OpenEuroLLM consortium kickoff

OpenEuroLLM is a 20-partner European consortium coordinated by Jan Hajic (Charles University) and co-led by Peter Sarlin (AMD Silo AI), funded under the EU Digital Europe Programme with an allocation of roughly €55M (about $60M). The consortium commenced work on February 1, 2025, with partners spanning European research institutions, companies, and EuroHPC supercomputing centres. The stated goal is an open multilingual LLM covering all official EU languages, on the theory that US frontier models have uneven coverage outside English and that European data-protection law constrains commercial reliance on US-hosted inference.

In December 2025, OpenEuroLLM became the first AI project to be granted strategic access to multiple EuroHPC supercomputers simultaneously, with 10 million GPU hours allocated across LUMI (Finland), Leonardo (Italy), Jupiter (Germany), and MareNostrum 5 (Spain). The training plan calls for an 8B model by summer 2026 followed by a larger model using the strategic compute allocation. The related EuroLLM initiative is targeting eventual model sizes at 7B, 30B, 65B, and 200B parameters plus a distilled 7B model, though OpenEuroLLM's own size targets beyond 8B have not been firmly committed.

The consortium sits inside a broader €307M EU AI investment package announced in early 2025 and is the largest publicly funded open-model effort in Europe. The partner list mixes academic labs (Tübingen AI Center, University of Edinburgh, Charles University) with industrial participants (AMD Silo AI) and the EuroHPC centres that supply the compute, the same hybrid structure used by US lab-and-university collaborations on open models.

The sovereignty argument is explicit: the consortium frames the work as European AI infrastructure that does not depend on US clouds for training or inference. The practical test will be whether the released models reach competitive quality at competitive sizes; the 8B summer 2026 milestone is the first checkpoint that will reveal whether the consortium model can match the pace of US single-lab releases.