On paper, it is the largest single AI infrastructure bet ever assembled: a 10-gigawatt data center campus on a former Cold War uranium-enrichment site in Piketon, Ohio, with a price tag north of $500 billion. And according to The Wall Street Journal, the load-bearing wall of the entire financing stack may be a single company that historically just sold the chips: Nvidia.
Reports over the weekend put Nvidia in active negotiations to guarantee roughly $250 billion in financing so OpenAI can secure a long-term lease and construction commitments for the site, developed by SoftBank's energy subsidiary, SB Energy, on federal land. A separate, reportedly parallel track could add another $350 billion to fund OpenAI's chip purchases at the same campus.
Anatomy of the Deal: The Financial Engineering
Strip away the superlatives and the structure is a textbook credit enhancement play. Nvidia would not be lending the money outright; it would be backstopping it — putting its balance sheet, one of the strongest in technology history, underneath lease and construction obligations that conventional lenders have shown little appetite to underwrite on their own.
The signal this sends to the debt markets is unambiguous. A $500 billion project with a single anchor tenant, on land owned by the U.S. Department of Energy, still needed a guarantor with a market cap the size of a small country's GDP. That is not a vote of confidence in the project's standalone credit profile; it is a vote of confidence in Nvidia's willingness to absorb the downside.
This is not Nvidia's first rodeo in this exact formation. Jensen Huang has run variations of the playbook with neo-cloud operators like CoreWeave and Nebius, and with Naver in South Korea — using vendor-financing and guarantees to ensure that whatever infrastructure gets built runs on Nvidia silicon. The Ohio deal is simply the same mechanism scaled by roughly an order of magnitude, layered on top of the $30 billion equity stake Nvidia took in OpenAI in February 2026.
Break the financing stack into its constituent layers:
- The lease backstop (~$250B): Nvidia guarantees OpenAI's long-term lease and the data center build-out, letting SB Energy raise additional debt from lenders who would otherwise balk.
- The silicon financing (~$350B): A separate, still-unconfirmed package to fund OpenAI's GPU purchases — effectively monetizing Nvidia's own hardware roadmap.
- The sovereign layer ($33B): Japan has agreed to contribute capital tied to tariff relief, with the U.S. and Japan splitting power revenue until Japan recoups its stake, after which the U.S. keeps 90% of the remainder.
Commerce Secretary Howard Lutnick is reportedly overseeing power allocation at the federal site, with Anthropic, Google, and Microsoft all expressing interest in the region's energy envelope. Washington's preference, according to multiple reports, is that most of the 10 gigawatts ultimately flows to OpenAI.
Ten Gigawatts in the Grid Calculus
To understand why this project is a step-change rather than an increment, run the arithmetic. Ten gigawatts is enough to power somewhere between 7.5 and 10 million homes. Industry estimates put global active data center capacity at roughly 67.7 GW, which means a single campus in southern Ohio would represent about 14.8% of the planet's total data center load — for one company, on one site.
The site's history matters for the engineering story. The federal government seized the Pike County farmland by eminent domain in 1952, with the Army Corps of Engineers filing condemnation on every tract the same day the site was announced, to build a gaseous diffusion plant for uranium enrichment. That infrastructure heritage — massive electrical distribution, security perimeters, and a workforce accustomed to high-consequence industrial operations — is precisely the profile hyperscale AI developers now covet.
The power architecture is equally consequential. This is not a facility plugging into a strained regional grid and hoping. The project pairs with new gas generation, with AEP involved in grid upgrades, and with a dedicated federal power allocation. For OpenAI, owning a 10-gigawatt site is a strategic shift away from renting capacity from hyperscalers like Amazon, Microsoft, and Oracle toward controlling the physical substrate of its own compute.
The market's initial verdict, however, was skepticism. NVDA shares fell roughly 4.45% the following Monday as investors digested the prospect of a $250 billion contingent liability, however well-structured. Wall Street's reaction was a reminder that even a balance sheet as fortress-like as Nvidia's has a price when it becomes the lender of last resort for the AI boom.
For engineers and architects watching the industry, the takeaway is structural. The era of buying GPUs and plugging them into leased racks is giving way to an era of vertically integrated AI factories — federal land contracts, dedicated gas turbines, sovereign capital, and silicon vendors doubling as financial guarantors. The Piketon campus, if it clears the remaining negotiations, will be the template that every subsequent megaproject is measured against.
Whether the music stops before the building finishes is the question nobody can answer yet. But the fact that Nvidia is willing to stand behind the orchestra tells you everything about how high the stakes have become.
Comments