Nvidia's plan to invest up to $100 billion in OpenAI, announced with fanfare in September 2025, has hit significant delays. The Wall Street Journal reported Friday that negotiations have stalled after some inside Nvidia expressed doubts about the transaction's structure. Meanwhile, Reuters claims OpenAI has grown dissatisfied with certain Nvidia chips and has been quietly exploring alternatives.
Both companies spent the weekend denying any rift. The denials were emphatic, frequent, and coordinated enough to suggest they're worried.
What was promised vs. what's happened
The September announcement was big: a letter of intent for Nvidia to invest up to $100 billion as OpenAI built out 10 gigawatts of computing infrastructure. The first gigawatt was supposed to come online in the second half of 2026, running on Nvidia's Vera Rubin platform.
Five months later, no contract has been signed. The "up to $100 billion" language, which sounded like a commitment, turns out to have been doing a lot of heavy lifting. Jensen Huang has been telling industry contacts that the figure was non-binding and never finalized. When asked directly whether the investment would hit $100 billion, he replied "no, no, nothing like that."
Huang did say Nvidia would make "a huge investment" and "probably the largest investment we've ever made." That could mean many things, and the vagueness seems intentional.
The inference problem
The more interesting story sits underneath the deal drama. According to Reuters, citing eight sources, OpenAI has been unhappy with Nvidia's hardware for inference tasks since last year. Inference, the process of running trained models to generate responses, is becoming the battleground as AI shifts from building models to deploying them at scale.
Sources told Reuters that OpenAI's teams attributed performance issues in Codex, its code-generation product, partly to Nvidia's GPU architecture. The complaint centers on speed. Seven sources said OpenAI wants hardware that can deliver answers faster for specific tasks like software development and AI-to-AI communication.
OpenAI reportedly needs alternative hardware for about 10% of its future inference computing. Not a massive share, but enough to matter. The company had discussions with startups Cerebras and Groq, both building chips with large amounts of on-chip memory that can speed up inference workloads.
Nvidia's defensive moves
Nvidia moved fast once OpenAI's reservations became clear. The chipmaker approached both Cerebras and Groq about potential acquisitions. Cerebras declined and signed a commercial deal with OpenAI instead. Groq took a different path.
By December, Nvidia locked up Groq with a $20 billion licensing agreement, effectively ending OpenAI's talks with the startup. Nvidia also hired away Groq's chip designers. The company called Groq's intellectual property "highly complementary" to its product roadmap, which is corporate-speak for "we bought what we needed."
That's a telling response. If Nvidia's inference hardware were already competitive, why scramble to acquire the technology your biggest customer was considering as an alternative?
The public relations blitz
Both CEOs went into damage control mode over the weekend. Huang, speaking to reporters in Taipei, called reports of tension "nonsense" and said he believes in OpenAI. "They're one of the most consequential companies of our time, and I really love working with Sam."
Sam Altman posted on X that Nvidia makes "the best AI chips in the world" and that OpenAI hopes to remain "a gigantic customer for a very long time." He added: "I don't get where all this insanity is coming from."
The insanity is coming from eight sources talking to Reuters and people familiar with the matter talking to the Wall Street Journal. But sure.
Circular financing concerns
Part of Nvidia's hesitation may stem from scrutiny over what analysts call circular financing. Nvidia invests in OpenAI, which uses that money to buy Nvidia chips, which boosts Nvidia's revenue, which justifies the investment. The loop looks tidy on paper but raises questions about how much of the AI boom represents real demand versus companies funding their own customers.
Nvidia has similar arrangements with CoreWeave and xAI. Huang recently said he regrets not investing more in Elon Musk's AI startup, calling it "an investment into a really great future company." That enthusiasm is easier to express when the investee is contractually obligated to become a major customer.
What happens next
The companies appear to be renegotiating. The Journal reported they're discussing a smaller equity investment, potentially tens of billions rather than $100 billion, as part of OpenAI's current funding round. That round is reportedly targeting $100 billion at a valuation above $800 billion, with Amazon, Microsoft, and SoftBank also in discussions.
OpenAI's infrastructure compute lead, Sachin Katti, posted that Nvidia's technology remains "foundational" and called the relationship "deep, ongoing co-design." Oracle clarified that whatever happens between Nvidia and OpenAI won't affect its separate $300 billion computing power agreement with OpenAI.
For now, both sides need each other. Nvidia's market cap crossed $4 trillion on the AI boom that OpenAI helped ignite. OpenAI needs chips to meet growth targets that assume continued scaling. But "need each other" and "like each other" are different things. The next few months should clarify which one this is.
