Mistral chief warns Europe's militaries could be switched off without sovereign AI

Europe's most valuable artificial intelligence company has issued a blunt warning: without sovereign AI capabilities, European militaries risk being rendered inoperable by foreign powers. Arthur Mensch, chief executive of Paris-based Mistral, told an audience of policymakers in Brussels on Tuesday that artificial intelligence has reached the same strategic tier as nuclear weapons. "If you don't have artificial intelligence in your systems, you actually don't have an army," he said.

The strategic vulnerability of foreign-controlled AI

Mensch's argument rests on a straightforward proposition. Modern military systems, from intelligence analysis and targeting to logistics and command-and-control, are becoming deeply dependent on large language models and other AI tools. If those tools are supplied by companies outside Europe, the supplier or its government retains the ability to restrict access, degrade performance or cut service entirely. "If these artificial intelligence systems are actually procured from foreign companies, then ... our militaries can be turned off," Mensch said. "Do we want our military forces to be turned off because we have general political misalignment sometimes?"

The question is not hypothetical. The United States dominates the global market for foundation models, cloud infrastructure and high-end semiconductors. European armed forces already rely heavily on American cloud providers for data processing and on US-designed chips for compute. That dependence has been manageable while transatlantic alignment held. But the return of a transactional US foreign policy, combined with live conflicts in Ukraine and the Middle East, has sharpened European thinking about what strategic autonomy means in the digital age.

Mistral's position in the European landscape

Mistral, founded in 2023 by former researchers from Meta and Google DeepMind, has become the flag-bearer for European foundation-model development. The company was valued at over €11 billion in its most recent funding round, making it the continent's most valuable AI startup. Its models are open-weight, a deliberate contrast to the closed systems of OpenAI, Anthropic and Google. That openness has made Mistral the default choice for European governments and companies that want to run models on their own infrastructure rather than sending data to American clouds.

But Mensch's appearance in Brussels was not merely a sales pitch. He presented a set of policy proposals intended to shape the European Commission's forthcoming technological sovereignty package, tentatively scheduled for the end of May. The package is expected to address cloud services, semiconductors and data-centre capacity, three layers of the stack that determine whether European AI can be trained, deployed and operated without leaving European jurisdiction.

Procurement as a lever for sovereignty

The central proposal from Mistral is that the bloc should use government procurement to ensure critical AI infrastructure remains European. "Most of Europe's AI workloads run on infrastructure controlled by foreign providers," the company's policy document states, leaving the bloc "vulnerable to geopolitical risks, supply chain disruptions and the loss of economic value." The remedy, Mistral argues, is to mandate European-controlled infrastructure for "critical workloads", public services, research and defence.

This approach mirrors the logic behind the European Chips Act and the EU Cloud Rulebook: create demand signals strong enough to sustain a domestic industrial base. But the AI layer adds urgency. Training a frontier model requires thousands of advanced GPUs, massive electricity supply and specialised engineering talent. Europe has some of each, notably the LUMI supercomputer in Finland and the Jean Zay system in France, but not at the scale or concentration found in the United States or China.

The safeguards debate: who draws the red lines?

A parallel debate concerns the guardrails placed on military uses of AI. Last month the US government clashed with Anthropic after the American company set red lines restricting the use of its models in certain weapons applications. The Trump administration viewed those restrictions as an infringement on sovereign defence decisions. Mensch took a different view: "It's not up to them to set guidelines but to their customers." In other words, AI providers should build capable systems; democratically elected governments should decide how to use them.

Audrey Herblin-Stoop, Mistral's senior vice president for global affairs, said it is too early to address whether European governments will pressure AI companies to lower their safeguards in defence applications, as the Trump administration did with Anthropic. "Let's prioritize at the moment the ability for us to continue to develop a very strong ecosystem and have our own AI," she said. The distinction matters. If European companies accept government-imposed weakening of safety controls, they risk undermining the trust that makes their open-weight models attractive to civilian customers and allies.

Geopolitical context: Ukraine and the Middle East

Mensch framed his argument against an "increasingly bleak geopolitical landscape and live wars in Ukraine and the Middle East." The war in Ukraine has already demonstrated the speed at which commercial AI tools, satellite imagery analysis, drone targeting, electronic warfare, can be adapted for combat. Ukrainian forces have integrated Western commercial technologies at a pace that traditional defence procurement cannot match. But that integration has also created dependencies: when a US company restricts access or a cloud provider changes terms, the effect on the battlefield is immediate.

The Middle East conflicts have reinforced the lesson. AI-driven surveillance, predictive maintenance and decision-support systems are now standard in advanced militaries. The countries that control the underlying models and compute hold a persistent advantage. For Europe, the choice is not whether to use AI in defence but whether to build the stack at home or rent it from abroad.

What the Commission package must deliver

The European Commission's technological sovereignty package, expected in late May, will be the first concrete test of whether Brussels can translate rhetoric into industrial policy. The package is likely to include: a cloud and AI infrastructure act setting procurement preferences for European providers; a revised chips strategy targeting the 2nm and below process nodes that Europe currently lacks; and a data-centre investment framework addressing the energy and permitting bottlenecks that have slowed European build-out.

Mistral's proposals align with the cloud and AI infrastructure component. The company wants a clear definition of "European-controlled", meaning not just a European legal entity but European ownership of the hardware, the data centres and the model weights. That definition would exclude European subsidiaries of US hyperscalers, a point of contention in previous negotiations over the EU Cloud Rulebook.