Italian AI startup Domyn has announced plans to release a fully open-source, reproducible large language model with over 400 billion parameters within a year. The company, in partnership with Germany's Fraunhofer Institute, has formed the EUROPA Alliance, which was selected for the European Commission's Frontier AI Grand Challenge and given access to the EuroHPC supercomputing network.
The core motivation? Reducing Europe's dependence on foreign-hosted AI systems. Open-source AI today is largely dominated by Chinese companies, while leading US models are mostly closed-source and require remote access. With fears over US export controls on models like Anthropic's, and restrictions imposed by Italy and the Czech Republic on DeepSeek's remote services, European businesses and governments are increasingly turning to locally deployable, free alternatives.
Training data will be collected from institutional partners, with initial government data-sharing agreements expected within weeks of meetings with several European heads of state. As for compute, Domyn CEO Uljan Sharka notes that training a frontier model from scratch requires far less power than serving inference to hundreds of millions of users — and Europe's public supercomputing network is already up to the task. The project has secured backing from Abu Dhabi's G42, Eurizon Capital, Rabobank, and BNY Mellon.