Nvidia is in talks to provide roughly $250 billion in financing guarantees that would allow OpenAI to lease a 10-gigawatt data center campus under development in southern Ohio, The Wall Street Journal reported on 26 July 2026. Including the chips that would fill the buildings, the project could cost more than $500 billion — making it the largest data center development ever announced.
The guarantee would cover the lease itself and the debt raised to build the site, but not the Nvidia hardware inside it. Separately, the two companies are discussing financing for as much as $350 billion of OpenAI chip purchases. People familiar with the talks told the Journal that negotiations are at an early stage, terms are not final, and the arrangement could still collapse.
Key Highlights
- Nvidia would guarantee approximately $250 billion of lease and construction financing for OpenAI
- A separate discussion covers financing up to $350 billion in Nvidia chip purchases
- The campus targets 10 gigawatts of demand, with a first phase of roughly 800 megawatts in 2028
- The site is a decommissioned uranium-enrichment facility on federal land about 50 miles south of Columbus, Ohio
- Developer is SB Energy, SoftBank's energy arm, which would also operate the power plant
- Total project cost, chips included, could exceed $500 billion
Why a Guarantee Instead of a Loan
The structure exists because of a credit mismatch. OpenAI is a private, unprofitable company without an investment-grade rating, and borrowing hundreds of billions of dollars at that profile is punishingly expensive. Nvidia, by contrast, carries one of the strongest balance sheets in the market.
By wrapping OpenAI's leasing and construction obligations in a Nvidia guarantee, lenders would price the debt against Nvidia's creditworthiness rather than OpenAI's income statement. The financing cost falls sharply, and the project becomes bankable. Nvidia has already invested roughly $30 billion in OpenAI, so the guarantee extends an existing relationship rather than starting a new one.
The Circular Financing Question
The arrangement has drawn immediate scrutiny because it places the chip supplier on the hook for its largest customer's debt. Investor Michael Burry summarised the criticism bluntly, remarking "around and around we go" — a reference to capital circulating among a small group of AI players rather than entering from outside the sector.
The mechanics are straightforward enough to state plainly: Nvidia invests in OpenAI, OpenAI builds infrastructure running Nvidia chips, Nvidia guarantees the debt funding that infrastructure, and the infrastructure locks in years of chip orders. If OpenAI's revenue grows into its commitments, the loop is a durable demand moat that no competitor can match — AMD cannot cosign $250 billion of anyone's debt. If revenue disappoints, the risk concentrates rather than disperses.
A Federal Site, Government-Controlled Power
The location is unusual. The campus would sit on a decommissioned uranium-enrichment site on federal land, chosen in part because existing power infrastructure and permits reduce the friction that has stalled other large builds. Electricity for the project is controlled by the U.S. government.
The generation stack traces back to the 2025 U.S.–Japan Strategic Trade and Investment Agreement, under which Japan committed roughly $33 billion to fund a natural gas power facility in exchange for lower tariffs. Proceeds from power sales would be shared between the two governments until Japan recovers its investment, after which the U.S. share rises to 90 percent. SB Energy would operate the plant under a fee arrangement.
Commerce Secretary Howard Lutnick is reported to be helping determine which company receives access to the electricity. OpenAI, Anthropic, Microsoft and Google are all said to be in the queue.
Impact
For OpenAI, a deal would mark the first serious step toward owning its compute rather than renting it from Microsoft, Amazon and Oracle. That shifts the balance of power in cloud negotiations and reduces a dependency that has shaped the company's economics since 2019.
For the wider market, the story is about how frontier AI capacity now gets financed. Equity rounds and revenue no longer cover buildouts at this scale; supplier guarantees and debt structures do. Teams building on top of AI infrastructure should read the signal accordingly — secure capacity commitments rather than assuming availability, and keep workloads portable across model providers.
Background
The Ohio project sits alongside a run of enormous 2026 infrastructure commitments, including Oracle's $40 billion order for 400,000 Nvidia GB200 chips for an OpenAI hub in Texas and OpenAI's own projected compute spending of roughly $750 billion through 2030. Power, not silicon, has become the binding constraint, which is why a site with pre-existing federal generation capacity is worth more than its acreage suggests.
What's Next
Nothing is signed. The Journal's sources stressed that terms remain unsettled and the deal could fail. Confirmation — or denial — from Nvidia and OpenAI is the near-term thing to watch, along with how the Commerce Department allocates the site's power among competing bidders.
Source: Bloomberg