Most of the generative AI you use every day — the models drafting your emails, transcribing your meetings, generating your images — was trained and is run by three or four American companies, and a fast-growing set of Chinese ones. This site tracks the other option: the models, companies and public infrastructure being built in Europe, openly enough that anyone can actually check what's inside.
"Sovereignty" in AI usually gets framed as a compliance question: keep the data in the EU, satisfy GDPR, tick the boxes. That's real, but it's the smaller half of the problem. The European Commission's own AI alliance has put the sharper point plainly: it isn't enough to build secure data centres if the models running on them were trained somewhere else. Downloading someone else's open-weight model and hosting it in Frankfurt is not the same as being able to train, maintain and improve models yourselves.
Infrastructure without capability just means Europe pays for the racks while someone else keeps the knowledge. That's the gap this site tracks — not just where you can legally run a model, but who in Europe can actually build one, and what's changing.
The European stack page is a running list of the model labs, specialist companies and public compute behind them — grouped so you can see where the gaps actually are.