Mistral co-founder Arthur Mensch argues that Europe must control the on/off switch of its critical AI systems. What that means in practice — and what businesses and policymakers can do right now.
One Sentence That Says It All
"What matters is keeping the on and off switch of critical systems in your own hands." With these words, Arthur Mensch — co-founder of French AI company Mistral AI — summed up, in a March 2026 Deutschlandfunk interview, what digital sovereignty is really about. Not techno-nationalism. Not isolationism. Control over infrastructure that has become as fundamental to society and business as electricity or clean water.
Mensch speaks from experience. Mistral AI is one of the very few European companies genuinely competing in the global AI race. Founded in 2023 in Paris by former DeepMind and Meta researchers, the company closed a €1.7 billion Series B round in September 2025 — the largest in European AI history. And yet, or perhaps precisely because of that, Mensch speaks candidly about dependencies Europe should not be willing to accept.
What "Digital Sovereignty" Looks Like in Practice
Klingt interessant?
The term sounds abstract. The reality behind it is anything but. According to Bitkom's *Digital Sovereignty 2025* report, which surveyed 603 German companies, nearly all of them rely almost exclusively on foreign hardware, software, cloud, and security solutions. Only a small fraction offer digital products on international markets themselves.
In concrete terms: if a European company builds its customer service on a US-based AI chatbot, it's ultimately a company in San Francisco or Seattle that decides whether and how that chatbot keeps running. When public agencies have outsourced their core processes to the Microsoft ecosystem — and that is the norm, as a 2025 study by the cyberintelligence.institute shows — operational capacity depends on a single proprietary platform.
There's an especially sensitive dimension here. As Mensch noted in the interview, chatbots are becoming a new kind of media. For many people, they are the primary gateway to information. When a chatbot filters and pre-structures the information landscape — as it inevitably does — its origin and design orientation carry consequences. "Chatbots could influence their users' communication," Mensch said. "That's particularly problematic during election campaigns."
Europe's Strategic Risk — and Its Blind Spots
On November 18, 2025, a summit on European digital sovereignty took place in Berlin. The federal government, industry associations, and tech companies agreed: dependence on non-European providers in cloud, software, and AI is a strategic risk.
Agreement on the diagnosis, sobriety when looking at the numbers. The *Index Digitale Souveränität* by consulting firm adesso — drawn from nearly 500 larger companies and public institutions — puts the average maturity level for digital sovereignty at 66 percent. That sounds respectable until you read the details: nearly 80 percent of respondents have no dedicated strategy. And close to 60 percent of executives see little or no engagement from the federal government in creating the right conditions.
Even more striking: a term circulating in expert circles cuts right to the core problem — "sovereignty washing." The Centre for Digital Sovereignty (ZenDiS) published a white paper showing how vendors market themselves with phrases like "sovereign cloud" or "European data center location," while actually meeting only partial aspects of genuine sovereignty. Storing data in Europe is not the same as operational independence, control over software updates, or the ability to switch providers without friction.
What European AI Companies Are Proving Right Now
Mensch pushed back against the narrative that Europe is technologically uncompetitive in AI development. He's right. The past two years have shown what's possible.
Mistral AI has built a genuine alternative to Silicon Valley's closed systems through its open-weight models — whose weights are publicly available. This is no longer an academic exercise: in October 2025, Mistral launched an enterprise AI platform where companies can develop and run their own AI tools. Luxembourg signed a strategic agreement with Mistral in August 2025 to build government services on sovereign infrastructure.
Aleph Alpha from Heidelberg — Germany's best-known AI player — has taken a different path: explainable, GDPR-compliant models that companies and public authorities can run on their own infrastructure. Founder Jonas Andrulis stepped down as CEO in October 2025 to focus on strategy and governance. The company is going through a profound transformation — but it represents an approach that remains geopolitically relevant: AI that isn't tied to American or Chinese server farms.
According to *Fortune Business Insights*, Europe's AI market is projected to grow from roughly $65 billion today to $338 billion by 2032 — an annual growth rate of over 26 percent. The pie is large. The question is who holds the fork.
What Businesses Can Do Today
Digital sovereignty is not a matter of ideology — it's risk management. Organizations that understand this act accordingly. Four levers have proven effective in practice:
Quantify vendor lock-in. How much would it actually cost to replace a critical AI service with an alternative today? An honest answer to that question reveals the true depth of dependency. APIs, data formats, contract clauses — all of it factors in.
Deploy open-source components strategically. According to Bitkom's *Open Source Monitor 2025*, three-quarters of companies see open-source software as a key lever for digital sovereignty. Open-weight models like Mistral's make it possible to run, customize, and audit models internally.
Actively evaluate European alternatives. Not because they're inherently superior, but because market comparison creates negotiating power and keeps strategic options open. Aleph Alpha, Mistral, and others now have enterprise offerings that make sense for many use cases.
Treat GDPR and the EU AI Act as design principles, not afterthoughts. Organizations that handle compliance reactively pay more — in time, money, and lost flexibility. Those that start with compliance as a foundation build systems that are not just legal in Europe, but genuinely trustworthy.
What Policymakers Still Need to Learn
Mensch's call for digital sovereignty isn't a critique of free markets. It's a critique of missing framework conditions. Europe has created a regulatory instrument without peer in the EU AI Act — but regulation alone doesn't build infrastructure.
What's missing is consistent investment in European computing capacity. According to a 2025 Scaleway report, Europe holds only around ten percent of global AI compute capacity. The US and China dominate the hardware base on which modern AI models are trained. As long as that remains true, even the finest European model will depend on external chips and infrastructure.
The *Digital Europe Programme* doubled its budget in 2025 to strengthen local AI initiatives. France is charging ahead with state-backed support for Mistral. Germany is lagging — which is paradoxical, given that the November 2025 Berlin summit demonstrated political will without yet fully operationalizing it.
Mensch also raised the defense context in his interview: in a world where adversaries deploy AI systems as weapons, Europe needs defenses against AI-powered drone attacks. Those who don't control these systems themselves are at the mercy of others. That may sound like a geopolitical thought experiment — but it's real security architecture.
What Remains
The debate over digital sovereignty risks disappearing between two poles: on one side, flowery summit declarations with no operational follow-through; on the other, techno-nationalism that stifles competition and slows innovation.
The right path runs between them. Arthur Mensch described it clearly: the goal is not to ban American or Chinese AI models. The goal is to have a choice. To be able to flip the switch yourself when it matters. And for that, Europe — businesses, public institutions, and policymakers together — needs more than good intentions.
The talent is there. The capital exists. What's missing is the political and entrepreneurial will to act. And time is getting short: the AI market is being carved up right now. Those who don't play today will buy tomorrow at the terms set by those who did.
