Show HN: Can Europe train a frontier AI model on the compute it owns? (github.com)

A recent report explores the feasibility of European countries training a sovereign frontier-class AI model using the extensive public compute resources they already possess, rather than waiting several years for new gigawatt datacenters to come online. The model suggests that by leveraging already operational EuroHPC supercomputers and national AI Factories, Europe could create a competitive AI model by around 2028. This approach utilizes a federated training process with low-communication strategies, significantly reducing the time to availability compared to the longer grid connection waits for new campuses. This initiative is noteworthy as it underscores Europe's capability to become a significant player in the AI landscape without being reliant on external entities like OpenAI or Anthropic. The report emphasizes that while the training efficiency of the current federated compute might be less than optimal, the urgency of deploying AI systems sooner rather than later positions this approach as a potential game-changer. Furthermore, it raises critical discussions around the accessibility of computing power and political decisions that could influence AI advancements across the continent.