No two companies are as important to the recent AI boom as OpenAI and Nvidia. OpenAI’s journey from research lab to chatbot juggernaut reoriented the entire tech industry around the thesis that large language models will change everything. Nvidia’s chips have been almost synonymous with scaling, on which OpenAI and others are collectively spending hundreds of billions of dollars. For the last few years, the companies’ fortunes have been aligned, and their relationship has been, for the most part, an obvious one: OpenAI, which has become one of the biggest startups in the world, is a major customer for Nvidia, which has become the largest public company in the world.
In 2025, their twin trajectories have left much of the rest of the economy behind, and their stated ambitions — necessary to keep the momentum going — have become extreme. In a recent blog post, Sam Altman said his company wanted to “create a factory that can produce a gigawatt of new AI infrastructure every week” to “maybe” be able to “figure out how to cure cancer,” or “provide customized tutoring to every student on earth.” OpenAI hasn’t had any trouble raising money so far, but when your investment projections start crossing into the trillions, financing starts to get tricky, particularly if you aren’t a company like Google or Meta, with an existing business throwing off tens of billions of dollars a year. Which might help explain this:
OpenAI and NVIDIA today announced a letter of intent for a landmark strategic partnership to deploy at least 10 gigawatts of NVIDIA systems for OpenAI’s next-generation AI infrastructure to train and run its next generation of models on the path to deploying superintelligence. To support this deployment including data center and power capacity, NVIDIA intends to invest up to $100 billion in OpenAI as the new NVIDIA systems are deployed.
Emphasis mine! The issue of circular financing in the AI world has been bubbling up for a while. As The Information reported in May, Nvidia has been engineering disorienting deals with customers for a while, most obviously with a company called Coreweave, which rents compute — basically, access to Nvidia hardware — to AI firms:
The chip giant invested $100 million in [Coreweave] in early 2023. It funneled hundreds of thousands of high-end graphics processing units to CoreWeave. And it agreed to rent back its chips from CoreWeave through August 2027.
In related news, just this week, Coreweave announced it had “expanded its partnership with OpenAI in a new deal worth up to $6.5 billion, bringing the total value of their agreements to $22.4 billion.” These announcements aren’t exactly attempts to obfuscate what’s going on here — “NVIDIA intends to invest… as the new NVIDIA systems are deployed” — but it’s worth stating even more plainly what’s happening here: Nvidia is investing money in one of its largest customers, which will use some of that money to buy or lease capacity from Nvidia. On Tuesday, we got a little more insight into how that would actually happen. OpenAI, Oracle (a cloud provider that depends on Nvidia), and Softbank (a major investor in Nvidia and OpenAI) fleshed out plans to build out more data centers, which would one day be full of Nvidia hardware paid for or leased by… OpenAI.
If you’re a layperson, this probably sounds a bit weird. Maybe our friends in finance can explain a hidden logic here? Rich Privorotsky of Goldman Sachs attempts to summarize the arrangement: “Nvidia invests up to $100bn for non-voting shares, and OpenAI uses the cash to buy Nvidia chips, with a plan to deploy what could be at least 10GW of Nvidia systems.” It is what it looks like, in other words: A small group of large companies handing money back and forth.
With a number of caveats — every cycle is different, and the earnings multiples aren’t as crazy yet — he also draws a parallel to the late 90s, which have been coming up a lot lately. “Vendor financing was a feature of that era, when the telecom equipment makers (Cisco, Lucent, Nortel, etc.) extended loans, equity investments, or credit guarantees to their customers who then used the cash/credit to buy back the equipment,” he writes. Indeed, vendor financing features prominently in tech bubble post-mortems. In 2002, Newsweek explained the strange role they played at the tail end of the 90s tech investment boom:
Swept up in the free-money spirit of the time, they were financing not only the customers’ purchase of their wares–the switches, routers and other nuts and bolts of the Internet–but the growth of the customers’ businesses, too. The standard practice of vendor financing thus became another wretched bubble excess: as if Ford were loaning customers money to buy cars–and boats, jewels and houses, too. “Vendor financing was the sickest sign of how things spun out of control,” says Tero Kuittinen, technology adviser to Finnish investment bank Opstock in Helsinki.
The parallels are obvious — or, at least, would look extremely obvious if the AI industry’s fortunes turned — but they’re not determinative. During the telecom buildout, vendor financing surged at a particular time. “Vendor-financing deals account for a relatively small fraction of the total $1 trillion in debt held by telecom carriers worldwide,” Newsweek noted at the time, but investment continued well after the an early 2000 market collapse “wiped out the silliest ideas for pet, toy and trinket Web sites,” on the thesis that “dot-coms might be dead, but the Internet itself would thrive,” a perfectly reasonable argument if timing doesn’t matter, which of course it does, quite a bit. Last month, Kuittinen argued that this time, things really are different, but not necessarily in a good way: