For years Nvidia's constraint was simple: it could not make enough chips. That problem has been replaced by a stranger one. The world now has more appetite for AI compute than it has land, power and money to deploy it, and Nvidia has decided to help supply all three. On 17 August the company agreed to provide up to $105 billion in financial guarantees backing OpenAI's 20-year lease at a vast new campus in Pike County, Ohio, being built by SoftBank-backed SB Energy. The site will host Nvidia compute exclusively, OpenAI will be the tenant, and Nvidia is also putting $1.5 billion directly into the developer.
Ohio is only part of it. A week earlier, Nvidia announced partnerships with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to build financing platforms meant to mobilise more than $500 billion in outside capital for AI infrastructure. The stated goal is to give Nvidia's customers cheaper capital and to turn Nvidia compute into an investable asset class. Add earlier moves, a $2 billion stake in CoreWeave, a fresh investment in data-centre developer Cloverleaf, and a shape emerges. Nvidia is becoming part chipmaker, part infrastructure developer and part bank.
The obvious joke writes itself. Nvidia helps finance the buildings, the buildings fill with Nvidia hardware, and that hardware shows up as demand on Nvidia's own books, generating cash to finance the next round. Nvidia reckons OpenAI's deployments alone represent roughly 12 gigawatts of its compute through 2030, rising toward 16 if Ohio fully expands, an opportunity it values at around $600 billion. Which raises the question worth sitting with: at what point does a supplier stop measuring demand and start manufacturing it? A customer that independently raises money and buys GPUs is clean demand. A customer whose developer, leases and financing vehicles all trace back to the seller is something more circular.
None of this is new in kind. Vendor financing, where a maker lends its customers the money to buy its products, is decades old. Carmakers do it, industrial-equipment firms do it, and the comparison that should give everyone pause is the telecom boom of 2000. Cisco, Lucent and Nortel competed to finance the carriers buying their gear. Cisco committed roughly $2.4 billion in customer loans, Lucent around $7 billion. It worked beautifully until the customers stopped paying, at which point Cisco reserved nearly $900 million for bad loans and billions across the industry were written down as networks went bankrupt.
Nvidia's answer is that AI compute is different: durable infrastructure on the scale of electricity, not a warehouse chained to one tenant. Its Ohio guarantees are structured to shrink as OpenAI pays and facilities come online, cover only defined slices of land, power and lease value, and rest on the claim that the campus could be handed to another customer if OpenAI ever walked away. That may hold. But the tension is real, and Wall Street has noticed. AI hardware ages in a few product cycles while the financial obligations tied to the concrete last for decades, and Reuters Breakingviews warned last month that Nvidia was being drawn into a dangerous game, with the original Ohio proposal reportedly floating guarantees as high as $250 billion before being scaled back.
All of it lands days before Nvidia reports quarterly earnings, with analysts expecting revenue near $92 billion, up about 96% on the year. The machine is still spinning fast. The entire strategy rests on one assumption, that there will always be somebody willing to pay for more intelligence. So far that has been a spectacular bet. It is also, increasingly, a bet Nvidia is funding on both sides of the table.