European sovereignty in the AI world. The French start-up Mistral Ai uses the dominance of the USA
The most tangible effect of this change in mood is the phenomenon of the French start-up Mistral AI. The company founded only two years ago forecasts over $ 100 million. annual revenues and does not hide that It largely owes it to customers who want to become independent of the Silicon Valley.
To meet the demand, Mistral AI is building its own 40 MW data center near Paris, equipped with 18 thousand. The latest Nvidia Grace Blackwell systems, with an expansion plan up to 100 MW within eighteen months.
The project is politically supported by President Emmanuel Macron, who encouraged French concerns to book computing power. This is how the first “AI factory” is to be created fully controlled by a European entity.
Mistrala's technology itself develops as quickly as its infrastructure. In June, the company presented two AI models with a built -in reasoning function, developed from scratch by the reinforcement learning method (the model learns by making decisions and receiving prizes or penalties, which allows it to optimize its activities in time to maximize profits). According to internal benchmarks, the Magistral Medium matches, and in places he exceeds, commercial models OPENAI and Google. In turn, available in the Open Source Magistral Small formula is now available on the Hugging Face platform.
Public Code sharing strategy is to attract corporations that want full control over the place of implementation and tuning models. This is an aspect that closed systems from behind the Atlantic do not guarantee.
Check also: In two years, what you do on your computer will do faster and cheaper. What is Agi?
Europe has bold goals
The growing popularity of the French start-up is part of the wider ambitions of the continent. In 2024, the European Commission adopted an act of artificial intelligence, the first comprehensive set of principles for this technology that It will be fully used in mid -2026.
The regulation is to guarantee transparency and security, but at the same time emphasizes the need for “strategic abilities” in terms of computing power and access to models. According to IDC forecasts, expenditure on AI in Europe will reach $ 144 billion in 2028, growing by an average of 30 percent annually. It is a scale that justifies the creation of native infrastructure.
In the context of sovereignty, however, not only Mistral counts. Amazon Web Services announced an investment of EUR 7.8 billion in the European Sovereign Cloud. It is about a separate cloud region built in Brandenburg, Supported only by EU staff and physically separated from the global AWS network.
EUROHPC-ju projects are followed by similar logic, which finance supercomputers from Lumi in Finland to Meluxina in Luxembourg, and the pilot “AI Factors” financed on the initiative of Investai.
But do Europeans rightly fear the technological hegemony of the USA? Is this justified? Supporters of sovereignty indicate the risk of geopolitical pressure – access to cloud models can be limited with one button, and the interruption of chip supply chain would immediately stop the development of innovation.
GDPR requirements and the upcoming AI ACT regulations additionally reward local data processing. The open, configurable architecture of Mistrala also allows you to shorten the algorithm audit path and easier to prove compliance with the law.
Critics, however, counter that Europe remains dependent on American components anyway – just look at the NVIDIA logo at each of 18 thousand. GPU in the center of Paris. In addition, the lower availability of high risk capital hinders the scaling of companies such as Mistral. Carnegie Endowment points out that too restrictive regulation may discourage investors and deepen the market fragmentation, even if the intention is to protect European values.
Don't miss: AI is a test for the management. The average cost reduction will be the average
More data centers will be created
In practice, the most important will be the pace at which European entities will expand the hardware and talent layer. The first stage of Mistral Compute is to be launched as early as 2026, which It can increase the total computing power of AI in the EU by up to 10 percent.
If the project led to a reduction in the cost of training models on the Old Continent, this will convince the next generation of start-ups. And this is what the lack of scale is considered to be the largest brake of European ambitions today.
The rational strategy for Europe is diversification, i.e. specifically the construction of its own centers and models AI, where control over critical data counts, while using the global ecosystem of hardware and open software. Mistral Ai shows that such a hybrid model can be not only politically desirable, but also commercially profitable.
The next two years will decide whether the French AI Factory will become the cornerstone of European technological independence, or only a symbol of a well -diagnosed, but still unresponsive problem. The problem of dependence on the US innovation.
Author: Grzegorz Kubera, Business Insider Polska journalist
