Europe does not need to win the chatbot race, but the autonomy one: Domyn's bet

In the debate over European AI, defeatism usually prevails: without the trillion-dollar data centers of the U.S. giants, what can Europe do? The plan that Domyn presented to Reuters on June 25, 2026, proposes changing the question. The Milan-based company—formerly iGenius, founded in 2016—announces, within a year at most, a fully open and reproducible frontier model, with more than 400 billion parameters and trained from scratch.

The most fruitful argument comes from its CEO, Uljan Sharka, who describes the EuroHPC network as an «undervalued strategic asset.» The technical distinction he defends is important and often overlooked: training a model, though intensive, is a process bounded in time; serving inferences at planetary scale is what requires massive, permanent infrastructure. If one accepts that premise, Europe would already have enough computing capacity to create foundational models, though not to run the next global chatbot. And that is precisely a different and more attainable goal.

The choice of open source and reproducibility is not ideological but geopolitically surgical. The project, articulated within the EUROPA consortium alongside Germany's Fraunhofer and selected in the European Commission's Frontier AI Grand Challenge, responds to a double vulnerability that the report itself illustrates well: Italy and the Czech Republic have restricted the remote use of DeepSeek but allow local deployments, and at the same time European concerns are growing over U.S. export controls on Anthropic's models. Dependence, in short, is not only on Beijing; it is also on Washington. A model that can run on one's own infrastructure is not subject to the regulatory decisions of any third country.

It is worth maintaining rigor where the article itself maintains it: size in parameters does not guarantee matching the frontier leaders, and Domyn has not detailed its financing, although it confirms the backing of Abu Dhabi's G42 and investors such as Eurizon. Declared ambition and demonstrated capability are different things, and a one-year timeline is demanding.

Even with those caveats, the proposal fits a diagnosis shared by other players on the continent, such as Mistral or OVHcloud—whose head recently spoke of a «second wave» of model builders driven by falling costs. Domyn's real contribution may not be the model itself, but the redefinition of the goal: sovereignty does not consist of cloning ChatGPT, but of ensuring that governments, companies and researchers can deploy high-level AI without asking anyone's permission. If Europe understands that its advantage lies in autonomy and not in size, this may be one of the most sensible bets of the year.