analysis OpenAI and Nvidia have signed a letter of intent wherein OpenAI agrees to buy at least 10 gigawatts of Nvidia systems for its datacenters, while the AI arms dealer returns the favor with an investment of up to $100 billion in the house that Altman built.
The first phase of the deal, which Nvidia and OpenAI jointly announced on Monday, will see OpenAI deploying Team Green’s Vera Rubin platform in its datacenters starting in H2 2026. To help defray the cost and to own a bigger piece of the AI pie, Nvidia will start buying into OpenAI “progressively as each gigawatt is deployed,” the press release explains. As part of the deal, Nvidia will become the “preferred strategic compute and networking partner” of OpenAI, whatever that means. The deal is non-exclusive on OpenAI’s part, we’re told, meaning it can use competing chips from AMD or others, should it care to do so.
The announcement has enough wiggle room to drive an AI-powered self-driving semi through – it’s a letter of intent for a strategic partnership, which is a non-binding kinda-sorta contract, and the deal calls only for Nvidia to invest “up to” $100 billion for as long as OpenAI keeps buying its chips. If the whole thing sounds a lot like the company whose stock price has benefited most from the AI bubble keeping said bubble inflated by helping to fund its own most important customer, well, you might be savvier than the average stock market participant, who collectively bid Nvidia’s stock up about 4% on the news. (Then again, the pop may be years away, in which case investors sitting on the sidelines are the real fools. Nobody knows!)
On a more mundane level, Nvidia has chips that OpenAI needs, and the AI leader gets some of the funding it will need to carry on burning through cash until it builds a market position that comes close to profitability, which Sam Altman recently admitted wouldn’t happen for “quite a while,” even as the firm is on track to hit about $20 billion in annualized revenue this year. If OpenAI does in fact turn the corner to become a self-sustaining business, then Nvidia owns part of what could become the most valuable software company in history.
Oracle is also part of this back-scratching arrangement. Just a couple of weeks ago, OpenAI reportedly committed $300 billion over five years to pay Oracle for about five gigawatts of compute capacity that likely also comes from Nvidia GPUs. It’s not clear where OpenAI will find that money, but that didn’t stop Oracle from boasting about a huge projected increase in demand for AI infrastructure, which inspired stockholders to send shares up 40% at one point, briefly making Oracle founder Larry Ellison the richest human in the world.
The two companies are already closely intermingled – some might say co-dependent. Last week, Nvidia said it was supplying the hardware for the British arm of OpenAI’s Stargate datacenters, albeit using Grace Blackwell Ultra GPUs rather than Rubin silicon. And in July, it announced a similar deal in Norway for another Stargate bit barn.
Meanwhile, it remains unclear how big and lucrative the demand for AI tools actually is, with reports increasingly suggesting that top-down company-mandated AI pilots are not offering much of a return on investment, even as individual workers are finding ways to use AI to make their own jobs easier or more productive. No matter! The train keeps rolling… ®