Nvidia Took $70 Billion Off the Table With OpenAI
February 22, 2026 by Asif Waliuddin
Nvidia Took $70 Billion Off the Table With OpenAI
According to Financial Times reporting, Nvidia has restructured its OpenAI investment from a $100 billion long-term commitment to a $30 billion milestone-based structure. A 70% reduction, contingent on OpenAI hitting defined performance targets.
This was buried in a "funding rounds" section of a market roundup. It should be on the front page.
The Hype
The narrative around AI investment is relentlessly bullish. OpenAI is the most valuable private company in the AI space. It has 200 million weekly users. It is projecting $20 billion in annualized revenue. Hyperscalers are spending $690 billion on AI infrastructure in 2026. The Stargate project has $400 billion in committed capital. Everything is up and to the right.
In this environment, a $30 billion investment from Nvidia in OpenAI sounds massive. And it is. Thirty billion dollars is an extraordinary commitment by any standard.
But it was supposed to be $100 billion. The reduction is the story, not the remaining number.
The Reality
Milestone-based investment structures are not neutral. They are a specific instrument that exists for a specific reason: the investor needs optionality to limit exposure if the recipient does not perform.
When an investor commits $100 billion unconditionally, they are saying: we believe in this company's trajectory enough to lock in capital regardless of near-term outcomes. When the same investor restructures to $30 billion with milestones, they are saying: we believe enough to start, but we need gates that allow us to stop.
The shift from unconditional to conditional is a change in conviction. Not necessarily a loss of conviction -- Nvidia clearly still believes in OpenAI enough to commit $30 billion. But the nature of the belief has changed from "we know this works" to "we think this works, but we want to verify along the way."
Now consider who is making this assessment.
Nvidia is not a typical financial investor in OpenAI. Nvidia is OpenAI's primary chip supplier. They have more visibility into OpenAI's infrastructure economics than any outside party and most inside parties. They know:
- How many chips OpenAI is ordering and at what cadence
- What OpenAI's inference cost structure looks like at the hardware level
- How efficiently OpenAI is utilizing its compute capacity
- What OpenAI's chip roadmap looks like 2-3 years out
- Whether OpenAI's multi-vendor strategy (AMD, Cerebras, Broadcom) is diluting Nvidia's position
Nvidia's information advantage is not marginal. It is structural. They see the unit economics that nobody else sees. And after seeing those economics, they decided to reduce their exposure by $70 billion and add milestone gates.
What Changed
The Financial Times reporting does not specify what triggered the restructuring. But we can reason about what changed between the original commitment and the revision.
Possibility 1: OpenAI's revenue trajectory is less certain than originally projected. The $20 billion ARR figure is impressive but may not be growing at the rate needed to justify $100 billion in ecosystem investment. Consumer subscription growth may be decelerating as AI novelty fades for casual users. Enterprise revenue is growing but requires expensive, custom deployments that compress margins.
Possibility 2: OpenAI's multi-vendor chip strategy reduces Nvidia's strategic return. If OpenAI is routing significant inference volume to Cerebras wafer-scale engines and cost-optimized workloads to AMD, Nvidia's share of OpenAI's compute spend is declining. A $100 billion investment in a company that is actively diversifying away from your products has a different return profile than a $100 billion investment in a captive customer.
Possibility 3: The broader AI infrastructure ROI picture is muddier than it looked six months ago. Nvidia sees the entire market, not just OpenAI. If they are observing that hyperscaler utilization rates, enterprise adoption timelines, or inference revenue growth are softer than the capex numbers suggest, that context informs their OpenAI-specific risk assessment.
None of these possibilities require that Nvidia has lost faith in AI. They require only that Nvidia has updated its risk model for OpenAI specifically -- and that the update was large enough to warrant restructuring a $100 billion commitment.
The Contextual Signal
This restructuring is happening in the same week as:
- Hyperscalers committing $690 billion in AI capex for 2026
- Microsoft disclosing $80 billion in unfilled Azure orders
- BlackRock rotating its AI ETF toward infrastructure plays and away from model vendors
The pattern: the money going into AI infrastructure is increasing. The conviction in any specific model vendor's long-term position is decreasing. The smart money is betting on "whoever wins AI needs chips and power" while hedging against "we do not know which AI company wins."
Nvidia's restructuring is this same trade, executed by the company with the most information. They are staying in the AI ecosystem (chips will be needed regardless) while reducing concentrated exposure to one model vendor's execution.
The Bottom Line
The $70 billion that Nvidia removed from its OpenAI commitment is one of the clearest signals in the current AI market. It does not mean AI is failing. It means the company with the best visibility into AI infrastructure economics decided it needed protection against OpenAI-specific execution risk.
When the chip company gets cautious about the AI lab, pay attention. Nvidia is not betting against AI. They are betting against certainty about which AI companies will generate returns at the scale their valuations imply.
The $30 billion that remains is still an enormous commitment. But the $70 billion that was removed tells you more about the state of AI economics than the $690 billion being poured in by everyone else.