Nvidia is negotiating $600 billion in financing tied to OpenAI infrastructure—a planned Ohio data center and GPU purchases that would dwarf any prior semiconductor supply arrangement. The structure has not closed. The announcement arrived in pre-market coverage with no term sheet disclosed and no named co-lenders, only the reported involvement of Nvidia's own balance sheet. The deal would exceed the $500 billion Stargate commitment announced in January and make Nvidia the largest creditor to the model developer it supplies.
The financing covers two assets: a greenfield data center in Ohio and a multi-year GPU purchase agreement. OpenAI would draw against the facility to buy Nvidia chips and fund construction, with repayment structured around future inference revenue. Nvidia disclosed no haircut, no equity kicker, and no independent valuation. The arrangement resembles vendor financing in enterprise software—except the borrower has no positive operating cash flow, the lender sells the capital goods, and the collateral is a not-yet-built facility in a market where utilization rates remain unproven outside hyperscaler environments. The $600 billion figure is eight times OpenAI's last private valuation and roughly equal to Nvidia's trailing twelve-month revenue.
Critics inside the capital-markets complex are calling it circular financing: Nvidia lends money so OpenAI can buy Nvidia chips, inflating both Nvidia's revenue and its loan book. The arrangement creates pro-forma demand. It also creates pro-forma credit exposure, because if OpenAI cannot service the debt from inference margins, Nvidia becomes both an unpaid supplier and a distressed lender. The structure is not illegal. It is optically difficult. In 2002, Cisco wrote off $2.25 billion in vendor financing to dot-com customers who never generated the cash flow to pay. In 2023, Tesla wound down its Solar Loan program after similar scrutiny. The difference here is scale and the fact that Nvidia is still trading at 37 times forward earnings, a multiple that assumes both revenue growth and capital discipline.
Allocators should watch three near-term events. First, whether Nvidia discloses the financing as a separate line item in the April 10-K or buries it in "other long-term assets." Second, whether any bank or sovereign fund joins as co-lender, which would validate the structure or at least distribute the exposure. Third, whether OpenAI's inference revenue—currently under $4 billion annualized—can support debt service on a $600 billion facility at any realistic rate. If the implied interest expense exceeds $30 billion annually and OpenAI's gross margin on inference is 60 percent, the company needs $50 billion in revenue just to cover the coupon. OpenAI has not published a path to that number. Nvidia has not published a credit memo.
The deal is not done. The optics are not good. The financing may still restructure or shrink before close. What remains is a $600 billion question: whether Nvidia is financing the AI buildout or financing its own order book.