Sovereignty claim in AI
The sovereignty claim in AI is the demand by a state or alliance of states to be able to build, operate, and control artificial intelligence itself, rather than depending on foreign providers. Behind this lie concerns about security, the economy, and the question of whose rules apply to a technology.
Almost all well-known AI systems, that is, programs that learn patterns from vast amounts of data, today originate from the USA or China. They run in data centers owned by these countries and are subject to those countries' laws. European governments and companies increasingly find this uncomfortable. The sovereignty claim in AI is the political demand to reduce this dependency. What is meant is: a country or an alliance of states wants, if in doubt, to produce the technology itself, operate it itself, and decide on its own rules. Sovereignty here does not mean isolation, but the ability to carry on without foreign permission in an emergency.
What is at stake when the provider is based abroad
The first reason is quite practical: dependency is expensive and risky. If a hospital, a government agency, or a car manufacturer builds its entire software on a single foreign service, it can hardly negotiate prices later on. If the service is shut down or blocked by sanctions, operations come to a standstill. Experts call this the lock-in effect: you can no longer get out of a system because switching would be too costly.
The second reason concerns data and law. American providers are subject to US law, even if their servers are located in Frankfurt. Under certain conditions, US authorities can demand access to data. For patient records, tax data, or a machine manufacturer’s construction plans, this is a serious problem. The European General Data Protection Regulation and the US access claim do not fit together seamlessly.
The third reason is economic. Whoever merely buys the technology does not profit from it. The profits, jobs, and expertise then arise elsewhere. That is why the sovereignty claim is always also industrial policy, not merely security thinking.
The four layers: chips, data centers, models, data
Sovereignty cannot be established at a single point. One can imagine the AI landscape as a stack of four layers. At the very bottom lie the chips, that is, the specialized processors on which AI computes. Above that lie the data centers, then the models themselves, and at the very top the data and applications. Whoever controls only one layer is not yet sovereign.
The bottom layer is the most difficult. The most powerful AI chips practically all come from the US company Nvidia, manufactured in Taiwan. Building up one’s own chip production costs tens of billions and takes years. This is exactly what the European Chips Act is about. The situation is simpler with data centers: supercomputers such as Jupiter in Jülich are located on European soil and are publicly funded.
At the model level, open software is an important lever. If a model, including its learned parameters, is freely available, anyone can download it and run it on their own machines. The provider then cannot shut it down. The French company Mistral has released several such models. A common misconception, however, is that open models are automatically sovereign: they were often nevertheless trained on American chips.
From Gaia-X to the Bundeswehr: where the claim appears
In the news, the term usually appears in connection with funding programs and major projects. Gaia-X was the attempt to create a European cloud infrastructure and is regarded by many as an example of how arduous such undertakings are. The EU’s AI Act regulates what AI systems are allowed to do in Europe, and is thus a form of sovereignty over rules rather than over technology. The term Sovereign AI, popularized by Nvidia CEO Jensen Huang, also belongs here, albeit with a sales interest attached.
For companies, the question is very concrete. Government agencies, banks, and hospitals are increasingly specifying in tenders that data must not leave Europe. Providers respond with so-called sovereign cloud offerings, in which European partners take over operations. How sovereign this really is, is being hotly debated.
It is important to distinguish this from two similar terms. Data protection regulates what may be done with personal data. Data sovereignty means control over one’s own data. The sovereignty claim in AI is broader: it concerns control over the entire technology, from hardware to application. Critics consider the goal expensive and partly unrealistic, because no country can master all layers on its own.