When the United States Claude Mythos blocks people from outside the USsends a clear message: it is a cutting-edge export technology subject to control, like chips are. And in that scenario, Europe is practically an observer: the old continent is a pioneer in legislating AIbut its infrastructure, business ecosystem around it, and LLM models are behind what the United States or China have. Simply put, Europe depends on third parties for the best AI.
But that can change. A little less than a month ago, Arthur Mensch appeared in May 2026 before the French National Assembly with another very clear message: if Europe wants to stop being an observer and descend into the mud, it has to do it now. Time is running against them. The deadline given by the co-founder and CEO of the main European artificial intelligence company is short: two years.
The warning from the CEO of Mistral AI. In his exhibition, Arthur Mensch gave a warning macroeconomic with concrete figures: Europe has approximately two years to build its own AI infrastructure, or it will be structurally subordinated to American technology companies. If it does not arrive in time, Europe will become “a vassal state.”
The future if we don’t achieve this is dark: “Once the supply is monopolized by American companies, we will suddenly run out of supply and will no longer be able to transform electrons into tokens.” His argument is technical but with a political background: whoever controls the calculation controls the economy. AI is not just another digital service: it is the infrastructure on which everything else will work. Like electricity or roads, but privatized and in foreign hands.
Why is it important. Because under this approach, being dependent on third-party AI is not only a mere technical issue, which is no small thing considering its use in critical sectors such as defense or banking, it is also a serious problem of productive sovereignty and balance of payments. The CEO of Mistral refers to AI as a strategic asset, in the same way that gas is. Europe had a hard time understanding the cost of its energy dependence on Russia and Mensch’s argument is that the old continent is making the same mistake.
In statements to CNBCdelved into the impact on the macroeconomy: “You cannot afford a trade deficit of one trillion if you really want to remain competitive in the race” because every euro that Europe pays to US companies for AI services is financing the competitor’s R&D. And that money is not coming back. Furthermore, we have already seen that using AI will be increasingly expensive, so much so that There are companies like Uber or Microsoft cutting licenses. Imagine if what depends on AI is your safety.
Context. The game board shows that Europe does not exactly start with a good hand: according to data from Epoch AI, collected both by the US Federal Reserve as by RAND Europethe United States controls 74% of global high-level computing for AI, China 14% and the EU just 4.8%. He Draghi report of September 2024 has already identified that much of the blame for Europe’s productivity gap with respect to the US lies in the technology sector, or rather, the absence of it. One year later, Draghi himself was pessimistic: barely had been fulfilled 11.2% of almost 400 recommendations.
In detail. Europe has already started with the plan action plan called “AI Continent” which is committed to tripling the capacity of data centers and deploying up to five gigafactories, but the question is whether it will be enough and if it will arrive on time. Without going any further, the 500,000 chips in these gigafactories planned are very far from what there is in the US: by the end of 2025, OpenAI I had already planned exceed one million chips.
Mensch did not stop at warnings, but made concrete proposals. The first of them: use public procurement as a lever. Given that 50% of European GDP is generated through public spending, it is clear that it is a magnificent instrument to catalyze this development. On the other hand, Mistral is exploring the development of its own chips and has already announced a new data center in France.
Yes, but. The main argument against Mensch’s words is obvious: he is one of the major stakeholders in the policies he proposes. Mistral has 1,000 employees, a valuation of 12 billion euros, a target of 1 billion euros in revenue by the end of 2026, this year it has invested 1 billion in R&D and approximately 75% of its sales are in Europe. On the other hand and for the moment, not having your own servers does not mean not being able to use AI, of course, being clear that it is a third party who dictates the conditions, prices and limits of that access.
In Xataka | Europe wanted to set an example to the world with its AI Law. What you are achieving is becoming evident
Cover | Flickr and Wikimedia Commons / ALEXANDRE LALLEMAND | Igor Omilaev | Markus Spiske
GIPHY App Key not set. Please check settings