Nvidia Will Co-Sign OpenAI's $500 Billion Data Center
OpenAI is negotiating a 10-gigawatt campus on federal land in Ohio — bigger than all seven existing Stargate sites combined — with Nvidia acting as financial guarantor of the lease. The circle of who funds, supplies, and backs whom keeps tightening.
The number that should stop you is not 10 gigawatts, though that's the one that will travel. It's the structure underneath it. According to reporting from The Information, OpenAI is in advanced talks to lease a single data-center campus in southern Ohio that could cost at least $500 billion to build — and Nvidia is expected to do three things at once: supply the chips, take an equity stake in OpenAI as capacity comes online, and act as the financial guarantor of OpenAI's lease. Read that sequence twice. The company selling the hardware is also backing the loan that pays for the building that houses the hardware it sells.
What's being built
The site is the former Portsmouth Gaseous Diffusion Plant in Pike County, Ohio — Department of Energy land, a Cold War uranium-enrichment relic now slated to host the largest AI training facility ever contemplated. The campus would be developed by SB Energy, a SoftBank unit, with OpenAI controlling the equipment under a 20-year lease and payments starting only once operations begin. The first phase is targeted for 2028; the full 10-gigawatt buildout is expected to take at least a decade.
To put 10 gigawatts in proportion: this one campus would exceed the combined capacity currently planned across all seven existing Stargate sites. It's not an addition to the program — it's a doubling of the whole thing in a single location. The first 800-megawatt phase still has to clear environmental and regulatory review before anyone pours concrete, but the intent is unambiguous. OpenAI is trying to build compute the way nations build power grids.
The circular financing question
Here is the part that has the markets uneasy. In September 2025, Nvidia and OpenAI signed a letter of intent for at least 10 gigawatts of Nvidia systems, with Nvidia committing up to $100 billion into OpenAI as that capacity comes online. The Ohio deal threads through the same loop: Nvidia guarantees OpenAI's lease and helps backstop SB Energy's financing, OpenAI uses the facility to buy and run Nvidia chips, and the revenue Nvidia books from those chips helps justify the investment that made the purchase possible.
Each leg of that triangle is defensible on its own. A chip vendor offering financing to anchor demand is ordinary. A hyperscaler taking a 20-year lease to lock in capacity is ordinary. A sovereign-scale developer like SoftBank's SB Energy carrying the construction risk is ordinary. The discomfort is in the geometry: when the supplier, the customer, and the guarantor are this tightly coupled, the same dollar can appear as revenue, as investment, and as collateral depending on where you stand in the ring. Critics call it circular. Defenders call it vertical integration of a capital-intensive supply chain. Both are describing the same diagram.
What makes it more than an accounting curiosity is scale. Morgan Stanley expects AI-related global debt issuance to roughly double to $570 billion in 2026. A meaningful share of the AI build-out is now being financed against future compute demand that has to materialize on schedule for the structure to hold. If demand keeps compounding, the circularity is just efficient capital formation. If it stalls, the same tight coupling that looks efficient becomes the transmission mechanism for stress — every leg of the triangle exposed to the others.
Why federal land changes the politics
Building on DOE land at a decommissioned nuclear site is not an accident of real estate. It's a signal that AI infrastructure has become a matter of industrial policy, not just private capex. Federal land comes with federal interest — in the jobs, in the strategic compute, and increasingly in the framing of data centers as national assets in a capability race with China. The same week, the UK used London Tech Week to announce a £400 million chip-buying program and a national compute strategy; the US answer is to hand a frontier lab a Cold War enrichment site and let it build the biggest training cluster on Earth.
That framing cuts both ways for OpenAI. Government land can mean streamlined siting and political cover. It also means the project's timeline is now hostage to environmental review, local power and water constraints, and the durability of whatever administration's priorities put it there. A 10-gigawatt campus needs the energy of a mid-sized city; the grid and cooling questions are not footnotes.
The real story
For two years the AI narrative has been about models — who has the smartest one, who shipped first. Ohio is a reminder that the binding constraint has quietly moved underneath the models, to power, land, chips, and the financing that ties them together. OpenAI isn't competing on cleverness here. It's competing on its ability to assemble half a trillion dollars of physical infrastructure and the capital structure to pay for it.
The bet embedded in this deal is enormous and specific: that demand for frontier compute in 2028 and beyond will be large enough, and durable enough, to justify the single most expensive private construction project most people have ever heard of — underwritten by the company that profits most if the bet pays off. If it works, it's the moment AI infrastructure graduated from corporate spending to nation-scale buildout. If it doesn't, the closed loop everyone is squinting at right now will be the first thing the post-mortem points to.