The post is a short GitHub memo asking whether Europe could train a cutting-edge AI model using compute it already controls, mainly by stitching together public supercomputers, national clusters, and cloud capacity instead of relying on a single hyperscaler. The basic claim is not that Europe is ahead, but that the hardware gap may be less absolute than it looks if fragmented resources can be coordinated.
Most people accepted the narrow point that Europe probably has more compute than the usual doom narrative suggests. They rejected the idea that this solves the real problem. The repeated answer was that frontier AI is now a coordination and capital problem far more than a spreadsheet-of-GPUs problem. You need an institution that can raise tens of billions, secure power, move fast across borders, hire aggressively with meaningful
equity, build a product, and keep training over multiple generations. Europe looks weak on almost every one of those dimensions. Several comments pointed to the deeper pattern: Europe still produces top researchers, but too many of them do frontier work inside US companies, and the best European successes often end up owned by American capital anyway.
That pushed the conversation away from the repo’s federated-compute idea and toward sovereignty. People worried less about whether a one-off training run is physically possible and more about what happens if US labs or the US government restrict access to top models,
weights, or services. In that framing,
Mistral and
DeepL were treated as useful but not enough. Mistral in particular was described as commercially pivoting toward enterprise tooling and custom models rather than truly chasing the frontier, which many took as evidence that Europe lacks the funding and market structure to stay in the top tier.
The sharpest divide was over whether Europe should even want to win this race. One camp said regulation, privacy protections, and slower adoption are features, not bugs, and that Europe should avoid copying the US model of growth at any cost. The dominant view was harsher. It held that Europe is confusing guardrails with strategy, regulating around technology it does not control, and risking a familiar outcome where it keeps its values on paper while becoming dependent on foreign systems in practice. Even people sympathetic to privacy and labor protections tended to land on the same operational conclusion: Europe can preserve those values only if it also builds enough industrial and AI capacity to avoid outsourcing the whole stack.