On September 9, 2025, ASML made its move: it paid €1.3 billion for an 11 percent stake in Mistral. By June 2026, Tech Policy Press estimated ASML's market value at nearly $700 billion, the highest of any European company. ASML was already the world's only supplier of the extreme-ultraviolet lithography machines needed to manufacture the most advanced chips. The Mistral investment gave it a foothold in the part of the AI supply chain Europe still lacked: leading AI models.

The Commission chose EUROPA, led by Domyn, for its Frontier AI Grand Challenge. The award reserves 2.5 percent of EuroHPC's total AI computing capacity for the consortium for one year and includes a dedicated cluster of 6,000 Nvidia Blackwell chips.

Germany's SPRIND is pursuing a different model through a separate €125 million competition. The opening stage provides €3 million, followed by rounds worth €8 million and €15.5 million. In May, Jano Costard defended the urgency of the effort: “Germany is leading this because we have no time to waste in waiting for other actors to get in that space.” EuroHPC, by contrast, spreads shared computing capacity across 19 AI Factories.

The programs embody competing theories of how Europe should produce a frontier model: concentrate resources behind one consortium, narrow the field through successive rounds, or distribute capacity among many users. Europe is financing all three without deciding which approach is supposed to deliver the result. Mistral, meanwhile, is absent from each of them.

OpenEuroLLM acknowledged in its March 6 report that it still faced “significant challenges” in securing enough computing power to train its final models. The shortage reflects a broader policy failure. Brussels has spread funding across programs built on conflicting assumptions, without deciding whether frontier-model training should be treated as shared public infrastructure or as a concentrated industrial campaign.

Epoch AI estimates that Anthropic spent $9.7 billion in 2025. CERN, by comparison, approved a 2026 budget of CHF 1.515 billion, or roughly €1.6 billion, while ESA approved €8.26 billion. Together, Europe's leading particle-physics and space institutions will spend about €9.9 billion, roughly what a single American AI company spent in one year. The comparison is not an argument for taking money from science or spaceflight. It shows the scale of investment Europe already accepts when it considers a project strategically important.

In May 2025, Europe adopted its SAFE defense instrument, which supplies €150 billion for defense procurement. Contractors must be based in the EU, EEA or Ukraine and maintain local executive management. The International Procurement Instrument already caps non-EU inputs at 50 percent in medical-device contracts worth more than €5 million. Yet Europe's Tech Sovereignty Package stopped short of imposing a comparable preference, largely to avoid provoking Washington. A Belfer Center brief found that only about 1 percent of public-sector data is stored in cloud services that exclude providers under non-EU control. Europe was willing to risk American displeasure over artillery procurement, but designed its AI policy to avoid the same confrontation over computing power.

The comparison ends there. Defense procurement can direct orders toward an existing European industrial base. Europe has no equivalent frontier-model training capacity at the scale it would need to procure. A purchasing rule cannot conjure a training run. The legal machinery already exists; the industrial capacity does not.

On July 22, TechFundingNews reported that Mistral was seeking new capital, with a proposed round of roughly €3 billion. The reported valuation approached €20 billion. Private capital may therefore be solving the problem. A European frontier model would still train on Nvidia silicon fabricated by TSMC, defeating any claim to full-stack sovereignty. A reported account of the May 12 National Assembly hearing says Arthur Mensch asked for European-preference procurement and priority electricity for European AI. Mensch requested regulatory proportionality as well. Those requests concerned state power rather than cash. I would change my mind if Mistral closes the round and secures multi-year power and compute at frontier scale. It must then disclose the run and finance its successor from European demand without procurement preferences.

Brussels should write a European preference into AI procurement and reserve sustained compute and power for Mistral, conditioned on frontier-scale performance. A European frontier lab gives the Atlantic alliance a second supplier and strengthens it against Beijing without decoupling from Washington. Continued fragmentation leaves Europe a customer of Washington and a market for everyone else. The fallback of Chinese open weights is a revocable policy choice, as July's Washington fight makes clear.

ASML recognized the missing layer with a €1.3 billion check. Brussels has built programs around the decision it still has to make.

Opinion

San Francisco

Editor-in-Chief and founder of Implicator.ai. Former ARD correspondent and senior broadcast journalist with 10+ years covering tech. Writes daily briefings on policy and market developments. Based in San Francisco. E-mail: editor@implicator.ai