Nvidia is in talks to put roughly $250 billion in financing guarantees behind OpenAI, a commitment that would anchor a 10-gigawatt data center campus in southern Ohio and hand the ChatGPT maker its first real path off Microsoft’s cloud. The Wall Street Journal reported the discussions Sunday, citing people familiar with the matter. For Nvidia, the backstop locks in demand for its chips at a scale that makes current order books look like a rounding error; for OpenAI, it is the down payment on infrastructure independence.
The project, developed by SoftBank’s energy subsidiary, is expected to cost more than $500 billion all in, including the silicon inside. Nvidia’s $250 billion guarantee covers the lease and debt financing for the campus itself but explicitly excludes the chips, those are the subject of a separate financing discussion worth up to $350 billion. The first phase, about 800 megawatts, is slated for 2028. Power allocation sits with the US government, funded in part by a Japanese pledge to invest $33 billion in a natural gas plant under a recent trade deal, with Commerce Secretary Howard Lutnick refereeing access.
The numbers are large enough to distort the gravity around them. AI infrastructure spending across the industry is on track to top $700 billion this year, and this single campus would absorb a material slice of that. OpenAI has been in advanced talks for weeks and is described as showing the strongest interest, though Anthropic, Microsoft and Google have also knocked on Lutnick’s door. The structure, a chipmaker guaranteeing the real estate so its customer can buy more chips, is circular in a way that would make a SPAC sponsor blush.
What makes this more than a headline is the signal: the hyperscaler model is cracking. OpenAI does not want to rent from Microsoft, Amazon and Oracle forever, and the capital markets are apparently willing to underwrite the exit. But the deal still hinges on debt investors buying into a 10-gigawatt bet on inference demand that has not yet materialized at anything like this scale. The chips are not guaranteed. The power is political. The timeline stretches to 2028.
Watch whether the chip financing closes on terms that look like equity risk disguised as debt. Watch whether Lutnick’s allocation process favors incumbents or the new entrant. And watch the $700 billion industry spend figure, if this project moves forward, the denominator just got a lot bigger.