Nvidia's $100 Billion OpenAI Deal Collapses Into a $30 Billion Equity Bet

Five months ago, Nvidia CEO Jensen Huang and OpenAI CEO Sam Altman stood together and announced what they called the biggest AI infrastructure deployment in history: a $100 billion, multi-year partnership that would see Nvidia invest in ten $10 billion installments while OpenAI deployed ten gigawatts of Nvidia computing capacity.

No contract was ever signed. No money ever changed hands. And now both companies have quietly walked away.

According to the Financial Times, Nvidia and OpenAI have abandoned the unfinished framework in favor of something far simpler: a $30 billion equity investment by Nvidia, tied to OpenAI's next funding round. The shift transforms Nvidia from a strategic infrastructure partner into a shareholder, and it raises uncomfortable questions about whether the original deal was ever more than a press release.

The Numbers

Who Benefits

Nvidia gets ownership without obligation

The original deal locked Nvidia into a decade of capital commitments tied to data center construction. That is a lot of risk for a company watching AI infrastructure spending strain the entire supply chain. An equity stake is cleaner. Nvidia gets exposure to OpenAI's upside without the operational headaches of co-financing ten gigawatts of computing capacity.

Jensen Huang was already managing expectations in late January, telling reporters the $100 billion figure was "never a commitment" but rather "an invitation." By early February, he was dismissing friction reports as "nonsense" while simultaneously negotiating the smaller deal.

"I really love working with Sam," Huang told Fortune. The original sum, he said, was "never a commitment."

OpenAI gets freedom to shop around

Here is what Nvidia's $30 billion buys: stock, not loyalty. Under the original framework, OpenAI's infrastructure buildout was anchored to Nvidia hardware. With a straight equity investment, that money can go anywhere, and OpenAI is already spending it elsewhere.

In October 2025, OpenAI signed a deal with AMD to deploy six gigawatts of AMD GPUs over multiple years and generations. AMD issued OpenAI a warrant for up to 160 million shares of AMD common stock, roughly a 10% stake. That deal followed Meta's own multibillion-dollar Nvidia commitment just days earlier, suggesting OpenAI was hedging against the kind of supplier concentration that makes procurement teams nervous.

OpenAI has also struck deals with Cerebras and is working with Groq, both of which specialize in inference-optimized silicon.

Who Pays

Nvidia's training monopoly meets an inference problem

The collapse was not about training. Nvidia's GPUs remain unchallenged for pre-training frontier models. The tension is about inference, the workload that generates revenue when millions of users interact with ChatGPT and GPT-5.2 in real time.

Reuters reported that OpenAI grew frustrated with how quickly Nvidia's hardware could deliver responses for certain tasks, particularly coding-related workloads. OpenAI was specifically concerned about latency for inference-heavy products like Codex. When your business model depends on serving billions of API calls per day, milliseconds matter, and OpenAI concluded Nvidia's general-purpose GPUs were not fast enough.

Specialized inference chips from Cerebras and Groq use large amounts of SRAM embedded directly in the silicon, reducing reliance on external memory. That architectural difference can dramatically cut the time chips spend fetching data rather than processing it. OpenAI's target: alternative chips eventually handling around 10% of its inference workload.

Shareholders face dilution math

OpenAI's broader funding round puts the numbers in perspective. The company is raising over $100 billion at a valuation exceeding $850 billion, with a pre-money value of $730 billion. The investor roster reads like a sovereign wealth fund's portfolio:

The funding arrives in installments throughout 2026, aligning with OpenAI's buildout plans. The company has told investors it anticipates needing approximately $600 billion in computing power between now and 2030.

That is not a typo. OpenAI believes it needs more than half a trillion dollars in compute over the next four years. For context, its annualized revenue recently exceeded $20 billion, with both revenue and computing access metrics tripling annually. The question investors should be asking is not whether the funding round closes, but whether tripling revenue forever is a reasonable assumption or just the latest version of "this time it's different."

What the Circular Money Problem Looks Like

Financial analysts have flagged a structural issue with Nvidia's investment. The capital flows in a circle: Nvidia invests $30 billion in OpenAI, and OpenAI uses a significant portion of that money to buy Nvidia GPUs. Nvidia books the revenue, which boosts its stock price, which makes its investment in OpenAI look smarter, which encourages more investment.

This is not fraud. It is not even unusual in enterprise technology. But it does mean the headline figures overstate how much net new capital is actually entering the AI ecosystem. When your biggest customer is also your biggest investor, the numbers are not as independent as they appear.

"We love working with NVIDIA and they make the best AI chips in the world," Sam Altman said in a post on X. "We hope to be a gigantic customer for a very long time."

The statement was posted the same week OpenAI was finalizing chip deals with two of Nvidia's competitors.

What Happens Next

The broader market context matters here. US tech stocks with heavy AI exposure declined roughly 17% since early 2026. That pullback is pushing everyone toward what the Financial Times called "more pragmatic financial arrangements," a polite way of saying the era of nine-figure letters of intent may be ending.

For anyone evaluating which models offer the best value today, the infrastructure story is the one to watch. OpenAI's chip diversification strategy could reshape inference pricing across the industry. If Cerebras and Groq deliver on their promises, the cost of running inference at scale drops, and that eventually flows through to API pricing.

The $100 billion deal that Jensen Huang and Sam Altman announced in September was a statement of ambition. The $30 billion equity bet that replaced it is a statement of reality. In AI, those two things are still very far apart.

Sources:

• Nvidia and OpenAI abandon unfinished $100bn deal in favour of $30bn investment - Financial Times
• Nvidia Dumps $100 Billion Plan for a Much Smaller OpenAI Investment Bet - Yahoo Finance
• OpenAI reportedly finalizing $100B deal at more than $850B valuation - TechCrunch
• Nvidia CEO denies 'nonsense' report he's unhappy with OpenAI - Fortune
• Nvidia, OpenAI appear stalled on their mega deal - CNBC
• OpenAI Reportedly Discontent With NVIDIA GPUs for Inference - TrendForce
• NVIDIA & OpenAI Near $30B Investment Deal After $100B Plan Stalls - Meyka
• OpenAI chip deal with Cerebras adds to roster of Nvidia, AMD, Broadcom - CNBC