Pooling Europe's Compute: The Promise of Distributed Training for European Frontier AI

This RAND Europe report assesses whether distributed training — techniques allowing geographically separated datacentres to function as a single training resource — can help close Europe's frontier AI compute gap, based on a literature review, expert consultations, and quantitative estimation of compute and power requirements. Key finding: distributed training eases two of Europe's three structural bottlenecks (power and political coordination) but does not add chips — as of early 2026 Europe has approximately 123,000 H100-equivalents of operational capacity versus roughly 1.4 million in the United States, growing to 3.2 million versus at least 19.4 million by 2030 under current plans. The report states distributed training changes 'whether this stock can be pooled into a single resource large enough for frontier training' rather than remaining fragmented across individually sub-scale clusters. It recommends mapping/preparing physical infrastructure (grid capacity assessments with ENTSO-E/DG ENER, cross-border fibre assessment with DG CONNECT, a European AI Infrastructure Reference) and simplifying cross-border regulation.

For EU policymakers and enterprises assessing AI sovereignty strategy, this quantifies the scale of Europe's compute deficit versus the US and offers a concrete near-term policy lever (distributed training infrastructure and regulatory harmonization) short of matching US/China capital investment.

Policy leads should evaluate the report's infrastructure-readiness recommendations (grid/fibre mapping, EuroHPC coordination) against national AI sovereignty roadmaps.