Nvidia Weighs $250 Billion Backstop for OpenAI's 10-Gigawatt Ohio Data Center

The Wall Street Journal reported Sunday evening that Nvidia is in advanced discussions to guarantee approximately $250 billion in debt financing for OpenAI’s planned AI data center campus in Piketon, Ohio — a former uranium enrichment facility being repurposed by SoftBank’s energy subsidiary, SB Energy. Separate talks are simultaneously underway covering up to $350 billion in Nvidia chip purchases that would populate the campus once it’s built. If the full arrangement closes, it would represent the largest single data-center financing commitment in history by a margin so wide as to be almost incomprehensible.

Reuters said it could not independently verify the Wall Street Journal’s reporting by deadline. No final agreement has been disclosed by either Nvidia or OpenAI. The news drove fresh discussion about the increasingly circular nature of AI infrastructure financing — but it also crystallized a strategic reality that both companies have been quietly building toward for over a year.

A Former Uranium Plant Becomes AI’s Largest Campus

The site itself is notable. Piketon, Ohio’s Portsmouth Gaseous Diffusion Plant processed enriched uranium for nuclear weapons and power generation from 1954 until 2001. The Department of Energy has been decontaminating the land for decades. SoftBank’s SB Energy subsidiary acquired rights to the site largely because it comes pre-equipped with massive electrical infrastructure — a factor worth hundreds of millions in saved permitting and construction time. The planned campus would operate at 10 gigawatts of continuous power, compared to the roughly 100–200 megawatts that a typical hyperscale data center consumes today.

Total development cost for the campus, including both construction and the AI chips needed to fill it, is estimated at over $500 billion, making it roughly equivalent to the GDP of Denmark. The 10-gigawatt power figure alone would make it a meaningful fraction of Ohio’s total electricity generating capacity.

Why Nvidia Would Do This

The logic for Nvidia is relatively straightforward, even if the scale strains belief. OpenAI lacks an investment-grade credit rating, which means it cannot easily borrow money at the rates required to finance infrastructure of this scale. Nvidia’s balance sheet is in a fundamentally different position: the company generated over $80 billion in revenue in fiscal year 2025, carries substantial cash reserves, and has an AAA-adjacent credit profile from the debt markets’ perspective.

By backstopping OpenAI’s lease and construction debt, Nvidia effectively becomes a creditworthy guarantor for a customer it needs to keep buying chips. The arrangement is structurally analogous to a car manufacturer financing loans for customers who couldn’t otherwise afford their vehicles — except the “car” in this case costs half a trillion dollars. In exchange, Nvidia essentially guarantees future demand for its own products for the better part of a decade, given the time horizon on construction and equipment lifecycles.

The separate chip financing discussions — potentially another $350 billion — would compound this logic. Nvidia chips placed in OpenAI’s own-operated campus are chips that can’t be redirected to a competitor. The deal would also insulate Nvidia from the risk that OpenAI becomes dependent on Microsoft, Amazon, or Oracle data centers that might over time begin procuring chips from AMD, Google (TPUs), or Nvidia’s other competitors.

OpenAI’s Infrastructure Independence Play

For OpenAI, the strategic dimension is equally clear. The company has operated almost entirely on rented compute — primarily from Microsoft Azure under the terms of a multibillion-dollar partnership that predates the current AI boom. While that arrangement gave OpenAI access to scale it couldn’t have built alone, it also created a structural dependency: Microsoft controls the infrastructure, sits on OpenAI’s board as an observer, and negotiates the terms of OpenAI’s most critical operational input.

A 10-gigawatt proprietary campus would change that. OpenAI would, for the first time, own the servers its models train and run on, giving it the ability to optimize infrastructure decisions independently of a cloud partner’s commercial interests. It would also give OpenAI the ability to offer different SLA structures to enterprise customers — guaranteed uptime, custom hardware configurations, latency profiles — that are harder to deliver as a tenant of someone else’s facility.

The timing matters: OpenAI’s partnership with Microsoft has been renegotiated multiple times since 2020, and the terms of future renewals are widely expected to be contested. A company that controls its own compute has a fundamentally different negotiating posture than one that doesn’t.

Circular Financing and the AI Liquidity Machine

The arrangement has attracted sharp commentary from financial analysts. Michael Burry, the investor who famously shorted mortgage-backed securities before the 2008 financial crisis and is known for his skeptical takes, posted on social media: “Around and around we go” — a pointed reference to the circular flow of capital in which chip suppliers finance the customers who buy their chips, who then generate revenue to repay the chip suppliers.

The circularity is real. Nvidia’s revenue comes largely from selling to a concentrated set of hyperscalers and AI labs. Those customers then raise capital (from SoftBank, Microsoft, venture investors, and now potentially Nvidia itself) to fund chip purchases. The health of Nvidia’s business is therefore intertwined with the continued availability of financing to its customers — a concentration risk that analysts have flagged with increasing frequency as the AI capex cycle has extended into its third consecutive year of acceleration.

At the same time, circular doesn’t necessarily mean unstable. If the compute being purchased is generating genuine economic value — which enterprise AI deployments increasingly appear to be doing — then the financing cycle is less bubble-like and more analogous to long-term infrastructure bonds backed by operating revenue.

What Happens Next

The Piketon campus would not be operational for several years — utility-scale data center construction at 10 gigawatts is a multi-year undertaking even with favorable permitting. SoftBank’s SB Energy subsidiary has the site and the power rights; Nvidia and OpenAI are apparently negotiating the financial structure under which construction would begin.

Whether the deal closes at the reported scale, at a reduced scale, or with modified terms remains to be seen. But the direction is unmistakable: the AI infrastructure industry is entering a phase where the sums involved have moved from hyperscale (hundreds of millions to single-digit billions) to something that doesn’t yet have a widely accepted name. Half a trillion dollars in a single facility represents a new category of capital commitment — one that only a handful of entities on earth can contemplate backing.

Both Nvidia and OpenAI declined to comment on the WSJ report.