PIKE COUNTY, Ohio — The site sits on the former Portsmouth Gaseous Diffusion Plant, a Cold War relic whose enrichment buildings have stood empty for decades. By the end of the decade, it is meant to house one of the largest AI data centers on earth: 8 gigawatts of computing capacity, leased by OpenAI for 20 years, built by SoftBank’s SB Energy, and underwritten in part by Nvidia, which has agreed to provide up to $105 billion in credit support.
The structure, disclosed in a securities filing on Aug. 17, is the most elaborate example yet of the financial engineering that has come to define the AI buildout. Nvidia will be the exclusive AI compute provider at the site, supplying GPUs, CPUs and networking through its DSX platform. The guarantees cover land, power and building costs for an initial 4.25 gigawatts of capacity, with an option to extend support to the remaining 3.75 gigawatts that would bring the campus to 8 gigawatts total.
The number that matters is smaller than the one originally floated. CNBC reported earlier this summer that Nvidia had discussed providing a backstop of up to $250 billion for OpenAI’s data-center ambitions; the Wall Street Journal then reported that investor concerns about the chipmaker’s exposure had pushed the figure down to below $120 billion. The final $105 billion cap, tied to the initial phase, reflects that pressure — a guarantee cut by more than half from the first proposal.
The mechanics matter more than the headline. Nvidia is not handing over $105 billion in cash; it is guaranteeing the residual value of leased infrastructure. If OpenAI defaults or the leases collapse, Nvidia would cover the gap between the guaranteed minimum value of the facilities and what SB Energy recovers by re-leasing or selling them. OpenAI has agreed to reimburse and indemnify Nvidia for amounts the chipmaker actually pays, a structure that puts the ultimate credit risk back on OpenAI itself.
Nvidia is also investing $1.5 billion directly in SB Energy, joining SoftBank and OpenAI among the developer’s shareholders. SB Energy, in turn, is building at least 10 gigawatts of new power generation to support the campus, with $4.2 billion in regional grid upgrades developed in partnership with AEP Ohio, the local utility. The project sits on federal land and is being developed with the cooperation of the U.S. Department of Energy and the Department of Commerce.
The deal gives each party something it badly wanted. OpenAI gets a giant, financed campus without putting the full bill on its own balance sheet. Nvidia locks in a customer for its most advanced hardware — capacity expected to come online in phases starting in 2028 — while keeping its cash commitment modest. SB Energy gets a signature project that validates its pivot from solar development to AI infrastructure. Ohio gets a data center that officials say will create tens of thousands of construction jobs and an $80 million community benefits fund, with OpenAI adding another $40 million.
The arrangement also illustrates the circularity that has made AI investors uneasy. Nvidia finances infrastructure that OpenAI leases; OpenAI uses that capacity to run models that, in theory, generate the revenue to pay the lease; Nvidia sells the chips that make the whole loop possible. When one leg weakens — if OpenAI’s growth slows, if model margins compress, if demand for the capacity does not materialize — the circle stops working, and the guarantees would be tested.
Wall Street has been asking exactly those questions. The original $250 billion proposal drew pointed scrutiny from investors worried that Nvidia, already valued as the centerpiece of the AI trade, was taking on financial risk that belonged on banks’ books. The trimmed guarantee, analysts said, is Nvidia’s answer: it keeps the strategic upside of backing OpenAI while capping the downside at a level the balance sheet can absorb.
The guarantees expire under a set of conditions spelled out in the filing — when the 20-year lease ends, when OpenAI achieves a strong credit rating, or when standard termination events occur. Nvidia’s payment obligations begin only when the facilities are ready for service, expected in 2028, meaning the chipmaker has years before any of its support could be called.
For the AI industry, the Ohio campus is a template. The scale of capital required for AI data centers has outpaced what even the largest technology companies want to carry alone, pushing deals toward layered financing: developers, tenants, equipment vendors and utilities each holding a piece of the risk. Nvidia’s $105 billion guarantee is the clearest example yet of a chipmaker acting as banker, landlord and supplier at once.
The risks will only become visible when the machines are running. Until then, the former uranium-enrichment plant in Pike County has become the most watched construction site in the AI economy — a bet, financed in layers, that the computing needs of artificial intelligence will justify the largest infrastructure guarantee in corporate history.
