Report author: US intervention could make EU frontier parity 'almost prohibitively costly'

anton_d_leicht · x · 2026-09-15

Anton Leicht defends the large compute figures in his European AI report: it targets actual frontier parity with OpenAI/Anthropic's internal models, not being a few months behind. Core assumption: US intervention would cut off rented hyperscaler compute beyond some point. The numbers are high because the plan must very likely succeed — otherwise Europe ends up having antagonized the US without a frontier model. He concedes frontier parity is almost prohibitively costly, and that a months-behind goal would need far less compute.

Related event: Researchers Clash Over EU 14GW Compute Plan: How Much Is Really Needed to Reach the Frontier(13 posts)→

Original post →

More from Infra

Infra channel →