The €200B gamble: Will Europe's AI sovereignty vision become reality?

What do you need for sovereignty?

Part one: This step involves enabling EU companies to run AI workloads and keep their data in the region. This is challenging, but there's a decent path to success, taking into account the following issues:

• GPUs and AI chips. There's an AI chip shortage. The suppliers of GPUs and AI chips are overwhelmingly American (or Chinese). It isn't critical that GPUs and AI chips be made in the EU to support sovereignty, but the chip suppliers will sell what they have based on supply and demand. The highest bidders will win.
• Cloud computing and data centers. The bulk of AI investments these days is in building the cloud infrastructure and data centers to meet demand, and that's where a great deal of the EU AI investment is directed. However, will there be enough EU investment to meet sovereign AI requirements? The estimated U.S. AI investment in 2025 was $1,420 per capita, while the EU was $470. However, most U.S.-based cloud compute vendors are building in-region partnerships and investing their own funds to provide cloud compute and data center resources in the EU to meet sovereign AI requirements.
• Systems and controls to manage fragmentation. The EU is made up of 27 member nations that might have further sovereign requirements or complexities when data and compute span multiple countries. The AI Act and EU umbrella theoretically ensure free flow, but we'll see if that holds. Regardless, sovereignty introduces complexity, particularly in data use.

Part two: Cultivating AI innovation and spawning homegrown AI suppliers might be interesting, particularly in terms of AI models in low-resource languages.

Most language models have been trained in English. According to Common Crawl, more than 46% of LLM training corpora is in English; roughly 6% is in each of the following languages: German, Russian, Japanese, French, Spanish and Chinese; all other languages make up the remaining 20%.

AI sovereignty in the EU could make it feasible for models based on low-resource languages, like Danish, Swedish, Italian, Polish and Dutch to sprout and possibly flourish, and to strengthen momentum for French-, Spanish- and German-based models. In theory, these specialized models will return more accurate responses for use cases in those jurisdictions. The more accurate the models are, the better the potential ROI.

A final thought

Enterprise adoption of sovereign AI in the EU will grow significantly in 2026. Because these markets are big, stand-alone, addressable markets, there's a good chance domestic AI ecosystems will gain traction and displace some U.S.- and China-based vendors.

Don't expect that with things like GPUs and CPUs, but rather in services and software -- AI models that go deep in languages other than English come to mind. To an extent, the smaller market sizes in countries like Denmark, Finland and Slovenia could offer regional vendors a better chance to thrive. When enterprises have access to better-tailored models in their language, theoretically, results will be more accurate, accelerating AI initiatives into production.

Europe's pursuit of sovereign AI is as much about strategic autonomy as it is about technological capability. The EU's massive investments in compute infrastructure, AI factories and gigafactories signal a clear intent to control its own destiny, ensuring that AI workloads, sensitive data and critical innovation stay within its borders.

Yet sovereignty is not achieved through infrastructure alone; it also depends on cultivating a vibrant ecosystem of homegrown AI suppliers and models that reflect Europe's linguistic, cultural and regulatory diversity. If the EU can successfully align its investments, governance frameworks and innovation agenda, it might not only reduce dependency on global AI superpowers but also redefine what trustworthy, inclusive and locally relevant AI looks like.

Mark Beccue is a principal analyst at Omdia, where he covers artificial intelligence.

Omdia is a division of Informa TechTarget. Its analysts have business relationships with technology vendors.