Nvidia Negotiates $250 Billion in Backing for OpenAI to Lease 10 GW Campus in Ohio
Nvidia is discussing securing $250 billion in funding for OpenAI to lease a 10 GW campus in Piketon, as revealed by the Wall Street Journal, in an arrangement that exposes the circular financing of AI.
According to the Wall Street Journal, Nvidia is in negotiations to secure approximately $250 billion in debt related to the lease that OpenAI plans to sign for a 10-gigawatt campus in Piketon, southern Ohio, developed by a energy subsidiary of SoftBank on the former uranium enrichment site of Portsmouth. If the deal is finalized under the reported terms, the chip manufacturer will fund both the hardware and the real estate that will house this hardware.
What’s on the Table
The backing positions Nvidia as the last-resort payer. If OpenAI delays or fails to honor the lease and the construction debt, it is Nvidia that covers the costs. The stated reason is economic: without operational profit and being private, OpenAI lacks investment grade and would be unable to raise the mountain of debt that the project requires on its own. With Nvidia backing it, the structure can now finance itself under terms close to a AAA issuance.
The $250 billion covers both the lease and the construction debt. In addition, the WSJ reports a second line in discussion, also with Nvidia, to finance up to $350 billion in chips that will occupy the campus. When combined with land, infrastructure, energy, and silicon, the project could exceed $500 billion, making Piketon the largest data center project ever announced. The first phase, of 800 megawatts, is expected to be completed by 2028.
The Link to the Japan-U.S. Package
The power grid that will supply the campus is supported by another front of the same axis. Approximately $33 billion in Japanese capital, facilitated by the trade agreement announced earlier this month between Washington and Tokyo, underpins the generation and transmission that Piketon will consume. SoftBank on the construction side, pension funds and Japanese utilities on the generation side, Nvidia backing the debt, and OpenAI as the final tenant. Each link depends on the others for the calculations to align, and the White House relies on the whole to justify the political promise of industrial rebirth in a city that lost nuclear enrichment in the 1990s.
The Signal for the Corporate Buyer
For the CIO signing a multiyear contract with OpenAI, the interpretation matters. A significant portion of the price paid per token is directly and increasingly indirectly tied to financial agreements that have not yet closed. Wall Street credit analysts are already classifying the structure as circular financing: Nvidia lends so that OpenAI can lease a property that Nvidia supplies with chips that OpenAI pays for with revenue that Nvidia helps generate. If one link stalls, the price of compute and the stability of the supply guiding today’s corporate roadmaps also stall.
This pattern repeats outside the U.S. In Japan, SoftBank has already signaled accumulated capex of over $40 billion in training infrastructure by 2028, and the Softbank Vision Fund remains exposed to its own balance sheet with OpenAI. In India, Reliance Jio and Adani announced partnerships in recent months with American hyperscalers anchored on identical premises of a continuous decrease in token costs. In Brazil, the pipeline of multiyear contracts between large banks, telecoms, and providers such as AWS and Microsoft embeds the same premise that the unit cost of compute will drop 20% per year, a thesis that the Ohio structure helps to support only if it materializes.
The Risk That No One Wants to Price
There is no final decision yet. The WSJ is clear: the terms are still being drawn up, and the conclusion is not guaranteed. Nvidia has not commented on the figures, OpenAI declined to comment, and SoftBank referred to the project’s official communication. The backing, if materialized, requires approval from Nvidia's directors and, very likely, disclosure to the SEC due to its scale. It also opens the window for antitrust scrutiny in three jurisdictions: the FTC in the U.S., the European Commission regarding the effect on the European cloud market, and the Japanese regulator concerning SoftBank's exposure.
The discussion reveals something larger than the absolute value. The scale of compute for frontier AI has surpassed what the conventional corporate debt market can finance on its own. Capital markets requested guarantees, and Nvidia responded that it would provide them. If the precedent is accepted, the next 10 GW campus will request exactly the same, and the value of the assumed risk will continue to rise.