Nvidia has quietly become the biggest venture lender in artificial intelligence, and the numbers are staggering. CFO Colette Kress told analysts on August 26 that the company has already deployed nearly $50 billion into AI labs that buy its chips, with commitments lined up for more than $500 billion through partnerships with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR. She expects those labs to account for roughly a quarter of Nvidia’s business next year.
The loop is straightforward. Nvidia invests in an AI lab. The lab uses that money — or the credit Nvidia’s involvement unlocks — to build a data center filled with Nvidia chips. The purchase shows up as Nvidia revenue. The stock price climbs, the cash pile grows, and the company invests again. Kress acknowledged on the earnings call that the company recognized the scale of support it was providing and knew some would call it circular financing. She said Nvidia sees it differently.
Her defense has three parts. Outside lenders still evaluate every deal independently, and Nvidia isn’t making loans directly. The chips go to customers who are investment grade or backed by someone who is. And if a customer fails, the equipment can be moved to another buyer. That last point is the load-bearing wall — it only holds while demand exceeds supply, which Nvidia says it currently does.
The scale of infrastructure commitments is concrete. Nvidia has signed agreements with SB Energy for land, power, and building capacity that will host only Nvidia equipment — the first phase supports 4.25 gigawatts and goes to OpenAI, whose total planned Nvidia compute runs to roughly 12 gigawatts through 2030. A second, unnamed lab gets credit support covering nearly two gigawatts. Kress also described a smaller-cloud arrangement where Nvidia rents capacity itself, guaranteeing income for lenders while taking a cut of revenue above that floor. She said Nvidia gets paid twice.
Jensen Huang’s response to a question about OpenAI designing its own chips — the Jalapeño processor — was that Nvidia sells a platform that works across any cloud, while rival silicon is built for one service. On the money question, his only regret was not investing more and sooner. The rationale underneath it all: Kress told Morgan Stanley’s Joseph Moore that an agent needs 15 to 100 times the computing power of a human using the same system, and Huang claimed AI tipped over to being mostly agentic in the past month. No data was offered.
That assumption supports guidance of $108 billion in revenue this quarter and preliminary expectations of around 70% growth in the year to January 2028 — limited by supply, not demand. Memory prices are climbing faster than expected, with margins guided down to 74% this quarter and bottoming at 71-72%. One thing to watch: the $500 billion figure is subject to definitive agreements, meaning binding contracts haven’t been signed. It’s an intention, not money in place. Nvidia reports again on November 17.
