Nvidia Is Co-Signing OpenAI's $500 Billion Debt — And the Numbers Don't Add Up the Way Anyone Is Saying
On Sunday July 27, the Wall Street Journal reported something that had to be read twice to believe. Nvidia — the chip company whose GPUs power the AI boom — is in talks to guarantee $250 billion in financing for OpenAI's planned data centre campus in Ohio. Not invest. Not lend. Guarantee. As in: if OpenAI cannot make its lease payments or service the construction debt on the world's largest planned AI facility, Nvidia steps in and covers the obligation. Bloomberg and Reuters confirmed the report within hours. The story has not gone away. And the more closely you examine what is actually being proposed, the more extraordinary it becomes.
This is not a normal deal. This is a chip company co-signing a loan for its biggest customer because the customer's credit rating is too poor to borrow on its own — while separately discussing another financing package of up to $350 billion to fund that same customer's chip purchases. From Nvidia. The total potential exposure is $600 billion. The circular nature of the arrangement is not a bug. It is the entire structure of how the AI industry is financing itself right now.
What Is Actually Being Proposed — The Deal Explained
The Ohio campus at the centre of this deal sits on a decommissioned uranium enrichment plant in Piketon, southern Ohio, roughly 50 miles south of Columbus. The facility is being developed by SB Energy — SoftBank's power and energy subsidiary — and is designed at full build-out to deliver 10 gigawatts of computing capacity. For context: 10GW is approximately equal to the entire current electricity consumption of Portugal. The first phase, targeted to deliver around 800 megawatts of capacity, is scheduled to come online in 2028. The electricity for the project is controlled by the US government and separately funded by Japan through a $33 billion commitment tied to the recent US-Japan trade deal — a geopolitical dimension of the campus that has received almost no coverage.
Here is the core problem that made Nvidia's involvement necessary: OpenAI cannot borrow money at scale on its own. Despite projecting approximately $25 billion in revenue for 2026, OpenAI remains deeply unprofitable and carries no investment-grade credit rating. When SB Energy approached debt markets to raise construction financing for the Piketon campus with OpenAI as the anchor tenant, lenders said no — or quoted terms that made the project unviable. The proposed Nvidia guarantee solves that problem by substituting Nvidia's creditworthiness for OpenAI's, allowing SB Energy to raise construction debt on terms it could not otherwise access.
What Nvidia receives in return is less tangible but arguably more valuable than any fee: guaranteed chip demand for years. A 10GW AI campus running Nvidia hardware represents a multi-year purchase commitment that locks in revenue visibility no quarterly earnings call can provide. The guarantee is, in this light, an extraordinarily expensive customer acquisition cost.
— Michael Burry, Substack post July 24, 2026 — who simultaneously disclosed he had expanded his short position in Nvidia
The Circular Financing Problem Nobody Wants to Name Directly
Michael Burry — the investor made famous by The Big Short for correctly calling the 2008 housing collapse before almost everyone else — disclosed an expanded short position in Nvidia on July 24 and reacted to the guarantee news with the quote above. His framing was slightly imprecise on the dollar amount but structurally accurate to what is being described: Nvidia finances OpenAI's ability to lease infrastructure, and OpenAI uses that infrastructure to buy Nvidia chips, and Nvidia uses those chip sales to service the guarantee exposure it took on to make the deal possible.
This is circular financing. And it is not unique to this deal — it is the entire architecture of how AI investment has been structured in 2026. Nvidia is reportedly in talks with OpenAI to provide $250 billion in financing for a data centre project in Ohio, once again raising questions about circular financing in the AI industry, as Yahoo Finance noted in its Big Tech earnings preview this week. ZeroHedge made the structural critique even more explicit: "The problem with the $2 trillion in circular AI financing is that it is all contingent on the frontier AI companies being money good on their $1.5 trillion in unfunded commitments — which they won't be if Chinese open LLMs grab market share."
That last clause is the part the industry does not want to confront. The entire financial architecture of the Piketon campus — and of the broader $500 billion commitment — is premised on OpenAI's revenues growing fast enough to service the obligations as they mature. If DeepSeek, Qwen, or another Chinese open-source model dramatically closes the capability gap with GPT-class models, the revenue trajectory that makes this structure solvent becomes significantly less certain.
What This Means for OpenAI — Breaking Free of Microsoft
For OpenAI, the Ohio campus represents something beyond computing capacity. For OpenAI, a deal would mark a first step toward controlling its own infrastructure rather than renting computing capacity from Microsoft, Amazon and Oracle. Since its founding, OpenAI has been almost entirely dependent on Microsoft Azure for its compute — a dependency baked into the terms of Microsoft's $13 billion investment. Building and operating its own 10GW campus would fundamentally change OpenAI's cost structure, its leverage in negotiations with cloud providers, and its long-term viability as an independent company rather than a de facto Microsoft subsidiary.
The Piketon campus, if fully built and operational by the early 2030s, would make OpenAI one of the largest computing infrastructure operators on Earth — comparable in scale to the hyperscale divisions of Amazon, Google, and Microsoft that took decades and hundreds of billions of dollars to build. The ambition is not modest. Neither is the risk.
What This Means for Nvidia — Upside and Exposure
Nvidia's stock dropped sharply on the news — Nvidia fell nearly 5% on Monday July 28 as investors processed the implications of a chip company assuming contingent liability of this scale. The concern is rational. If the guarantee is formalised, Nvidia would carry a $250 billion contingent obligation on its balance sheet — an exposure it could not book as revenue and could not offset with equity upside, since Nvidia is not taking an ownership stake in the Piketon campus in exchange for the guarantee.
The bull case for Nvidia is equally clear: a formalised 10GW commitment from OpenAI to run Nvidia chips locks in demand that makes the company's current valuation look conservative. For Nvidia, the arrangement would help guarantee demand for its chips for years to come. Every GPU that goes into Piketon is a Nvidia GPU. The revenue visibility created by a multi-year, 10-gigawatt anchor tenant commitment is worth billions in multiple expansion on its own — which is why Nvidia's board is apparently willing to consider an arrangement that, viewed purely as a balance sheet instrument, looks alarming.
| Party | What They Get | What They Risk |
|---|---|---|
| Nvidia | Guaranteed chip demand for years — 10GW of Nvidia hardware locked in | $250B contingent liability if OpenAI cannot pay — carried off balance sheet |
| OpenAI | Own infrastructure — breaks Microsoft Azure dependency — $25B+ annual savings at scale | Debt obligations contingent on revenue growth that has not yet materialised at this scale |
| SoftBank / SB Energy | Largest data centre project in history — access to US government power at Piketon | Construction execution risk — project depends on OpenAI remaining a viable anchor tenant |
| Microsoft | Nothing — this deal is designed to reduce OpenAI's dependence on Microsoft | Losing OpenAI as an exclusive Azure customer is a multi-billion dollar revenue risk |
| US Government | Domestic AI compute sovereignty — largest AI campus in the world on US soil | Minimal direct financial exposure — power access decisions carry geopolitical weight |
The Broader Context — Big Tech Earnings Week Begins
The Nvidia-OpenAI story broke at the worst possible time for investor nerves — the opening of Big Tech earnings week, with Microsoft, Meta, Apple, and Amazon all reporting results this week. Google raised its expected 2026 capital spending range to between $195 billion and $205 billion, saying the investment is necessary to meet demand for AI computing and support growth across Google Cloud, Gemini, and its consumer services — a number so large it pushed Google's free cash flow negative. Intel delivered stronger-than-expected quarterly results and raised its outlook, offering fresh evidence that the company is beginning to capture more of the spending flowing into AI infrastructure.
The pattern is consistent across every major company reporting this week: AI capital expenditure is accelerating, not stabilising. Google is spending $200 billion on infrastructure. Microsoft is building data centres across three continents. Amazon is expanding AWS at a rate that would have seemed impossible four years ago. And now Nvidia — whose chips power all of it — is discussing guaranteeing $250 billion of its biggest customer's debt to ensure the spending continues.
The Bull Case — Why This Could Work
- 10GW of anchor compute locks in Nvidia chip demand for a decade
- OpenAI's $25B 2026 revenue trajectory makes the debt serviceable if growth holds
- US government power access gives the project a strategic backstop
- Breaking from Microsoft Azure could save OpenAI billions annually at scale
- Nvidia's revenue visibility justifies the contingent liability in analyst models
- First-mover infrastructure at this scale creates durable competitive advantage
The Bear Case — What Could Go Wrong
- OpenAI remains unprofitable — revenue growth is not guaranteed to hold
- Chinese open-source models could compress OpenAI's pricing power and margins
- $250B contingent liability cannot be offset with equity — it is pure downside exposure
- Circular financing means the whole structure unravels if one party cannot perform
- Michael Burry short position signals serious structural scepticism from a credible voice
- Phase 1 not operational until 2028 — two years of construction risk before revenue flows
Nvidia co-signing OpenAI's debt is the most consequential financial story in tech since Microsoft's $13 billion OpenAI investment — and it is structurally more complex and more fragile. The deal, if completed, would create the largest data centre campus in human history on a former uranium enrichment site in Ohio, financed by a chipmaker guaranteeing its own customer's lease so that customer can buy more of the chipmaker's chips. It is circular, it is enormous, and it is precisely what happens when an industry's capital requirements outgrow conventional financing structures before the revenue to support them has materialised. The bull case is that OpenAI's revenue grows fast enough to make the structure sound in retrospect. The bear case is that it doesn't — and that $250 billion in contingent liability lands on Nvidia's balance sheet at the worst possible time. The answer will define the AI industry for the next decade. We will find out which version is right starting in 2028.