Show HN: Can Europe train a frontier AI model on the compute it owns?
Summary
The Show HN post outlines a EuroMesh repository that analyzes whether Europe can train a frontier AI model using sovereign public compute. It argues that federated, low-communication training across existing exaflop-scale resources could yield a frontier-class model around 2028, despite grid-power constraints and long lead times for large gigawatt campuses. The work is a sourced, reproducible analysis rather than peer-reviewed, with data and code to rebuild the results.