Nvidia Is Financing $105 Billion of an OpenAI Data Center It Will Also Supply
A securities filing disclosed on Monday that Nvidia will provide up to one hundred and five billion dollars in financing for an OpenAI data center in Ohio. The facility starts at 4.25 gigawatts of compute with an option on a further 3.75. Nvidia supplies the hardware. SB Energy builds and operates the site at the PORTS-Pike campus in Pike County under a twenty-year lease to OpenAI. Capacity phases in from 2028.
Read the structure rather than the headline number. A chip vendor is lending its customer the money to buy its chips, on a scale that dwarfs the customer’s revenue, for delivery two years out.
This is vendor financing. It is a well-understood arrangement with a well-understood history, and it is not inherently improper. It accelerates buildouts that would otherwise wait on the capital markets, it locks in a platform standard, and when demand materializes as forecast everyone involved does very well. Telecom equipment makers used it extensively at the end of the 1990s for exactly these reasons.
The reason it warrants attention is what it does to the revenue signal. When a supplier funds its customer’s purchases, the resulting sales are partly a reflection of the supplier’s willingness to extend credit rather than an independent measurement of end demand. Investors reading chip revenue as evidence of AI adoption are reading a number that now contains a financing decision inside it. That is not fraud and nobody is hiding it, the filing is how we know. It just means the number answers a narrower question than it appears to.
The scale is the other thing. This single arrangement is larger than last September’s letter of intent between the same two companies, which committed up to a hundred billion across ten gigawatts. Add the hyperscaler capital budgets, which run to roughly seven hundred billion this year, and the aggregate is large enough to show up in the bond market as competition for capital, which is part of why thirty-year Treasury yields hit a nineteen-year high last week.
Two things to watch, both of which resolve well before 2028. Whether the power actually arrives, since grid interconnection queues are the binding constraint across the industry and a meaningful share of planned projects face delay. And whether inference demand grows into the capacity, because the capacity is being committed now regardless.