Two months after Nvidia and OpenAI unveiled their eyepopping plan to deploy a minimum of 10 gigawatts of Nvidia programs—and as much as $100 billion in investments—the chipmaker now admits the deal isn’t really ultimate.
Talking Tuesday at UBS’s International Know-how and AI Convention in Scottsdale, Ariz., Nvidia EVP and CFO Colette Kress advised traders that the much-hyped OpenAI partnership remains to be on the letter-of-intent stage.
“We still haven’t completed a definitive agreement,” Kress stated when requested how a lot of the 10-gigawatt dedication is definitely locked in.
That’s a hanging clarification for a deal that Nvidia CEO Jensen Huang as soon as known as “the biggest AI infrastructure project in history.” Analysts had estimated that the deal might generate as a lot as $500 billion in income for the AI chipmaker.
Kress’s feedback recommend one thing extra tentative, even months after the framework was launched.
A megadeal that isn’t within the numbers—but
It’s unclear why the deal hasn’t been executed, however Nvidia’s newest 10-Q provides clues. The submitting states plainly that “there is no assurance that any investment will be completed on expected terms, if at all,” referring not solely to the OpenAI association but additionally to Nvidia’s deliberate $10 billion funding in Anthropic and its $5 billion dedication to Intel.
In a prolonged “Risk Factors” part, Nvidia spells out the delicate structure underpinning megadeals like this one. The corporate stresses that the story is barely as actual because the world’s means to construct and energy the information facilities required to run its programs. Nvidia should order GPUs, HBM reminiscence, networking gear, and different elements greater than a 12 months prematurely, usually through non-cancelable, pay as you go contracts. If clients reduce, delay financing, or change route, Nvidia warns it might find yourself with “excess inventory,” “cancellation penalties,” or “inventory provisions or impairments.” Previous mismatches between provide and demand have “significantly harmed our financial results,” the submitting notes.
The most important swing issue appears to be the bodily world: Nvidia says the supply of “data center capacity, energy, and capital” is essential for purchasers to deploy the AI programs they’ve verbally dedicated to. Energy buildout is described as a “multi-year process” that faces “regulatory, technical, and construction challenges.” If clients can’t safe sufficient electrical energy or financing, Nvidia warns, it might “delay customer deployments or reduce the scale” of AI adoption.
Nvidia additionally admits that its personal tempo of innovation makes planning more durable. It has moved to an annual cadence of latest architectures—Hopper, Blackwell, Vera Rubin—whereas nonetheless supporting prior generations. It notes {that a} quicker structure tempo “may magnify the challenges” of predicting demand and may result in “reduced demand for current generation” merchandise.
These admissions nod to the warnings of AI bears like Michael Burry, the investor of “the Big Short” fame, who has alleged that Nvidia and different chipmakers are overextending the helpful lives of their chips and that the chips’ eventual depreciation will trigger breakdowns within the funding cycle. Nevertheless, Huang has stated that chips from six years in the past are nonetheless operating at full tempo.
The corporate additionally nodded explicitly to previous boom-bust cycles tied to “trendy” use instances like crypto mining, warning that new AI workloads might create comparable spikes and crashes which might be onerous to forecast and may flood the grey market with secondhand GPUs.
Regardless of the dearth of a deal, Kress burdened that Nvidia’s relationship with OpenAI stays “a very strong partnership,” greater than a decade previous. OpenAI, she stated, considers Nvidia its “preferred partner” for compute. However she added that Nvidia’s present gross sales outlook doesn’t depend on the brand new megadeal.
The roughly $500 billion of Blackwell and Vera Rubin system demand Nvidia has guided for 2025–2026 “doesn’t include any of the work we’re doing right now on the next part of the agreement with OpenAI,” she stated. For now, OpenAI’s purchases circulation not directly via cloud companions like Microsoft and Oracle reasonably than via the brand new direct association specified by the LOI.
OpenAI “does want to go direct,” Kress stated. “But again, we’re still working on a definitive agreement.”
Nvidia insists the moat is undamaged
On aggressive dynamics, Kress was unequivocal. Markets currently have been cheering Google’s TPU – which has a smaller use-case than GPU however requires much less energy – as a possible competitor to NVIDIA’s GPU. Requested whether or not these sorts of chips, known as ASICS, are narrowing Nvidia’s lead, she responded: “Absolutely not.”
“Our focus right now is helping all different model builders, but also helping so many enterprises with a full stack,” she stated. Nvidia’s defensive moat, she argued, isn’t any particular person chip however your complete platform: {Hardware}, CUDA, and a always increasing library of industry-specific software program. That stack, she stated, is why older architectures stay closely used at the same time as Blackwell turns into the brand new customary.
“Everybody is on our platform,” Kress stated. “All models are on our platform, both in the cloud as well as on-prem.”
