Nvidia May Guarantee $250B So OpenAI Can Rent a Data Center

The Wall Street Journal reported this week that Nvidia is in talks to guarantee roughly $250 billion in financing so OpenAI can lease a data center campus SoftBank is building in southern Ohio — on the site of a former uranium-enrichment facility about 50 miles south of Columbus. The campus is planned to eventually reach 10 gigawatts of capacity, and once you add in the cost of the Nvidia chips needed to fill it, the whole project could run past $500 billion. That would make it the largest single data center buildout ever announced. It also reignited a fight that’s been simmering all year over whether Nvidia is quietly financing its own sales.

What Nvidia is actually guaranteeing

It helps to be precise about what a “$250 billion guarantee” means here, because it’s not Nvidia handing OpenAI a check. OpenAI doesn’t have an investment-grade credit rating, which makes lenders nervous about financing a project this size on OpenAI’s promise alone. Nvidia’s guarantee is aimed at that specific problem: it would backstop the lease and construction financing for the Ohio campus, giving banks and bondholders a much stronger counterparty to rely on than OpenAI by itself. That’s separate from a second, reportedly larger arrangement — worth up to $350 billion — where Nvidia would help finance OpenAI’s actual purchases of Nvidia chips for the site.

The power for the first phase, about 800 megawatts due by 2028, is set to come from a natural-gas plant backed by $33 billion in Japanese investment tied to a recent US-Japan trade agreement. So this one project touches a foreign chipmaker’s balance sheet, a decommissioned nuclear site, a Japanese energy investment, and US debt markets, all to keep one company’s model training and inference running.

Why “circular financing” is the phrase everyone’s using

The criticism, from investor Michael Burry among others, is straightforward: Nvidia sells the chips, invests in or guarantees financing for the companies buying those chips, and books revenue on both ends of what looks increasingly like one closed loop. Burry’s reaction to the report was blunt — “Around and around we go… Nvidia to guarantee $200 billion of ChatGPT’s spending on $NVDA chips.” Jim Chanos, another well-known short seller, made the same point in slightly more technical terms: Nvidia is effectively financing its own chip sales, which means a chunk of Nvidia’s revenue growth depends on Nvidia’s own guarantees rather than independent demand.

Nvidia has publicly pushed back on the “circular financing” framing, and it’s worth taking that pushback seriously rather than assuming the critics are automatically right. A guarantee isn’t a purchase — Nvidia isn’t buying OpenAI’s compute or manufacturing demand out of thin air, it’s making a real project easier to finance by putting its own balance sheet on the line. That’s a normal thing suppliers do for major customers in capital-intensive industries; chipmakers, aircraft manufacturers, and telecom equipment vendors have financed their customers’ purchases for decades without it being described as a scheme. The more useful question isn’t whether this arrangement is circular in some technical sense — it’s whether Nvidia’s guarantee obligations are large enough, relative to Nvidia’s own balance sheet, that a stumble at OpenAI could hit Nvidia directly instead of staying OpenAI’s problem alone. That’s a real and answerable question, and it’s the one the Bank for International Settlements flagged in its 2026 annual report when it called out AI infrastructure financing arrangements as a source of systemic risk — not because any single deal is fraudulent, but because so much of the industry’s growth now rests on a small number of companies guaranteeing each other’s spending.

Why this is happening at all

Step back from the financing mechanics and the underlying reason is simple: training and running frontier models takes an amount of compute that no company’s cash flow can fund up front, so the entire industry runs on debt and guarantees instead. We wrote last year about OpenAI floating a government equity stake partly as a way to shore up its position with a stakeholder whose backing carries weight with lenders and regulators — this Nvidia guarantee is the same underlying problem, solved with a different partner. OpenAI needs capacity it can’t yet pay for outright; Nvidia has an overwhelming interest in OpenAI getting that capacity, since more capacity generally means more chip orders. The guarantee is one way to align those two needs without OpenAI having to raise the money on its own credit.

That alignment of interests is exactly what makes the “is this circular” question hard to settle cleanly. Of course Nvidia benefits from OpenAI’s growth — that’s true of any supplier and any major customer, and it would be true even without a guarantee attached. The guarantee just makes the relationship visible and quantifiable in a way that ordinary vendor-customer dynamics usually aren’t. Whether that visibility reveals a genuine systemic risk or just an unusually well-documented version of an ordinary business relationship is going to depend on details that haven’t been made public yet — how large Nvidia’s total guarantee exposure is across all its AI customers, not just this one deal, and what happens to that exposure if AI demand growth slows down before 2028.

The skeptical read

Nothing about a supplier guaranteeing a customer’s financing is new or inherently sketchy — this happens constantly outside AI, and Nvidia has real standing to argue it’s just doing that at a larger scale than usual because the deals involved are larger than usual. But “usual, just bigger” is also exactly what people said about several financing structures that didn’t hold up when growth slowed, and the Bank for International Settlements didn’t flag AI infrastructure financing for no reason. The honest position is that this deal alone doesn’t prove either story. What would actually settle it is a number nobody’s published yet: how much of Nvidia’s total guaranteed exposure across every AI customer would survive a slowdown in AI spending growth without becoming Nvidia’s own balance sheet problem. Until that number is public, “circular financing” and “ordinary supplier support” are both readings of the same set of facts, not competing facts themselves — which is worth remembering the next time a headline number this large shows up without the exposure math behind it, the same caution we’d apply to any single benchmark or funding claim in this space.

Sources: WSJ via Yahoo Finance, Proactive Investors, TipRanks on Burry’s reaction, Benzinga on Chanos.