energy, a tenant of SB Energy, a design collaborator with SB Energy, a strategic customer whose name anchored SB Energy’s growth story, and—according to the later warrant disclosure, which dates the instruments’ issuance to this same January window [2]—a holder of contingent equity upside in SB Energy. Meanwhile, SB Energy agreed to become an OpenAI customer, purchasing its APIs and deploying ChatGPT across its workforce. [7] Capital flowed from the laboratory to the developer; rent will flow from the laboratory to the developer; software revenue flows from the developer to the laboratory; and warrant value flows from the developer’s appreciation back to the laboratory. The transaction contains two-way revenue and two-way capital, braided together so tightly that neither party’s income statement can any longer be read in isolation from the other’s balance sheet. That braiding, not the dollar amounts, is the historical innovation.
2.3 Nvidia Enters the Capital Stack
The Ohio project then added the dimension that transformed a bilateral partnership into a systemic architecture: the entry of the chip supplier into the infrastructure capital stack. Nvidia’s August 17 announcement made clear that the company is not merely selling accelerators into a datacenter constructed by somebody else. It has secured land, power and shell capacity at PORTS-Pike; committed $1.5 billion of equity to SB Energy; obtained exclusivity, such that the campus will host only Nvidia AI compute infrastructure, subject to limited exceptions; and agreed to provide conditional credit support—residual-value guarantees capped at $105 billion—covering the initial 4.25 gigawatts, with an option covering the remainder. [8][9]
It is essential, for analytical precision, to state what the $105 billion figure is and is not, because much early commentary conflated it with an investment. The guarantees are a credit backstop tied to the residual value of the leased infrastructure: they phase in as facilities reach ready-for-service milestones from roughly 2028 onward, they decline as OpenAI makes its lease payments, and in the base case—in which the tenant simply pays its rent—Nvidia pays nothing under them. [11] The actual cash investment is the $1.5 billion equity stake, later supplemented, according to the IPO reporting, by roughly $3 billion of additional commitments through private transactions tied to the listing, one of which reportedly grants Nvidia the right to buy shares at a ten percent discount to the offering price. [3] The guarantee is not a gift; it is the purchase of a market. Nvidia’s own filing quantifies the prize: each infrastructure generation at the site could represent approximately 1.5 million Nvidia GPUs and $150 billion to $200 billion of Nvidia revenue, across multiple upgrade cycles over twenty years. [9]
This suggests a major strategic shift that deserves to be named plainly: the GPU supplier is helping to manufacture the market into which its future GPUs will be sold. Jensen Huang, announcing the project, described the intended output in industrial vocabulary—
“OpenAI will build and operate a world-class AI factory at PORTS-Pike.” — Jensen Huang, CEO of NVIDIA [26]
—and the company’s quarterly results explain why the wager seems rational to its author. In the quarter ended July 26, 2026, Nvidia reported revenue of $96.2 billion, up 106 percent from a year earlier, with data-center revenue of $89.0 billion and gross margins of 75 percent, and it guided the following quarter to approximately $108 billion. [18] Huang’s framing of the demand environment was categorical:
“Its tokens are productive and profitable. Now, compute is revenue.” — Jensen Huang, CEO of NVIDIA [18]
A company generating cash at that rate, with a market capitalization measured in trillions, can afford to deploy its balance sheet as a strategic weapon—guaranteeing the facilities that will house its chips, financing the customers that will buy them, and collecting equity in the developers that will build around them. That is a far broader corporate role than semiconductor design, and it is the Layer Two half of the Power Warrants architecture.
2.4 The Triangle of Capital
The three-party structure can now be assembled into the paper’s central diagram, which I call the Power-Warrant Triangle.
| Vertex | Party | What It Contributes | What It Receives |
| Power Developer | SB Energy | Land, generation, transmission funding, datacenter shells; builds, owns and operates the campus under 20-year leases. [8] | $2.5 billion-plus of equity from OpenAI, SoftBank and Nvidia; a $400 billion contracted backlog; IPO credibility. [3][4] |
| AI Laboratory | OpenAI | Long-duration compute demand: 17 leases, ~8 GW-IT, 20-year terms; $500 million equity; anchor-tenant credibility. [4][5][12] | Powered compute capacity at unprecedented scale; reported $5.5 billion of warrants; single-digit post-IPO stake. [2] |
| Compute Platform | Nvidia | Exclusive accelerator architecture; $1.5 billion equity plus IPO commitments; up to $105 billion of conditional residual-value guarantees. [8][9] | A captive multi-cycle market estimated at $150–200 billion of revenue per infrastructure generation; equity upside; IPO discount rights. [3][9] |
The three parties collectively reduce risks that none could easily absorb alone. SB Energy’s construction and development risk is reduced because its capacity is pre-sold for twenty years to a tenant whose lease is credit-supported by the most valuable company in the world. OpenAI’s utilization and availability risk is reduced because a dedicated developer, backed by SoftBank’s capital and Nvidia’s guarantees, is legally committed to delivering powered shells on a schedule. Nvidia’s technology and demand risk is reduced because the resulting campus is contractually exclusive to its platform for two decades. The synthesis can be compressed into a single expression: Demand + Infrastructure + Compute + Capital = Bankable AI Capacity. Each vertex supplies the element the financing market would otherwise find missing, and the warrants, guarantees and equity stakes are the adhesive holding the vertices together.
2.5 Circularity Versus Industrial Coordination
This paper would fail in its duty if it treated the concerns about circular financing as a public-relations problem to be dismissed rather than an analytical question to be examined, because those concerns are now voiced by the most serious institutions in global finance. When Nvidia first announced its intention to invest up to $100 billion in OpenAI in September 2025, Bernstein’s veteran semiconductor analyst Stacy Rasgon warned investors immediately that
“The action will clearly fuel ‘circular’ concerns.” — Stacy Rasgon, Senior Analyst, Bernstein Research [39]
By mid-2026, the pattern had a name and a literature. Bloomberg mapped the web of cross-investments among Microsoft, OpenAI, Nvidia, Oracle, AMD, CoreWeave and SoftBank as a defining feature of the cycle, while also noting the defense advanced by asset managers such as Janus Henderson: that in a market where compute is genuinely scarce, pairing long-term purchase commitments with financing is less a shell game than a “virtuous circle” that lines up suppliers, builders and customers against exploding demand. [41] Axios, reporting in July 2026 on the then-contemplated Ohio guarantee—which was initially discussed at approximately $250 billion before being negotiated down to the eventual $105 billion cap—observed that Nvidia was, in effect, using its financial heft to buoy customers who are losing money at historic rates, and that markets punished the stock on the news. [40]
The Bank for International Settlements elevated the issue from market commentary to systemic-risk doctrine. Its Annual Economic Report 2026, released in late June, named the sustainability of the AI investment boom as one of the principal pressure points threatening global financial stability, estimating that the five largest technology firms will spend more than one trillion dollars on AI-related capital expenditure across 2025 and 2026—beyond their earnings and free cash flow, and therefore increasingly funded with debt—and singling out circular financing arrangements in which chipmakers and hyperscalers take equity stakes in AI labs or computing firms that then commit to multi-year purchases from those same investors. [28][29] The BIS warned that the terms of such arrangements are often poorly disclosed, creating the risk of the same underlying asset being pledged multiple times, [46] and its central warning deserves quotation on its own line:
“…turn the capex boom into a protracted investment bust.” — Bank for International Settlements, Annual Economic Report 2026 [30]
A companion BIS working paper modeled the buildout as an investment race in which competitive pressure drives aggregate capital expenditure so high that the sector’s net economic surplus can turn negative in adverse scenarios. [31]
But the opposite interpretation also deserves rigorous statement, because industrial history is largely a history of coordinated finance. The transcontinental railroads were financed by land grants and construction companies owned by the railroads’ own promoters; the early electricity industry grew through holding companies in which equipment manufacturers held equity in the utilities that bought their turbines; the jet age was launched by airframe makers, engine makers and airlines that co-financed each other’s risk; and the semiconductor industry’s equipment ecosystem has always involved suppliers investing in customers’ capacity. Vertical financial entanglement at the technological frontier is the historical norm, not the exception, and it exists because frontier infrastructure exhibits exactly the coordination problem the Power-Warrant Triangle solves: no party will commit first without assurance that the others will follow. The honest analytical question is therefore not whether coordination exists—it plainly does—but where the boundary lies between coordination that accelerates genuine capacity and circularity that manufactures the appearance of demand. This paper proposes a working test with three prongs. First, additionality: does the arrangement cause real physical assets—generation, transmission, buildings—to exist that would not otherwise exist? Second, external validation: does revenue ultimately arrive from parties outside the circle, at Layer Five, in amounts sufficient to service the capital? Third, transparency: are the instruments, guarantees and dependencies disclosed in sufficient detail for outside investors and regulators to price them? By the first prong, the Ohio architecture scores well—ten gigawatts of generation and $4.2 billion of transmission are unambiguously real. [23] The second prong remains an open empirical question examined in Section 3.5. The third prong is where, as the BIS insists, the ecosystem currently falls short, and where policy attention properly belongs.
2.6 The Macroeconomic Weight of the Architecture
One further fact belongs in this section, because it converts the Power Warrants story from a corporate-finance curiosity into a macroeconomic phenomenon. The capital being coordinated through these structures is now large enough to move national accounts. Harvard’s Jason Furman calculated that investment in information-processing equipment and software—only about four percent of U.S. GDP—accounted for fully 92 percent of U.S. GDP growth in the first half of 2025, and that growth excluding those categories ran at roughly 0.1 percent annualized; he cautioned, fairly, that
“absent the AI boom we would probably have lower interest rates [and] electricity prices” — Jason Furman, Professor, Harvard University [36]
and thus some offsetting growth elsewhere, but the direction of the finding stunned even sympathetic observers. [36] ING’s economists, applying stricter definitions that subtract imported hardware, still attribute roughly 36 to 44 percent of U.S. GDP growth in mid-2026 to the technology-investment complex. [37] The four largest hyperscalers—Amazon, Microsoft, Alphabet and Meta—guided to a combined roughly $725 billion of capital expenditure for 2026, up 77 percent from the record $410 billion of 2025, with Microsoft alone setting calendar-2026 capex near $190 billion; Jefferies analyst Brent Thill, defending the spending against skeptics, put the bull case in five words:
“The bear thesis is garbage.” — Brent Thill, Analyst, Jefferies [17]
When private structures of the kind described in this section steer capital flows of this magnitude—flows that rival the infrastructure budgets of nations—their governance ceases to be a private matter. That recognition motivates Section 4. First, however, the architecture must be located within the full five-layer framework.
Section 3: Power Warrants Through the Five-Layer AI Economy
3.1 Layer One — Energy Becomes Investable AI Infrastructure
Energy is no longer external to AI strategy; it has become the first layer of the AI economy itself, and the Ohio development illustrates the scale of the shift with unusual clarity. To support the PORTS-Pike campus, SB Energy and SoftBank have committed to building at least ten gigawatts of new energy generation—yielding approximately eight gigawatts of IT capacity—and to investing at least $4.2 billion in new regional grid infrastructure through a partnership with AEP Ohio explicitly designed to protect ratepayers. [8] The transmission program, announced in March 2026 alongside the U.S. Department of Energy, involves new 765-kilovolt lines—the highest-capacity transmission technology in the American grid, capable of moving roughly six times the energy of a standard 345-kilovolt line—with power expected to begin flowing to the site in 2029, and with SB Energy committed to paying the full $4.2 billion so that the costs are not shifted onto Ohio households. [23] AEP Ohio’s president, Marc Reitter, emphasized precisely this feature, describing a framework aimed at
“protecting our ratepayers from the costs associated with this new infrastructure.” — Marc Reitter, President, AEP Ohio [23]
The generation plan is equally striking: the new natural-gas generation serving the campus is funded in part through the U.S.–Japan Strategic Trade and Investment Agreement and will be owned by the United States government, while SB Energy funds the grid upgrades in full. [24] The energy layer of this project, in other words, is simultaneously private, bilateral-diplomatic and federal—a combination without obvious precedent in American power development.
Nor is the Ohio project unique in demonstrating that Layer One has become a field of corporate strategy. Across the hyperscale economy, Microsoft signed a twenty-year power purchase agreement for the entire 835-megawatt output of the restarted Three Mile Island Unit 1—the first resurrection of a retired American reactor for a single corporate client, now accelerated toward operation in 2027; Google contracted with Kairos Power for a fleet of small modular reactors totaling roughly 500 megawatts; Amazon invested in the Susquehanna nuclear campus and the X-energy SMR program; and Meta assembled what is described as the largest corporate nuclear procurement in U.S. history, up to 6.6 gigawatts across TerraPower, Oklo, Vistra and Constellation. [42][43] Under the Power Warrants framework, these transactions occupy Stage Two—capacity reservation—but their twenty-year durations and their financing significance for the generators involved show Layer One steadily migrating toward Stages Three and Four. Future AI laboratories and hyperscalers will increasingly treat generation ownership, fuel security, grid interconnection, transmission rights, capacity payments, tariff design, behind-the-meter generation, nuclear agreements and storage portfolios not as procurement details but as competitive assets to be owned, optioned and, where possible, capitalized.
3.2 Layer Two — Chips Acquire Power-Sector Exposure
Nvidia illustrates a remarkable Layer Two evolution that would have seemed implausible even three years ago: a semiconductor company has become one of the most important actors in American electricity development. The progression can be expressed as an ascent through abstraction layers in reverse—chip design led to system design, system design to datacenter architecture, datacenter architecture to infrastructure finance, and infrastructure finance, finally, to power availability itself. Nvidia’s own strategic language now treats “land, power and shell” as a named category of strategic resource for AI factories, to be secured, guaranteed and allocated like wafer capacity. [8]
The implication is profound for how Layer Two competition will be fought. Semiconductor rivalry in the Power Warrants era increasingly involves balance-sheet capacity and infrastructure orchestration, not merely transistor performance. AMD competed for OpenAI’s six gigawatts not only with its MI450 roadmap but with a ten-percent equity warrant; [20][22] Nvidia competed for the eight-gigawatt Ohio campus not only with its Vera Rubin platform but with $105 billion of credit support and exclusivity provisions. [9] Even the second-order consequences of the buildout now loop back through Layer Two’s own supply chain: Nvidia’s CFO, guiding gross margins lower into early fiscal 2027, told investors candidly that
“Memory scarcity today is being driven in large part by the AI buildout itself.” — Colette Kress, CFO of NVIDIA [19]
—a sentence in which the boom’s inputs, outputs and side effects chase one another in a single clause. A chip company that guarantees datacenters, invests in power developers, finances customers and absorbs memory inflation caused by its own success is no longer a component supplier. It is the systems integrator of the entire five-layer economy, and its equity stakes and guarantee book are Power Warrants instruments as surely as any stock certificate.
3.3 Layer Three — Datacenters Become Financial Bridges
The datacenter is the point at which the five layers physically intersect, and under the Power Warrants architecture it becomes a financial bridge as well as a technical one. Physically, the description is straightforward: electricity enters from Layer One; cooling removes the heat; Layer Two accelerators process the data; Layer Four models consume the compute; Layer Five applications monetize the intelligence; and capital, flowing in every direction at once, connects the whole structure. Financially, however, the datacenter has become the vessel into which every party’s claims are poured. The building’s twenty-year lease is the tenant’s obligation, the developer’s revenue, the lender’s security, and the guarantor’s contingent liability, simultaneously. The developer’s equity—held partly by the tenant and partly by the chip supplier—is a claim on the lease payments the tenant itself will make. The warrants held by the tenant are a claim on the market’s valuation of the lease the tenant signed. A single physical campus in Pike County, Ohio therefore supports at least five distinct, interlocking financial claims held by at least four parties plus the public markets, and the draft IPO documents through which those claims are now being disclosed are, in a real sense, the first published map of a Power Warrants balance sheet. [3][4] Layer Three has become the place where the AI economy’s engineering diagram and its capital-structure diagram are the same drawing.
3.4 Layer Four — Frontier Models Become Infrastructure Tenants, and Infrastructure Contracting Becomes a Competitive Moat
Model laboratories historically competed through algorithms, research talent, proprietary data and access to computing. The Power Warrants era introduces a further competitive dimension whose importance is easy to underestimate: infrastructure contracting capability—the organizational, financial and political capacity to commit credibly to multi-decade, multi-gigawatt obligations and to extract equity-linked consideration for doing so. A laboratory that can sign seventeen leases totaling eight gigawatts, [4] anchor a developer’s fifty-billion-dollar IPO, [2] attract a $105 billion guarantee from its chip supplier, [9] and negotiate warrant packages from both its silicon vendors [20] and its landlords [2] enjoys advantages that no benchmark score captures: earlier capacity, cheaper effective compute (because warrant appreciation offsets rent), preferential geography, and a supplier ecosystem financially incentivized to keep it winning.
This raises an important concern for the structure of the model race. If the decisive inputs to frontier progress are increasingly physical and financial rather than purely intellectual, the race may narrow toward organizations capable of financing industrial-scale ecosystems—laboratories attached to trillion-dollar platforms, sovereign wealth, or capital markets access of IPO scale. OpenAI’s own trajectory illustrates the threshold effects: the company filed a confidential S-1 in June 2026 reportedly targeting a valuation near one trillion dollars, [45] a sum unintelligible except as the capitalization of exactly the contracting capability described here. Smaller laboratories, however brilliant, cannot issue demand that functions as collateral, and therefore cannot harvest Power Warrants. The framework thus predicts—uncomfortably—that financial architecture will become a barrier to entry at Layer Four at least as formidable as model quality, a prediction whose competitive-policy implications are taken up in Section 4.
3.5 Layer Five — Applications Ultimately Pay the Bill
The entire structure remains economically sustainable only if intelligence produces revenue, and this is the layer where the Power Warrants architecture must ultimately face its audit. The demand signals are genuinely extraordinary: ChatGPT reached approximately one billion weekly active users in July 2026, with more than fifty million consumer subscribers reported earlier in the year, and OpenAI’s company-wide annualized revenue pace exceeded $40 billion by August 2026. [44] The application economy above the models—enterprise agents, API consumption, coding systems, scientific tools, advertising pilots and the early machine-to-machine economy—is growing at rates with few historical parallels.
Yet the skeptical arithmetic is equally sober, and it comes from serious quarters. Daron Acemoglu of MIT, the 2024 Nobel laureate in economics, estimates that AI will deliver on the order of half a percentage point of total factor productivity gains over the coming decade—a fraction of the projections embedded in current valuations—and has warned of the ecosystem’s interlocking deals in unsparing terms:
“These models are being hyped up, and we’re investing more than we should.” — Daron Acemoglu, Institute Professor, MIT; Nobel Laureate in Economics [34]
His deeper worry, expressed in the same interview, is that the circular arrangements among chipmakers, cloud providers and laboratories could eventually reveal themselves to be structurally fragile—a house of cards, in his phrase—if application revenue disappoints. [34][35] The IMF’s chief economist, Pierre-Olivier Gourinchas, crediting the AI investment boom with holding global growth at 3.3 percent in 2026 even as trade tensions dragged, [32] nevertheless framed the valuation risk with institutional precision:
“Markets may well be ahead of fundamentals.” — Pierre-Olivier Gourinchas, Chief Economist, International Monetary Fund [33]
The Five-Layer Power Warrants chain can therefore be written in both directions. Forward: Energy → Chips → Datacenters → Models → Applications → Revenue. And backward: Application Revenue → Model Expansion → Compute Demand → Infrastructure Valuation → Financial Upside. When both directions flow, the circle is virtuous and the warrants are real wealth; when Layer Five falters, the same circle transmits the shortfall instantly through every layer’s balance sheet at once, because the layers no longer merely trade with one another—they own one another. That double-edged circularity is one of the defining characteristics of the emerging AI industrial economy, and it is the reason the architecture cannot remain a purely private concern.
Section 4: Power Warrants Become a Public-Policy Question
4.1 The State Is the Silent Sixth Participant
Although the Five-Layer AI Economy contains five economic layers, every large infrastructure project in a modern democracy operates inside a sixth, governmental layer, and the Power Warrants architecture is unusually dependent on it. Governments control or decisively influence permits, taxation, electricity regulation, environmental review, land use, transmission approval, water rights, workforce incentives, economic-development subsidies and community-benefit requirements—and in the Ohio case, the government’s role goes far beyond regulation. The PORTS-Pike campus sits partly on remediated land controlled by the U.S. Department of Energy; the new natural-gas generation serving it is funded in part through the U.S.–Japan Strategic Trade and Investment Agreement and will be owned by the United States government; and the Department of Energy stood beside SB Energy and AEP Ohio at the March 2026 announcement of the transmission framework. [23][24] The federal government is, in a meaningful sense, a co-developer of the largest Power Warrants project in existence. A financial architecture in which private parties exchange warrants, guarantees and equity atop federally owned land, federally owned generation and state-approved transmission cannot coherently be described as a purely private matter, and this section develops the public questions it raises.
4.2 Ohio as a National Test Case
Pike County, Ohio may become one of the most consequential experiments in American AI industrial policy, and the symbolism of the site is almost too rich to require commentary. The former Portsmouth Gaseous Diffusion Plant enriched uranium for the national-security economy of the twentieth century; its successor campus will concentrate computation for the intelligence economy of the twenty-first. Yesterday’s atomic infrastructure geography is literally becoming tomorrow’s intelligence infrastructure geography, down to the reuse of the federal water system. [25]
The announced local stakes are substantial. OpenAI’s agreement to secure approximately eight gigawatts-IT at the campus is projected to create roughly 35,000 construction jobs over the six-year buildout through 2032 and approximately 2,500 long-term operating positions, with commitments to prioritize local workers, contractors and suppliers and a memorandum of understanding with North America’s Building Trades Unions. [12][25] The community-benefit architecture has grown in parallel: SB Energy’s original $40 million community fund was matched by an incremental $40 million from OpenAI, creating an initial $80 million community benefits fund directed at priorities identified by residents—schools, public safety, health care, utilities, workforce training, housing, veterans’ services and small business—alongside a separate commitment of up to $84 million in Codex credits extended to roughly 844,000 eligible Ohio college and technical-school students, and a pledge of annual public reporting on local hiring, community investment, water usage and energy consumption. [12][25][26] At the groundbreaking, SoftBank’s chairman Masayoshi Son addressed the region’s most sensitive anxiety—electricity bills—with a personal pledge:
“I will commit right here, right now, that we will protect the electricity bill.” — Masayoshi Son, Chairman, SoftBank Group [27]
Whether these commitments prove durable across a thirty-year asset life, changing corporate ownership, and the possible turbulence of the AI market cycle is precisely what makes Ohio a test case rather than a settled precedent. The structure is genuinely innovative; the question is whether innovation in community architecture can keep pace with innovation in financial architecture.
4.3 Ratepayers Versus AI Infrastructure: The Question of Who Pays
The central political question of the AI power buildout, in Ohio and everywhere else, will increasingly reduce to two words: who pays? The PORTS-Pike structure offers the most ratepayer-protective answer yet attempted at scale. SB Energy has committed to paying the full $4.2 billion for the new 765-kilovolt transmission infrastructure specifically to avoid increases in transmission rates for Ohio residents; the campus is designed to generate its own power rather than draw down the regional grid; and both AEP Ohio and the Department of Energy have presented the framework as a template under which data-center development pays for its dedicated infrastructure rather than shifting costs onto households and small businesses. [23][24] Observers of Ohio’s data-center politics have noted that this structure differs materially from earlier projects in the state, where transmission costs risked migrating to ordinary customers absent protective tariffs—here, the cost-shift problem is addressed by the deal’s structure itself rather than by after-the-fact rate design. [27]
The Power Warrants lens, however, sharpens the question rather than settling it. If AI laboratories can receive billions of dollars of financial upside from the infrastructure companies serving them—$5.5 billion of warrant value in the reported SB Energy case [2]—then elected officials will increasingly, and reasonably, ask why any portion of the enabling grid, generation or water systems should ever be socialized onto residential customers, and whether jurisdictions that host such projects should themselves negotiate for a share of the upside rather than settling for tax revenue and benefit funds. The Ohio template answers the first question well and the second question not at all. As electricity prices become a national political issue heading through the 2026 midterm environment and beyond, the disclosure standard set by the SB Energy IPO documents—which made the warrant arrangement publicly knowable at all—may prove to be the template’s most important export.
4.4 Community Equity and Infrastructure Equity: The Political Asymmetry
Power Warrants introduces an asymmetry that policy analysis should state explicitly rather than leave implicit. On one side of the ledger, the corporate participants in these architectures may receive warrants, equity stakes, long-term contracts, investment returns, preferred access, discount rights and strategic optionality—claims that are liquid, appreciating, portable and heritable. On the other side, host communities receive jobs, tax revenue, infrastructure, community funds and educational credits—benefits that are real and substantial, but that arrive bundled with environmental effects, housing pressure, construction disruption, water demands and long-term land-use transformation, and that are neither liquid nor appreciating. The corporate claims scale with the project’s success; the community benefits are largely fixed at signing. If SB Energy’s valuation doubles after its IPO, OpenAI’s warrants roughly double with it, while Pike County’s $80 million fund remains $80 million. [2][25]
The policy challenge is therefore not the presence of community benefits—the Ohio package is, by historical standards, generous and unusually well-structured—but the absence of community participation in upside. A range of instruments could close the gap without deterring investment: host-community equity trusts holding small stakes in project companies; valuation-milestone payments mirroring the vesting schedules of the corporate warrants; property-tax structures with escalators tied to campus revenue; or state co-investment vehicles of the kind several sovereign funds already operate. The deeper principle is symmetry: if demand can be converted into equity for the tenant, hosting can be converted into equity for the community, and the same financial creativity that produced the Power Warrant can produce its civic counterpart.
4.5 Five Questions for Governors and Federal Policymakers
Drawing the threads of this section together, I propose that policymakers evaluate every future gigawatt-scale AI infrastructure project against five questions, each of which maps directly onto a feature of the Power Warrants architecture documented in this paper.
Question One — Additionality. Is new generation actually being added, in firm capacity and on a binding schedule, sufficient to cover the project’s load? The Ohio answer—at least ten gigawatts of new generation for eight gigawatts of IT load—is the standard against which other proposals should be measured. [8]
Question Two — Cost Responsibility. Who pays for transmission, substations and grid upgrades, under what legally enforceable commitment, and for how long? Voluntary pledges decay; the $4.2 billion SB Energy transmission commitment is meaningful precisely because it is embedded in the AEP Ohio framework rather than in a press release alone. [23]
Question Three — Financial Transparency. Does the AI customer hold equity, warrants, discounts or other financial interests in the infrastructure developer, and are those interests disclosed with the specificity the BIS has demanded—vesting schedules, valuation triggers, guarantee conditions and cross-default exposures? [29][46] The SB Energy draft IPO documents demonstrate that such disclosure is possible; policy should make it routine rather than incidental to a listing.
Question Four — Community Participation. Does the host region participate meaningfully in the economic upside, through instruments that appreciate with the project, rather than solely in fixed benefits that do not?
Question Five — Systemic Exposure. What happens if the AI tenant fails, restructures, reduces demand or changes technology? Who bears the residual value of an eight-gigawatt single-tenant campus—the guarantor, the lenders, the developer’s public shareholders, the federal landowner, or, in extremis, the ratepaying public? The Nvidia guarantee answers this question for the first 4.25 gigawatts of Ohio; [9] most projects elsewhere have no answer at all.
These questions are deliberately nonpartisan. They will be as relevant to a governor courting investment as to a regulator scrutinizing it, and their systematic application would do more to legitimize the Power Warrants economy than any quantity of corporate communication.
Section 5: From Power Warrants to the Financialization of Intelligence Infrastructure
5.1 The Infrastructure Developer Becomes an AI Derivative
If a power-and-datacenter developer’s value becomes heavily dependent on a frontier AI laboratory, then investors buying shares in that infrastructure company are, whether they intend it or not, making a leveraged bet on the laboratory’s future—and SB Energy’s IPO will be the first pure public test of this proposition. Consider what a purchaser of SB Energy shares actually buys: a company with roughly $140 million of first-half operating revenue, a $3.2 billion first-half net loss, no operating data centers, and a $400 billion backlog concentrated overwhelmingly in a single private tenant whose own finances the prospectus flags as a principal risk, supported by a residual-value guarantee from that tenant’s chip supplier. [3][4] Strip away the corporate form and the security resembles a structured note whose coupons are OpenAI lease payments and whose principal protection is a conditional Nvidia guarantee. The infrastructure company functions, in substantial part, as a financial proxy—a derivative—on private AI growth that public investors cannot otherwise access.
This creates a genuinely new asset class, which the market will likely name before academics do: AI-linked infrastructure equity. Its defining property is that it transmits Layer Four and Layer Five outcomes into Layer One and Layer Three valuations with almost no damping. It offers public investors exposure to the AI boom through hard assets, which is its appeal; and it offers them concentration, opacity and reflexivity, which is its danger. The BIS’s warning about poorly disclosed arrangements and multiply pledged assets [46] reads, in this light, less like a general caution than like a pre-review of exactly this asset class’s prospectus risk factors.
5.2 The AI Laboratory Becomes a Quasi-Infrastructure Company
The reverse transformation occurs simultaneously, and it is just as consequential. As frontier laboratories accumulate stakes in energy developers, warrant positions in chip suppliers, equity in datacenter operators and options across the physical economy, they gradually cease to resemble conventional software companies, and their balance sheets begin to resemble those of industrial conglomerates or, more precisely, of the great holding companies of earlier infrastructure eras. OpenAI’s disclosed and reported portfolio already spans the reported $5.5 billion SB Energy warrant position and single-digit equity stake, [2] the AMD warrant convertible into approximately ten percent of that chipmaker, [20][22] equity received through its CoreWeave arrangements, [34] and its historical profit-participation structures with Microsoft—a collection of claims on other companies’ futures that any diversified holding company would recognize as a strategic investment book.
The consequences of this hybridization run in every direction. For valuation, it means a frontier laboratory’s worth is no longer only a multiple of its revenue but also the mark-to-market of its infrastructure claims—a fact that OpenAI’s reported pursuit of a trillion-dollar public listing makes concrete. [45] For governance, it means the laboratory’s incentives now include the share prices of its suppliers and landlords, a web of interests that boards, auditors and eventually securities regulators will have to learn to map. For competition policy, it means that vertical relationships in AI increasingly carry equity ties of exactly the kind antitrust doctrine has historically scrutinized in other network industries. And for the laboratories themselves, it means the model is no longer the only asset that must be defended; the portfolio must be defended too, and portfolios have gravitational effects on strategy that pure research organizations have never had to manage.
5.3 Infrastructure Appreciation Could Subsidize Intelligence
Power Warrants introduces a provocative future possibility that deserves to be stated carefully, because if it materializes it may be remembered as one of the most important financial innovations of the AI economy. Suppose an AI laboratory’s massive demand helps make an energy-and-datacenter developer much more valuable—as OpenAI’s leases have already done for SB Energy, whose warrant package appreciated from an estimated $3.6 billion to $5.5 billion in under six months on the strength of the Ohio contracts. [2] If the laboratory holds warrants or equity in that developer, some of that appreciation can economically offset part of the laboratory’s own infrastructure expenditure. The tenant’s rent inflates the landlord’s valuation; the landlord’s valuation inflates the tenant’s warrants; the warrants, once vested and monetized, refund a portion of the rent.
The laboratory would therefore be simultaneously creating the expense and participating in the asset appreciation caused by the expense—a closed thermodynamic loop of corporate finance in which some fraction of every dollar of compute cost returns as investment gain. In the limit, one can imagine frontier laboratories managing their infrastructure warrant books as strategic treasuries, timing monetizations against training-run expenditures, and reporting an “effective cost of compute” net of infrastructure gains. There is nothing improper about the mechanism in principle—it is vendor financing’s mirror image, and every element can be disclosed—but its systemic meaning should be understood: it lowers the private marginal cost of compute expansion below its social resource cost during booms, which accelerates buildouts, and it reverses violently during busts, when warrant books deflate exactly as compute commitments come due. The subsidy, in other words, is procyclical, and procyclical subsidies are how industries overshoot.
5.4 The Risk of Concentrated Interdependence
The structure’s systemic risk therefore deserves its own unflinching statement. Imagine—no longer hypothetically—a network in which the AI laboratory depends on the datacenter developer for capacity; the developer depends on the laboratory’s leases for essentially all of its future revenue; the developer’s lenders depend on those leases as security; the chip supplier guarantees $105 billion of the infrastructure and books $150 to $200 billion of expected revenue per upgrade cycle from the same campus; the developer purchases the chip supplier’s systems; the laboratory holds appreciating financial interests in the developer; and the developer’s public shareholders hold, at one remove, a claim on the laboratory’s continued solvency. [2][3][9] A problem originating at any node—an AI demand disappointment, a construction failure, a financing shock, a technological discontinuity that strands a GPU generation—propagates through the entire structure not through markets, which can reprice gradually, but through contracts and guarantees, which trigger discontinuously.
This is precisely the topology the Bank for International Settlements identified when it warned that a repricing of AI-related assets could propagate rapidly through a financial network in which non-bank intermediaries hold historically large exposures, and that a major equity-market correction could carry larger macroeconomic consequences today than in the past. [28][30] The Five-Layer AI Economy, under full Power Warrants integration, becomes a Five-Layer balance sheet—and consolidated balance sheets fail in a consolidated way. The honest conclusion is not that the architecture is doomed; it is that the architecture has knowingly traded idiosyncratic risk for systemic risk, purchasing coordination and speed at the price of correlation, and that the institutions best positioned to monitor the trade—securities regulators, utility commissions, central banks—are only beginning to acquire the disclosure tools the task requires.
5.5 The 2027–2030 Power Warrants Scenario
Projecting the framework forward yields a plausible sketch of the infrastructure landscape at the decade’s end. By 2030, the market could contain several competing, partially closed infrastructure ecosystems, each organized around a frontier laboratory or hyperscaler and each bound by its own lattice of warrants, guarantees and exclusivities. One laboratory may be aligned with particular nuclear developers through equity-linked restart and SMR agreements descended from today’s Three Mile Island and Kairos templates. [42] Another may have secured natural-gas generation and storage portfolios in partnership with sovereign capital, on the Ohio federal-ownership model. [24] A hyperscaler may own generation outright, completing the vertical integration its twenty-year PPAs already imply. A semiconductor company may guarantee whole campuses constructed exclusively around its systems, extending the PORTS-Pike exclusivity structure across a portfolio of sites. [9] And a frontier laboratory may hold warrants or options across several infrastructure providers simultaneously, managing the book as a strategic treasury in the manner sketched in Section 5.3.
Competition among frontier AI organizations would then have shifted decisively in character—from who has the smartest model toward who has secured the strongest long-duration portfolio of energy, compute, infrastructure and financial rights necessary to keep improving the model. Benchmarks will still matter, but they will function the way engine performance functioned in twentieth-century aviation: necessary, celebrated, and wholly insufficient without the routes, the airports, the financing and the fleet. The organizations that understood earliest that the AI race had become an infrastructure-rights race—and that infrastructure rights could be owned, optioned and warranted rather than merely rented—will hold advantages compounding across every layer of the economy this paper has described. That is the world the Power Warrants framework predicts, and 2026 is the year its foundations were poured, in public, in Ohio.
Section 6: What Have We Learned?
Pillar 1 — Electricity Is Becoming an Asset, Not Merely an Expense
The first lesson of Power Warrants is that power is changing its economic position inside artificial intelligence at the deepest level of the corporate ledger. In traditional computing, electricity appeared principally on the operating-cost side; in frontier AI, access to electricity determines whether billions of dollars of accelerators can produce revenue at all, which means that secured power now carries option value in the strict financial sense—it is the right, but not the obligation, to deploy compute profitably in the future. The IEA’s projection that data centres will consume more electricity than all American energy-intensive manufacturing combined by 2030 [14] is the macro expression of this shift; the $5.5 billion warrant package attached to an eight-gigawatt lease [2] is its micro expression. The company that controls future power controls future compute, and the company that holds financial rights connected to that power participates in its appreciation. Electricity has crossed from the income statement to the balance sheet, and it will not cross back.
Pillar 2 — AI Demand Is Becoming a Form of Capital
A credible multi-gigawatt AI customer does far more than consume infrastructure; its commitment gets infrastructure financed. Its lease supports borrowing, as SB Energy’s $400 billion backlog supports an IPO that its $140 million of operating revenue never could. [4] Its reputation attracts co-investors, as OpenAI’s tenancy attracted Nvidia’s $1.5 billion and Ares’ continued participation. [3][8] Its expansion plans justify transmission, as eight gigawatts of leases justified $4.2 billion of 765-kilovolt construction. [23] And its name reshapes regional electricity planning, as a single announcement redrew the load forecast of Appalachian Ohio. Artificial-intelligence demand has begun to function as a form of industrial capital formation, and the Power Warrant is simply the instrument through which the holders of that demand-capital collect their return on it. Economists have long understood that a committed buyer creates value for a seller; what is new is that the buyers have institutionalized the collection of that value in equity form, at billion-dollar scale, as a standard term of trade.
Pillar 3 — The Five Layers Are Becoming Financially Interlocked
Energy companies are moving toward datacenters; chip companies are moving toward infrastructure finance; datacenter developers are aligning their entire capital structures with model laboratories; model laboratories are investing downward into energy and hardware; and application revenue at Layer Five is asked to support the whole arrangement. The Five-Layer AI Economy is therefore no longer simply a vertical technology supply chain in which each layer sells to the one above it. It is becoming a network of cross-ownership, guarantees, contracts, financing commitments and strategic dependencies in which each layer holds claims on the others—a development the BIS has now formally classified among the pressure points of global financial stability, [28][29] and which this paper has traced transaction by transaction from the January SB Energy investment to the August warrant disclosure. Interlocking is not inherently pathological; it is how industrial systems coordinate at the frontier. But an interlocked system must be analyzed as a system, and 2026 is the year that analytical obligation became unavoidable.
Pillar 4 — Corporate Upside Requires Public Accountability
Power Warrants creates legitimate opportunities for corporations to reduce infrastructure risk and to participate in economic value they genuinely help create; nothing in this paper argues otherwise. But the same structure makes transparency a load-bearing element of the entire edifice. If corporations obtain warrants, equity, discounts and strategic upside from projects requiring enormous quantities of electricity, land, water and public infrastructure—including, in the Ohio case, federal land and federally owned generation [24]—then policymakers and the public must be able to see who receives the upside and who absorbs the downside, with the specificity of vesting schedules and guarantee conditions rather than the vagueness of press releases. Ratepayer protection of the kind AEP Ohio and SB Energy engineered, [23] community participation of the kind Section 4.4 proposes, and disclosure of the kind the SB Energy IPO documents inaugurated [2][3] are not obstacles to the Power Warrants economy. They are its license to operate, and the firms that internalize this earliest will face the least political friction across the decade.
Pillar 5 — The Next AI Advantage May Be Financial Architecture
The decisive competitive advantage of 2027 through 2030 may not belong exclusively to the laboratory with the highest benchmark score. It may belong to the organization that assembles the strongest combination of capital, energy, chips, datacenters, models and applications into a mutually reinforcing structure—the organization, in short, that practices financial architecture as a first-class engineering discipline. The evidence assembled in this paper points consistently in that direction: AMD won six gigawatts of deployment partly with a warrant; [20] Nvidia secured a twenty-year exclusive campus partly with a guarantee; [9] OpenAI converted tenancy into $5.5 billion of contingent equity; [2] and the hyperscalers converted twenty-year power commitments into the effective ownership of America’s nuclear renaissance. [42] The frontier model is becoming one component—the most glamorous component—of a much larger industrial system, and Power Warrants is the name of the discipline that designs the rest of it.
Pillar 6 — The Warrant Is Becoming a Standard Instrument of the AI Economy
A pattern repeated three times in thirteen months is no longer an anomaly; it is a market convention forming in real time. October 2025: AMD issues OpenAI a warrant for up to ten percent of the company, vesting with gigawatts deployed. [20][22] January 2026: SB Energy issues OpenAI warrants valued at $3.6 billion, vesting with post-IPO valuation milestones. [2] Through 2025 and 2026: equity-linked compute arrangements proliferate across the CoreWeave, Nvidia–OpenAI and related transactions that Bloomberg’s mapping of the circular-deal economy documents. [34][41] The direction of travel is unmistakable: wherever a frontier buyer’s commitment is the scarce input that makes a seller’s growth story credible, the buyer now demands—and receives—equity-linked consideration as a standard term. Historians of finance may eventually rank the compute warrant alongside the convertible bond and the production payment as an instrument that a specific industrial revolution invented because it needed it. The practical implication for every participant in the AI economy, from suppliers to regulators to investors, is to assume that the next gigawatt-scale agreement will carry a warrant, and to build analysis, disclosure and valuation practice accordingly.
Pillar 7 — Geography Is Destiny Again, and Communities Are Learning to Negotiate
The final lesson is territorial. The Power Warrants economy re-anchors the most valuable industry in the world to specific places—to Milam County, Texas and Pike County, Ohio, to transmission corridors and water systems and building-trades halls—after three decades in which software famously floated free of geography. That re-anchoring restores a form of leverage that communities and states had largely lost: the leverage of the host. Ohio’s experience shows both the promise and the current limits of that leverage—$4.2 billion of privately funded transmission, an $80 million community fund, $84 million of student credits, union agreements and annual public reporting on one side; [12][23][25][26] no appreciating claim on the project’s success on the other. The next generation of host negotiations will be conducted by communities that have read the SB Energy prospectus, and they will ask for the same instrument the tenant received. The evolution from community benefits to community equity is the natural completion of the Power Warrants framework, and the jurisdictions that pioneer it will define the social contract of the AI infrastructure age.
Conclusion — Why “Power Warrants” Fits the Emerging AI Economy
Artificial intelligence began the decade as a software story. Then it became a semiconductor story, as the world learned to count GPUs. Then it became a datacenter story, as the world learned to count gigawatts. Now it is becoming an energy story, as the world learns to count transmission lines, gas turbines and restarted reactors. But the events surrounding OpenAI, SB Energy and Nvidia in 2026 suggest that the next stage will be more complicated still, and that its proper name is a financial architecture story—a story about who owns the claims on the machine, not merely who builds or uses it.
The August 31 report concerning approximately $5.5 billion in SB Energy warrants issued to OpenAI matters because it provides an unusually vivid example of the boundary between AI customer and infrastructure owner becoming indistinct. OpenAI, if the reporting is accurate, did not merely secure access to infrastructure; it obtained an instrument giving it financial participation in the future value of the infrastructure company itself—an instrument that has already appreciated by more than half since issuance, on the strength of OpenAI’s own leases. [1][2] Reuters emphasized that it could not independently verify the Journal’s account, and this paper has treated the specific transaction with corresponding care throughout. [1] But the broader structural direction does not depend on any single report, because it is independently visible in the disclosed record: in the January investments and the Texas lease, [5] in the seventeen Ohio leases and the eight gigawatts, [4][12] in the $4.2 billion transmission commitment, [23] in the $105 billion of guarantees in Nvidia’s SEC filings, [9] and in the AMD warrant sitting in a 10-K for anyone to read. [21]
Nvidia’s role makes the transformation clearest of all. The company whose accelerators sit inside the AI factory is helping to secure the land, power and shell required to construct the factory, providing conditional credit support at a scale exceeding the market capitalization of most public utilities, investing in the infrastructure developer, and ensuring by contract that its own computing architecture occupies the resulting capacity for twenty years. [8][9] OpenAI provides the demand; SB Energy provides the development platform; Nvidia provides the computing technology and the financial reinforcement; and capital moves among the three while long-term contracts bind their economic futures together. Around them, the wider economy has organized itself in sympathy: three-quarters of a trillion dollars of hyperscaler capital expenditure in a single year, [17] a nuclear renaissance underwritten by twenty-year corporate offtakes, [42] national growth accounts visibly reshaped by the buildout, [36][37] and the central bank of central banks formally listing the structure among the chief risks to global financial stability. [28][30] More than two hundred economists and AI experts, including seventeen Nobel laureates, gathered by Stanford’s Digital Economy Lab, chose this same moment to urge the world to
“build the incentives, guardrails, and institutions needed to steer AI” — Statement organized by the Stanford Digital Economy Lab, July 2026 [38]
toward broadly shared benefit—and the argument of this paper is that the incentives, guardrails and institutions of the financial architecture beneath AI deserve a central place in that project.
This is why Power Warrants fits the paper, and why I believe it names the era. The phrase begins with a literal financial instrument disclosed in a draft prospectus, but its larger meaning extends across every layer of the AI economy. Power represents the increasingly scarce physical foundation beneath intelligence—the electrons, corridors and campuses without which the most sophisticated model is inert mathematics. Warrants represent the financial claims, options and economic participation that now accompany access to that foundation—the mechanism by which demand is converted into ownership. Together, they describe a future in which frontier AI laboratories will no longer simply ask how much electricity can we buy? They will increasingly ask what financial rights can we secure in the infrastructure our demand causes to be built?
That distinction matters more than any single valuation in this paper. If the answer increasingly includes warrants, equity, guarantees, preferential capacity, development rights and ownership optionality—and the record of 2025 and 2026 says that it does—then the AI economy will have crossed another important boundary. The most powerful artificial-intelligence companies will no longer merely sit at the top of the Five-Layer AI Economy consuming what the layers below them produce. They will reach downward into those layers—financing them, reorganizing them, guaranteeing them and, in a growing number of cases, obtaining financial interests in them. The Five-Layer AI Economy will therefore cease to function merely as a supply chain and will become an interconnected system of technology, infrastructure and ownership, with all of the coordination benefits and all of the correlated fragilities that consolidated systems carry.
Power Warrants is the term that captures that transition. The work of the coming years—for scholars, for investors, for regulators, and for the communities on whose land the intelligence economy is being built—is to make sure the transition is understood, disclosed and governed while its terms are still being written.
Footnotes and Endnotes:
[1] Reuters (syndicated via Business Recorder), “OpenAI issued warrants worth $5.5 billion in SB Energy, WSJ reports,” August 31, 2026. https://www.brecorder.com/news/40437215/openai-issued-warrants-worth-55-billion-in-sb-energy-wsj-reports
[2] Investing.com (citing The Wall Street Journal), “SB Energy offered OpenAI $5.5 bln in warrants to secure data-center deal — WSJ,” August 31, 2026. https://www.investing.com/news/stock-market-news/sb-energy-offered-openai-55-bln-in-warrants-to-secure-datacenter-deal–wsj-4882171
[3] Quartz (via Yahoo Finance), “SoftBank SB Energy gave OpenAI $5.5 billion in stock warrants,” August 31, 2026. https://finance.yahoo.com/technology/ai/articles/softbank-sb-energy-gave-openai-115437263.html
[4] Yahoo Finance, “OpenAI Got $5.5B in SB Energy Warrants to Lease Nvidia-Backed AI Data Centers,” August 31, 2026. https://finance.yahoo.com/technology/ai/articles/openai-got-5-5b-sb-190151986.html
[5] OpenAI, “OpenAI and SoftBank Group partner with SB Energy,” January 9, 2026. https://openai.com/index/stargate-sb-energy-partnership/
[6] CNBC (Greg Brockman statement), “OpenAI and SoftBank announce $1 billion investment in SB Energy as part of massive AI buildout,” January 9, 2026. https://www.cnbc.com/2026/01/09/openai-and-softbank-group-announce-1-billion-investment-in-sb-energy-.html
[7] Reuters (via U.S. News & World Report), “OpenAI, SoftBank Invest $1 Billion in SB Energy as Stargate Buildout Expands,” January 9, 2026. https://money.usnews.com/investing/news/articles/2026-01-09/openai-softbank-invest-1-billion-in-sb-energy
[8] NVIDIA Newsroom, “NVIDIA Guarantees SB Energy’s PORTS-Pike Technology Campus in Ohio to Exclusively Host NVIDIA AI Compute,” August 17, 2026. https://nvidianews.nvidia.com/news/nvidia-guarantees-sb-energy-s-ports-pike-technology-campus-in-ohio-to-exclusively-host-nvidia-ai-compute
[9] NVIDIA Corporation, Q2 Fiscal 2027 CFO Commentary (Form 8-K exhibit), U.S. Securities and Exchange Commission, August 26, 2026. https://www.sec.gov/Archives/edgar/data/0001045810/000104581026000073/q2fy27cfocommentary.htm
[10] CNBC, “Nvidia backing $105 billion in financing for OpenAI data center in Ohio,” August 17, 2026. https://www.cnbc.com/2026/08/17/nvidia-financing-open-ai-data-center-ohio.html
[11] Data Center Knowledge, “Nvidia Backs OpenAI’s Ohio Data Center Buildout With $105B Guarantee,” August 2026. https://www.datacenterknowledge.com/data-center-construction/nvidia-backs-openai-s-ohio-data-center-buildout-with-105b-guarantee
[12] OpenAI, “OpenAI joins PORTS-Pike project,” August 17, 2026. https://openai.com/index/openai-joins-ports-pike-project/
[13] FINRA (Financial Industry Regulatory Authority), “SPAC Warrants: 5 Tips to Avoid Missed Opportunities” (definition of warrants), 2021. https://www.finra.org/investors/insights/spac-warrants-5-tips
[14] International Energy Agency, Energy and AI — Executive Summary, World Energy Outlook Special Report. https://www.iea.org/reports/energy-and-ai/executive-summary
[15] International Energy Agency (Fatih Birol statement), “AI is set to drive surging electricity demand from data centres…,” April 2025. https://www.iea.org/news/ai-is-set-to-drive-surging-electricity-demand-from-data-centres-while-offering-the-potential-to-transform-how-the-energy-sector-works
[16] Brookings Institution, “Global energy demands within the AI regulatory landscape,” June 2026. https://www.brookings.edu/articles/global-energy-demands-within-the-ai-regulatory-landscape/
[17] Tom’s Hardware (citing Financial Times and Jefferies’ Brent Thill), “Google, Microsoft, Meta, and Amazon capex spending to hit $725 billion in 2026, up 77% from last year,” April 2026. https://www.tomshardware.com/tech-industry/big-tech/big-techs-ai-spending-plans-reach-725-billion
[18] NVIDIA Corporation (Jensen Huang statement), “NVIDIA Announces Financial Results for Second Quarter Fiscal 2027,” August 26, 2026. https://www.sec.gov/Archives/edgar/data/0001045810/000104581026000073/q2fy27pr.htm
[19] CNBC (Colette Kress statement), “Nvidia earnings takeaways: Huang forecasts 70% fiscal 2028 revenue growth,” August 26, 2026. https://www.cnbc.com/2026/08/26/nvidia-nvda-earnings-report-q2-2027-live-updates.html
[20] Advanced Micro Devices, Inc., “AMD and OpenAI Announce Strategic Partnership to Deploy 6 Gigawatts of AMD GPUs,” October 6, 2025. https://ir.amd.com/news-events/press-releases/detail/1260/amd-and-openai-announce-strategic-partnership-to-deploy-6-gigawatts-of-amd-gpus
[21] Advanced Micro Devices, Inc., Form 10-K/A (OpenAI warrant disclosure), U.S. Securities and Exchange Commission. https://www.sec.gov/Archives/edgar/data/2488/000000248826000021/amd-20251227.htm
[22] CNBC, “OpenAI looks to take 10% stake in AMD through AI chip deal,” October 6, 2025. https://www.cnbc.com/2025/10/06/openai-amd-chip-deal-ai.html
[23] AEP Ohio (Marc Reitter statement), “AEP Ohio Partners to Bring $4.2B in New Electric Infrastructure in Appalachian Ohio Without Raising Customer Rates,” March 20, 2026. https://www.aep.com/news/stories/view/10823/
[24] PORTS-Pike Technology Campus (project site; U.S. DOE / SB Energy materials), “PORTS-Pike Technology Campus.” https://portscampus.com/
[25] Data Center Frontier, “PORTS-Pike Takes Shape as an 8-GW AI Infrastructure Model,” August 2026. https://www.datacenterfrontier.com/hyperscale/article/55398883/ports-pike-takes-shape-as-an-8-gw-ai-infrastructure-model
[26] Dayton Daily News (Jensen Huang statement; Codex credits), “OpenAI announces role in Pike County AI campus expected to create 35,000 construction jobs,” August 2026. https://www.daytondailynews.com/local/openai-announces-role-in-pike-county-ai-campus-expected-to-create-35-000-construction-jobs/article_03f4525c-a96b-51f6-8961-e1a094df0141.html
[27] Stop Ohio Data Centers (Masayoshi Son groundbreaking statement; project documentation), “Piketon, Ohio: PORTS Technology Campus 10 GW Data Center,” 2026. https://stopohiodatacenters.org/pike-county
[28] Fortune, “The central bank of central banks just released its flagship annual report — and it sees a $1 trillion AI investment boom headed for a reckoning,” June 29, 2026. https://fortune.com/2026/06/29/bis-central-bank-warning-hyperscaler-data-center-1-trillion-gamble-recession/
[29] Business Standard, “BIS flags AI spending boom as growing threat to global financial stability,” June 30, 2026. https://www.business-standard.com/technology/tech-news/bank-for-international-settlements-flags-ai-spending-boom-as-growing-threat-to-global-financial-stability-126063000669_1.html
[30] Bloomberg (via Yahoo Finance), “AI Bust Risks Ripple Effects From Growth to Credit, BIS Says,” June 28, 2026. https://finance.yahoo.com/economy/policy/articles/ai-bust-risks-ripple-effects-090000030.html
[31] Bank for International Settlements, Working Paper No. 1367, “The AI investment race,” 2026. https://www.bis.org/publ/work1367.pdf
[32] Malay Mail (Reuters/AFP reporting; Pierre-Olivier Gourinchas statements), “Global growth gets a 2026 upgrade, but IMF says AI hype could still unravel the boom,” January 19, 2026. https://malaymail.com/news/money/2026/01/19/global-growth-gets-a-2026-upgrade-but-imf-says-ai-hype-could-still-unravel-the-boom/206024
[33] International Monetary Fund Media Center (Pierre-Olivier Gourinchas), “IMF — World Economic Outlook Press Briefing,” April 2026. https://mediacenter.imf.org/news/imf—world-economic-outlook-press-briefing/s/50073691-4f97-4b23-a3ab-35990f7e468f
[34] NPR (Daron Acemoglu interview), “Here’s why concerns about an AI bubble are bigger than ever,” November 23, 2025. https://www.npr.org/2025/11/23/nx-s1-5615410/ai-bubble-nvidia-openai-revenue-bust-data-centers
[35] Fortune (Daron Acemoglu interview), “Nobel Laureate Daron Acemoglu on the ‘brainless’ AI discourse…,” June 21, 2026. https://fortune.com/2026/06/21/nobel-laureate-daron-acemoglu-ai-productivity-capitalism-democracy/
[36] Fortune (Jason Furman analysis), “Without data centers, GDP growth was 0.1% in the first half of 2025, Harvard economist says,” October 7, 2025. https://fortune.com/2025/10/07/data-centers-gdp-growth-zero-first-half-2025-jason-furman-harvard-economist
[37] ING Think, “How much is AI contributing to US economic growth?,” August 2026. https://think.ing.com/opinions/how-much-is-ai-contributing-to-us-economic-growth/
[38] Stanford Digital Economy Lab (Erik Brynjolfsson, Director; statement of 200+ economists and 17 Nobel laureates), July 2026. https://digitaleconomy.stanford.edu/
[39] Business Standard (Stacy Rasgon, Bernstein Research), “Nvidia-OpenAI deal sparks concerns over circular financing in AI boom,” September 24, 2025. https://www.business-standard.com/amp/companies/news/nvidia-openai-deal-sparks-concerns-over-circular-financing-in-ai-boom-125092401589_1.html
[40] Axios, “Nvidia reignites ‘circular’ AI concerns as it weighs OpenAI financing guarantee,” July 27, 2026. https://www.axios.com/2026/07/27/nvidia-openai-financing-ai-jensen-huang-ssi
[41] Bloomberg, “AI Circular Deals: How Microsoft, OpenAI and Nvidia Keep Paying Each Other,” January 2026. https://www.bloomberg.com/graphics/2026-ai-circular-deals/
[42] SMR Intel, “Every Nuclear-Powered Data Center Deal: Google, Amazon, Meta & Microsoft (2026),” 2026. https://smrintel.com/nuclear-data-center-deals/
[43] Data Center Frontier, “Data Center Nuclear Power Update: Microsoft, Constellation, AWS, Talen, Meta,” 2024–2026. https://www.datacenterfrontier.com/energy/article/55239739/data-center-nuclear-power-update-microsoft-constellation-aws-talen-meta
[44] Quash (compiling OpenAI, The Verge and CNBC disclosures), “ChatGPT Statistics: Users, Revenue, Downloads and Usage,” August 2026. https://quashbugs.com/blog/chatgpt-statistics-users-revenue-downloads-and-usage
[45] AI Business Weekly, “OpenAI Statistics 2026: Users, Revenue & Valuation,” July 2026. https://aibusinessweekly.net/p/openai-statistics
[46] Capacity, “What is circular financing in AI infrastructure, and should telecoms and data centre operators be worried?,” August 2026. https://capacityglobal.com/news/what-is-circular-financing/
