Can Europe train a frontier AI model on the compute it owns?

EuroMesh is an open-source effort trying to answer a concrete question: can you aggregate the GPU and HPC compute that European institutions and cloud providers actually own today and use it to train a frontier-scale model. The repo documents the methodology, the inventory estimates, and the distributed training architecture questions that would need to be solved. It is not a product you deploy this weekend. It is closer to a research artifact and a conversation starter about compute sovereignty, which is a real policy and engineering question as AI regulation tightens in Europe. The audience for this is narrow but specific: ML researchers thinking about federated or distributed training at scale, and technical founders who operate in regulated European markets and have a stake in how the infrastructure layer develops. If neither of those describes you, file it as background reading. -> Best for: ML researcher or AI engineer working on distributed training or European compute infrastructure