Nvidia is reportedly discussing a vendor financing deal worth up to $250 billion: a guarantee that would let OpenAI lease a $500 billion, 10-gigawatt data center being built in Ohio. CNBC confirmed the talks on July 27, 2026 ET, with a separate $350 billion in discussion to help OpenAI buy Nvidia’s own chips.
Vendor financing means the seller lends the buyer money to buy its own product — and that money loops back around as the seller’s revenue. Here’s why: OpenAI is still unprofitable and privately held, so it doesn’t have an investment-grade credit rating — the score bond markets use to decide whether a company is safe enough for long-term loans. Without one, the banks funding this data center’s construction won’t lend on OpenAI’s word alone. So they lend on Nvidia’s instead. That’s what a backstop (a lender-of-last-resort promise, not cash paid up front) buys: a lower interest rate, because the bank is really underwriting Nvidia’s balance sheet.
This isn’t new. In the 19th century, British steelmakers took U.S. railroad bonds instead of cash for rail shipments — betting the railroads they helped build would eventually pay them back. In 1905, GE spun up a financing arm just to lend money to power utilities so they’d buy GE turbines. In the early 2000s, telecom equipment maker Lucent lent carriers roughly $8 billion to buy its gear — and when demand for bandwidth stalled, the carriers stopped paying, the orders dried up, and Lucent went down with them. [As of July 2026], a similar loop is forming between Nvidia, OpenAI, and Oracle: Nvidia backs OpenAI, OpenAI leases from Oracle, Oracle buys Nvidia chips to fill the space.
None of this makes the deal fraudulent — every link in the chain is disclosed. The risk is that the whole structure depends on one thing staying true the entire time: that AI demand keeps growing fast enough to justify the debt before anyone has to actually collect on the guarantee.
My take: Around and around we go — the guarantor is also the buyer.
Not advice. Just how I see it.