The Wall Street Journal's report that OpenAI has fallen short of its internal revenue and user-growth targets, including the missed goal of one billion weekly ChatGPT users by year-end, has dragged Oracle, Broadcom and Advanced Micro Devices lower and put Sam Altman's $660 billion compute build-out under fresh scrutiny. For European policymakers and operators, the share-price moves are not the story. The argument that the entire transatlantic AI capex cycle has been overpriced has just been made in print, with named numbers, by the leader's own finance team.
Brussels has spent two years being told that the AI Act would push talent and capital westward across the Atlantic. Sarah Friar's reported warning that OpenAI may struggle to fund its compute commitments unless growth accelerates does not invalidate that argument, but it does change its weight. If the world's most-valuable private AI company cannot reliably grow into the cost base it has already committed to, the European compliance overhead looks less like a tax and more like a price for predictability. That is a different conversation in Berlin, Paris and The Hague than the one civil servants were having on Monday.
Mistral is the most direct beneficiary. Arthur Mensch's company crossed a €400 million annualised revenue run rate this quarter, against €20 million a year ago, and has just closed an €830 million debt facility for its Paris data centre. None of that scales with OpenAI's headline ARR. What Mistral has been selling, an open-weights European alternative with sovereignty guarantees attached, has just been priced more attractively because the incumbent's option value fell. The same logic carries into last week's Cohere and Aleph Alpha merger, which closed at a roughly $20 billion combined valuation with Schwarz Group's STACKIT cloud as its European infrastructure spine. Two weeks ago that deal read as the quiet sale of Germany's national AI champion. Today it reads as decent timing.
"Without its own AI, Europe will become a vassal state of the United States and China." Arthur Mensch, CEO, Mistral AI, in remarks reported by MLex this month
The European listed exposures sit on three different rungs of risk:
- ASML is the most insulated. Its EUV order book is multi-year and multi-customer; TSMC's appetite is the variable, not OpenAI's monthly figures.
- SAP is the cleanest long. Christian Klein has bet enterprise AI on a multi-model strategy in which Microsoft Copilot is one option among several, and a slower OpenAI to Microsoft revenue arc gives Joule more room to land.
- Adyen has no direct exposure but stands to benefit from any rotation back toward European software with rational unit economics.
- Schneider Electric and Siemens, both of which have priced in aggressive European data centre power spend, do face a real downside read-through if the broader infrastructure buildout slows.
The political read-through is sharper. President Macron's bet on Mistral has been derided in Washington as flag-planting; today it looks closer to vindication. Robert Habeck's team in Berlin and the Dutch government in The Hague can credibly raise their ambitions for the InvestAI initiative and for the European Investment Bank's quietly expanded AI credit line. Henna Virkkunen's AI Office, which had been negotiating the General Purpose AI Code of Practice from a perceived position of weakness, now negotiates from a stronger one. The cost-of-compliance discount the Commission has been holding back just became more valuable, because the cost of doing nothing has become more visible.
None of this is decisive. OpenAI still has an order of magnitude more revenue than every European model lab combined, and Friar and Altman pushed back on the WSJ piece in a joint statement calling it "ridiculous." The window opened this week is real, but it is narrow.
The AI in Europe View
We expect the next ninety days to determine whether this is a turning point or a footnote, and the realistic outcome is somewhere in between. OpenAI's miss does not rewrite the AI hierarchy; the company will still finish 2026 with revenue an order of magnitude greater than any European competitor. What the WSJ report does is hand Brussels and Paris a temporary repricing of risk that they have to convert quickly. The General Purpose AI Code of Practice, the AI Office's enforcement budget, the EIB's InvestAI envelope, and the next round of Mistral and Cohere and Aleph Alpha financing will all be negotiated under a new comparison: not against an American incumbent compounding at a hundred per cent a year, but against one whose chief financial officer has reportedly told her board that the compute bill may not be payable.
Brussels' mistake, on past form, will be to read the news as confirmation that the AI Act has already won. It has not. The realistic outcome is that European AI policy clears the next two years only if the Commission converts this moment into hard procurement commitments, hard public-cloud guarantees and a hard line on the Code of Practice. Sovereign AI is not yet sustainable as a market position. It will become so only if European public buyers, European banks and European regulators move at the same speed Washington's investors are now backing away from. We are less confident than the consensus that they will.
Comments
Sign in to join the conversation. Be civil, be specific, link your sources.