Nvidia is becoming OpenAI's bank: the $250B Ohio backstop, decoded

This morning, Type0 noted that Nvidia's $250 billion backstop for OpenAI's Ohio build looked less like a chip-order win and more like a warning about who absorbs the credit risk in an AI infrastructure boom. Since then, the vendor-credit picture has only sharpened.

The Wall Street Journal reported Monday that Nvidia is in talks to backstop roughly $250 billion of OpenAI's lease and construction debt for a planned 10-gigawatt data center in southern Ohio — a guarantee structured to help OpenAI secure construction financing despite its below-investment-grade credit profile. The Ohio campus is being developed by SoftBank's energy subsidiary, with a first phase targeted at 800 megawatts of capacity and currently scheduled to begin operating in 2028.

The $250 billion figure is a debt guarantee, not a cash outlay by Nvidia and not a purchase order. It would sit behind OpenAI's lease and construction financing, addressing lender concerns that banks typically resolve by demanding a stronger-balance-sheet tenant or much higher rent before underwriting a 10-gigawatt build. Without a backstop, that math does not work.

Nvidia is also discussing a separate financing package that could total about $350 billion to support OpenAI's purchases of Nvidia AI chips, the WSJ reported. The two lines are distinct: one is the bricks-and-mortar loan guarantee, the other is the chip-purchase financing. The total project, including the Nvidia chips expected to power the facilities, could ultimately cost more than $500 billion.

The structure inverts the usual AI capex order. Normally, a hyperscaler finances the buildings, fills them with rented or owned chips, and signs a long-dated offtake contract with a model provider. Here, the model provider's chip supplier is guaranteeing the buildings so the model provider can lease them, while also separately financing the chips that go inside. Nvidia is becoming a credit institution for its own customer.

Cisco lent money to customers to keep orders flowing during the 2001 telecom bust and took the write-downs. Boeing and Airbus routinely provide manufacturer-financed support to airlines. GE Capital once carried enough commercial-paper exposure to threaten the parent's credit rating during the financial crisis. The difference in this case is the scale relative to the underwriter's market cap and the concentration of risk in a single counterparty. A $250 billion backstop on Nvidia's balance sheet, even as a contingent guarantee, is a different category of vendor financing than a fleet-financing facility for narrowbody jets.

The 9.2-gigawatt gap between the first phase and the headline 10-gigawatt figure tells the reader how much of this build remains unfunded and unbuilt, and how much of the $250 billion backstop is forward exposure rather than a current commitment.

Nvidia shares traded down about 2% at $202 on Monday morning, roughly 8% year-to-date. The market reaction was muted relative to the size of the commitment being discussed. If Nvidia is now the bank for the AI capex cycle, the next leg of the cycle is no longer priced only on chip volumes, but on the credit standing of the largest model provider and the willingness of its supplier to absorb that credit.

The WSJ reporting is explicit that the negotiations are ongoing, the terms are not finalized, and there is no assurance a deal will close. The next signal is whether the structure shows up in Nvidia's next 10-Q as a contingent guarantee, an off-balance-sheet commitment, or simply as a forward-looking sentence in the risk factors.