June 15, 2026
Gridlocked, but make it AI
Show HN: Can Europe train a frontier AI model on the compute it owns?
Europe says it can build its own AI now, but the comments are yelling “good luck”
TLDR: A new project says Europe could train a powerful homegrown AI sooner by pooling public supercomputers instead of waiting years for giant new facilities to get electricity. Commenters were far less convinced, mocking the plan as chatbot-grade theorizing and saying Europe’s real problem is politics, not hardware.
A bold new Show HN post tried to answer a very big question in very plain terms: does Europe really need to wait for giant new data centers, or can it stitch together the computing power it already has and train a serious AI model sooner? The repo’s answer is an emphatic yes — at least as a temporary fix. The pitch is that Europe already owns lots of public supercomputers, and because huge new AI campuses can take years to get connected to the power grid, a patchwork “team effort” could beat a shiny new mega-site by about five years.
But the real fireworks were in the comments, where the community instantly split into camps of skeptics, cynics, and drive-by comedians. One camp basically said, “Cute spreadsheet, but have you met European politics?” The harshest doubters argued that if major countries in Europe struggle to cooperate on military hardware during a geopolitical crisis, then getting them to unite around one giant AI project sounds like fantasy football for policy nerds. Another crowd was even more brutal, dismissing the entire repo with variations of “thanks ChatGPT” and “why not just post the chatbot transcript?” Ouch.
Then came the practical hot take: why train anything at all? One commenter argued Europe should just copy or shrink an existing top model and call it “sovereign AI” immediately. So the drama wasn’t really about math — it was about trust, politics, and whether this is a serious plan or just very polished cope.
Key Points
- •EuroMesh argues that Europe could use existing public compute as a stopgap to train a sovereign frontier-class AI model instead of waiting for new 1 GW datacenters.
- •The article states that Europe already has tens of exaflops of public AI compute across EuroHPC supercomputers and 19 AI Factories.
- •Its model estimates that federated low-communication training could enable a frontier-class model around 2028, versus around 2033 for a newly built gigawatt campus.
- •The model has three layers covering training-efficiency penalty, time-to-availability of compute, and regional scoring on time, cost, carbon, and feasibility.
- •The repository is presented as reproducible and transparent, but the article notes caveats including shared access constraints, heterogeneous systems, unproven training at larger scales, and non-peer-reviewed status.