Nvidia Isn't Just Selling Chips to OpenAI Anymore. It's Guaranteeing the Debt.

Traditional infrastructure debt markets — the ones that routinely fund power plants, toll roads, and data centers — looked at OpenAI's Ohio campus and said no. OpenAI is an unprofitable private company without an investment-grade credit rating, and the $500 billion buildout its backers are planning needs a guarantor with real credit standing before institutional lenders will write checks. Nvidia, according to reporting from the Wall Street Journal, is in talks to be that guarantor — to the tune of $250 billion in construction financing, with another $350 billion in chip purchase financing under separate discussion.

This is not a vendor relationship. A vendor sells product and gets paid. A guarantor pledges its own balance sheet against the possibility that the buyer can't service the debt. What Nvidia is reportedly considering is a structural position in OpenAI's financial architecture: if the Ohio campus generates less revenue than its debt service requires, Nvidia is on the hook. That's a fundamentally different kind of exposure than chip sales, and it reflects how dependent the two companies have become on each other's success.

The campus itself is being developed by SoftBank's energy subsidiary on the site of a former uranium enrichment plant in Piketon, Ohio. The full build targets 10 gigawatts of capacity — enough to power roughly 7.5 million American homes. Phase one targets approximately 800 megawatts, with completion scheduled for around 2028. The scale is designed to support OpenAI's next generation of model training and inference, workloads that require compute density that existing cloud infrastructure cannot provide. But training and inference revenue doesn't materialize before the campus is built, which is why the financing structure matters: someone has to fund the gap between construction and cash flows.

The historical precedent here is instructive and uncomfortable. During the 1990s telecom buildout, equipment vendors like Lucent and Nortel extended vendor financing to telecoms that couldn't otherwise afford the infrastructure to run their networks. When the revenue projections didn't materialize — when demand forecasts turned out to be wrong — the telecoms defaulted and the vendors collapsed under the weight of uncollectable receivables. Lucent's market cap fell from $258 billion to under $1 billion between 1999 and 2002. The mechanism Nvidia is reportedly entering is structurally similar: lender to a customer whose revenue projections are the primary justification for the loan.

Whether AI infrastructure economics are different enough from telecom economics to change the outcome is the real question. OpenAI's API and subscription revenues are real and growing, enterprise adoption of AI is structurally different from speculative consumer broadband bets, and the applications running on compute are demonstrably valuable. But the gap between those facts and a $600 billion combined guarantee is large enough to warrant skepticism. Nvidia's position in this deal isn't driven by generosity — it's driven by the fact that OpenAI is its largest customer, and a stalled Ohio campus means stalled chip sales at exactly the scale Nvidia needs to justify its own infrastructure investments and market valuation.

For the broader AI ecosystem, the Ohio deal reveals something the headline numbers obscure: frontier AI at scale is not yet self-funding. Revenue curves are real but still growing. Compute requirements are outpacing what the revenue can finance through traditional capital markets. The gap between those two things is currently being filled by a combination of equity investment, government support, and — if the reporting holds — vendor financing from the company that sells the chips. It works if the revenue curves eventually catch up. It doesn't if they don't. And Nvidia, more than any other company, has the most to lose from the second scenario — which is precisely why it's the one being asked to guarantee the gap.