Europe and Its Fight for Digital Sovereignty

Amid the global whirlwind unleashed by generative Artificial Intelligence, giants in computing, data, resources, and human capital are emerging and driving the major transformation we are witnessing. Alongside them come challenges, opportunities, and threats that directly test Europe’s ability to remain at the forefront of the economic, political, and technological disruption shaping our era.

Large Language Models (LLMs) have created a new geopolitical battlefield in the world: data and compute.

Data and Our Sovereignty

For the past 20 or 30 years, we have repeated in the technology and business world that data is the oil of the 21st century. What we did not imagine is that data could be, or already is, the essential fuel for an economic and technological revolution that makes the Industrial Revolution look like a mere footnote compared with the consequences we are seeing today, consequences whose true scope we are still simply unable to measure.

New generative AI models require massive amounts of data. The more and the better the data we have, the better the models we can build. But access to that data is the key ingredient for creating models that are more specialized, personalized, and capable. That is why data sovereignty is becoming increasingly fundamental. Our data is becoming more valuable by the day, and the data on which models have not yet been trained is even more valuable. Every sector holds business data containing knowledge, trends, conclusions, and behaviors so unique that they become not only a core strength for each company, but also a new risk that must be addressed.

Our data must remain our data. If we want true economic, business, and technological sovereignty, we must commit not only to storing that data in datacenters located in Europe, but also to building the infrastructure, investment, and European companies that align with the continent’s geopolitical interests.

Imagine an industrial company that trains and specializes a model on its own highly specific operations: that is the company using its competitive advantage to strengthen its market position. If that data ends up in a commercially available model, the company’s competitive advantage becomes diluted, because its know-how can then be leveraged by any competitor. And even if agreements are in place to prevent your data from being used for training, the current reality is that we remain at the mercy of foreign infrastructure to guarantee that security. As bitcoiners say: “Not your keys, not your coins.”

Europe’s regulatory framework is the most ambitious in the world: GDPR has generated more than €5.88 billion in fines since 2018, including landmark penalties such as €1.2 billion against Meta for unlawful transfers of data to the U.S., or €530 million against TikTok in 2025 for sending European user data to China. But rules alone do not build sovereignty. Seventy percent of the European cloud market is controlled by AWS, Microsoft Azure, and Google Cloud, while European providers account for only around 15% combined of their own market.

Compute Capacity: The Modern Steam Engine

We need GPUs. We need GPUs. WE NEED GPUs.

We need energy. We need energy. WE NEED ENERGY.

This is the major bottleneck the sector faces in developing better models, better tools, and the ability to run them. Right now, there is a fragile balance that allows the largest labs and companies to operate. Historic compute agreements are being signed, such as the one by Mistral AI, which in 2026 secured €830 million in debt financing from seven banks to acquire approximately 13,800 NVIDIA GB300 chips and build its own sovereign datacenter in Bruyères-le-Châtel, with 44 MW of capacity scalable to 100 MW, yet the industry still complains about chip shortages. Sam Altman, CEO of OpenAI, has publicly stated on multiple occasions that the lack of compute capacity is the main constraint on the development of his models, and Elon Musk’s xAI director described it in 2025 as “the scarcest resource on the planet.”

So where is Europe in this race?

The gap is brutal. The United States accounts for 81% of global private AI investment—$109.1 billion in 2024 alone—and has three exaflop-scale supercomputers (El Capitan, Frontier, Aurora). China, despite U.S. restrictions on NVIDIA chips, operates at least two undeclared exascale systems and is projected to control 22% of global semiconductor manufacturing by 2030. Europe, for its part, holds only 5% of global AI compute capacity, and has seen its biggest semiconductor bet—the Intel megafactory in Magdeburg, planned with €30 billion in investment—cancelled in July 2025.

The positive note comes from two fronts. In supercomputing, JUPITER (Jülich, Germany) became in November 2025 the first European exaflop-scale supercomputer, ranking #4 worldwide in the TOP500. It is joined by a network of 11 supercomputers under the EuroHPC umbrella, backed by a budget of approximately €7 billion for the 2021–2027 period. In semiconductors, the ESMC plant in Dresden, a joint venture between TSMC, Bosch, Infineon, and NXP, with more than €10 billion in investment, is progressing as planned toward production in 2027, becoming the continent’s first foundry with FinFET capability.

But the numbers do not lie: the six leading global hyperscalers are approaching $500 billion in capital expenditure worldwide in 2026. Microsoft alone committed $80 billion in 2025. OpenAI’s Stargate project anticipates $500 billion over four years. By comparison, Europe’s public initiative InvestAI is mobilizing €200 billion in public-private funding, including €20 billion for 4–5 AI Gigafactories, figures that, on an individual basis, are already surpassed by the investment of a single American private company.

Europe must invest in large-scale datacenters that allow it to manage its compute capacity in a sovereign way. This is not only an economic issue—it is a geopolitical one. Compute must be managed as a strategic resource, not only at the public level, but also in the private sector.

Arthur Mensch, CEO of Mistral AI: The Clearest Voice on Europe’s Challenge

No one articulates the urgency of this situation better than Arthur Mensch, CEO of Mistral AI, the French startup founded in 2023 that has become Europe’s leading bet in the LLM space. With a valuation of approximately €13.7 billion and around €300 million in annual recurring revenue, Mensch is not speaking in theory.

At Davos 2025, he was direct: “There are many countries looking for an alternative to U.S. providers and Chinese providers, and we provide that advantage.” And when speculation arose about a potential sale of the company, he made it clear that Mistral is not for sale and that the plan is an IPO.

At VivaTech 2025, when introducing Mistral Compute—its sovereign AI infrastructure—he raised the stakes: “We will offer a fully independent platform. Users will no longer need to depend on certain U.S. providers.” And he explained why: “It is essential that we can master and adapt these systems to our values” and, more importantly, “it is unthinkable that foreign actors should hold the keys to a system that we will deploy in critical infrastructure; this is a matter of national security.”

His strongest message came at the India AI Impact Summit in February 2026: “We are at risk today in the world. We do not want to live in a world where three or four massive companies truly own the deployment and creation of AI, own access to information, and own the infrastructure itself.” Mensch argued that every government, hospital, and public institution needs its own “on/off switch”, without depending on external providers.

On competition with China, he dared to challenge the Western consensus at Davos 2026, calling it a “fairy tale” to believe that Chinese technology lags behind the West: “The capabilities of Chinese open-source technology are probably creating stress among CEOs in the U.S.”

And regarding regulation, his evolution has been striking: after criticizing the AI Act in 2023 as “the worst possible taxonomy,” by 2025 he described it as “manageable,” accusing major U.S. technology companies of using it as a barrier to entry: “What we are seeing is a push for regulation from the very same people who have an interest in making sure no one else can enter the market.”

Conclusion

Europe has the foundations to compete: the most advanced regulatory framework in the world, top-tier talent, unique strategic assets such as ASML, whose monopoly in EUV lithography has no global equivalent, and growing political will. But the structural gap is deep: 5% of global AI compute, 15% of its own cloud market, total dependence on AI accelerators manufactured in the U.S. and Taiwan, and private investment that is only one-tenth of that of the United States.

Europe’s realistic path is not to outscale the U.S. and China in absolute terms, but to compete through efficiency, sovereignty, regulatory trust, and vertical specialization, serving the growing market of companies and governments that demand AI solutions aligned with European regulation and free from foreign jurisdictional risk. In that niche, sovereignty is not a limitation: it is the value proposition.