Nvidia is in talks to secure $250bn of OpenAI data center debt

Nvidia is in talks to provide nearly $250 billion in securities for OpenAI, the Wall Street Journal reported on Sunday, citing people familiar with the matter. The backstop will allow OpenAI to lease a 10-gigawatt campus in southern Ohio being developed by energy company SoftBank.
The reason such a guarantee is required is straightforward: OpenAI does not have an investment-grade credit rating. Nvidia’s balance sheet will be able to represent one, allowing lenders to price the debt against the manufacturer’s credit rather than the borrower’s.
What does money include?
The $250 billion figure applies to data center lease debt and construction debt, not to the Nvidia chips that will fill the building. Those are the subject of a separate interview, with the Journal reporting chip purchases could reach $350 billion.
The entire project is expected to cost more than 500 billion dollars including silicon. Its first phase, about 800 megawatts, will be completed in 2028.
Nvidia’s support will support financing vehicles designed to secure program financing for lenders. Reuters said it could not immediately confirm the report, while Nvidia, OpenAI, and the US Commerce Department did not respond to requests for comment.
Why is the site political?
TNW City Coworking Space – Where your best work happens
A workplace designed for growth, collaboration, and endless networking opportunities at the heart of technology.
The Ohio campus power is controlled by the US government and separately funded by Japan under a recent trade deal, which coincides with Tokyo’s pledge to invest $33 billion in the natural gas industry. Commerce Secretary Howard Lutnick is involved in deciding who gets access.
OpenAI has been in advanced discussions for the site for several weeks and is among the most enthusiastic bidders. Anthropic, Microsoft, and Google have also approached Lutnick in recent weeks, according to the Journal.
Circulation problem
Michael Burry, an investor known for shorting the real estate market before 2008, posted on X: “Around and around we go. Nvidia will guarantee $200 billion for ChatGPT to use $NVDA chips.” He added to his Nvidia short position on Friday at $210.28.
Tech analyst Ed Zitron disagreed, calling it “such a crazy thing to do on so many levels and almost as big, especially considering it’s being built by SoftBank.” Both were responding to the same structural feature: a chip supplier underwriting its customer’s ability to continue buying chips.
Nvidia has already committed more than $40 billion to AI equity positions, including nearly $30 billion to OpenAI. Its investment in Ilya Sutskever’s Safe Superintelligence is arguably the purest version of the pattern, where Nvidia supports companies that have invested in Nvidia hardware.
A structure with an antecedent
Vendor guarantees on data center obligations are no longer common, although the scale is here. Google has secured about $44 billion in data center rent for other companies, a figure Nvidia’s proposed backstop would exceed by nearly six times.
OpenAI’s own currency structure is growing exponentially. Eight banks are now backing SoftBank’s record $40 billion loan to OpenAI, and SoftBank has separately sought secured margin lending for its OpenAI stake.
Much of this responsibility is outside of normal credit disclosures. Research has put Big Tech’s off-balance-sheet AI liabilities at about $1.65 trillion, and Meta alone is worth about $420 billion, nearly three times its reported debt.
What OpenAI stands for
For OpenAI, the deal will be the first step toward owning its own computing instead of leasing it from Microsoft, Amazon, and Oracle. Controlling infrastructure is the difference between negotiating and dictating power.
For Nvidia, calculating demand certainty. Ensuring building locks in the age of chip orders from customers would be difficult to finance, which is why critics find the arrangement uncomfortable.
A question of scale
AI infrastructure spending is expected to exceed $700 billion this year, and the industry is expected to spend more than $3 billion by 2028 on data centers, much of it funded by the chips inside them. The central risk is not that AI fails to matter, but that buildout outpaces demand or that returns outpace debt amortises.
Nothing is signed. The Journal noted discussions about the site and its financing are ongoing and subject to change, which with a $250 billion guarantee is a big caveat.



