There is a rare sentence in tech: the world's most valuable chip company invested in the world's most valuable AI startup, and now its investors want that undone. The Decoder reports that investor pressure is forcing Nvidia to shrink its bet on OpenAI. The stake dates back to OpenAI's record funding round last year, when Nvidia joined a round reported at around $50 billion and a $500 billion valuation, investing close to $10 billion of its own money.
The setup always looked strange. Nvidia produces the AI accelerators that every frontier lab needs, including OpenAI's direct rivals. Then it became a major shareholder in OpenAI, one of its largest customers. A chip supplier that owns part of its biggest customer is a supplier that other customers — Anthropic, xAI, Mistral, Meta — have to trust with their orders and their roadmaps.
Investor pressure has three understandable causes. First, regulatory tail risk: antitrust authorities on both sides of the Atlantic have spent years mapping the ties between Big Tech and frontier AI labs, from Microsoft-OpenAI to Amazon and Google in Anthropic. The EU kept Microsoft's exclusivity clauses under competition review even after deciding the partnership did not count as a merger. Second, concentration: Nvidia's share price already reflects the entire AI trade, and adding a stake in one lab on top of that multiplies exposure. Third, neutrality: every dollar Nvidia keeps in OpenAI is an argument for Anthropic, xAI or Mistral to look elsewhere for chips.
What does Nvidia lose by trimming? In practice, little. OpenAI remains dependent on Nvidia hardware at almost unimaginable scale, and Nvidia stays the dominant vendor that can deliver at that scale. The equity stake bought influence, but it also bought suspicion. A smaller stake restores the cleaner position Nvidia had before the boom — the neutral arms dealer of the AI era.
Anthropic's numbers complicate the bubble story
The timing matters. Just as Nvidia investors demand less OpenAI exposure, Anthropic's figures are undermining the bear thesis. The Decoder summarises them as defying bubble warnings — and the public numbers back that reading. Anthropic's annualised revenue run rate reportedly crossed $7 billion at the end of 2025, roughly five times where it stood a year earlier. For a company whose flagship product is only a few years old, that is the kind of curve usually seen only in retrospect.
Why this is relevant to the bubble debate: Anthropic is a pure-play AI company. It does not sell phones, ads or cloud storage. The bulk of its revenue comes from people and businesses paying for Claude access and API tokens. Sceptics have argued for two years that AI spending is infrastructure chasing imagined demand. Anthropic's growth says that, for frontier models at least, somebody is paying — and paying repeatedly.
The darker reading still holds. Anthropic spends enormous sums on Nvidia chips and cloud capacity; revenue growth without profit is not the same as a healthy market. But the bubble narrative now has to explain a company that turned roughly zero commercial revenue a few years ago into a multi-billion-dollar run rate without the cushion of a search monopoly or an OS bundle.
The European angle: a neutrality dividend
The Nvidia-OpenAI story was never only Silicon Valley business. The European Commission has spent years mapping exactly this kind of entanglement: Microsoft-OpenAI, Google-Amazon-Anthropic, Nvidia-OpenAI. Its consistent warning is that when one company is simultaneously chip supplier, cloud provider, investor and customer, competition can disappear without a single acquisition being announced. The EU AI Act does not regulate who owns AI labs, but it imposes transparency and documentation duties on general-purpose model providers — the first wave of obligations has applied since August 2025 — and those duties matter more when model supply is concentrated.
If Nvidia ends up owning less of OpenAI, European users get a small but real neutrality dividend. European AI runs on Nvidia hardware — rented from hyperscalers, from national compute centres, or from EuroHPC's AI factories. The less entangled Nvidia is, the more credible its promise to serve all customers equally. That includes European labs such as Mistral and Aleph Alpha, and the open-weights ecosystem many European developers use as a hedge. In our own AI Arena benchmarking tests, we keep measuring what open models can do on a single GPU; the reason is practical — European developers want options that cannot be revoked by a pricing change in California.
There is also a currency reality. OpenAI and Anthropic price APIs in dollars, with European VAT on top, so a weaker euro makes every token more expensive. More neutrality between suppliers does not fix the dollar denomination, but it does give European CTOs negotiation leverage — and it keeps alternative providers and open weights alive as credible threats.
What to watch next
Three things, from a European buyer's perspective. First, how far the reduction goes: the reports point to shrinking, not a clean exit, and Nvidia's GPU supply agreements with OpenAI remain the real relationship to watch. Second, whether Anthropic keeps publishing growth numbers like the recent ones — the bubble debate will live or die on whether the curve continues, not on a single quarter. Third, Brussels: the European Commission has not yet reacted to the Nvidia-OpenAI stake, but its practice of sending information requests to AI partnerships suggests it will be watching closely.
These two headlines are really one story. Investor pressure on Nvidia is the market asking for fewer single points of entanglement in the AI supply chain. Anthropic's revenue numbers are the market showing that demand is real. For European AI users, both developments run in the same direction: a more neutral supplier landscape and a demand base solid enough to keep the ecosystem alive. The AI bubble may yet deflate — but it is not doing so this quarter.
Is Nvidia leaving OpenAI completely?
The Decoder's report describes a shrinking, not a full exit. Nvidia remains OpenAI's most important hardware supplier, and OpenAI has few alternative chip vendors at the required scale. A reduced equity stake mainly lowers regulatory and conflict-of-interest risk.
Will the reduced stake change GPU prices for European customers?
Not directly. GPU prices are set by supply, demand and long-term contracts, not by shareholdings. But a more neutral Nvidia reduces the perception — and the risk — that OpenAI gets preferential treatment, which matters for the long-term health of European compute providers and open-weights projects.
Does the EU AI Act regulate investments like Nvidia's stake in OpenAI?
No. The AI Act regulates the use and deployment of AI systems inside the EU, not foreign shareholdings. Investment structures fall under EU competition law, which is why the Commission has been examining Big Tech's AI partnerships through merger and antitrust lenses.
