1GW of training ≈ 400-500k B300s, and frontier labs don't need 3,000 staff
eliebakouch · x · 2026-09-14
eliebakouch dissects a frontier-lab wishlist: Anthropic/OpenAI rent heavily, with Nebius-like EU players or Nvidia-backed firms as first-line compute suppliers. He estimates 1GW-year ≈ 400-500k B300 GPUs — likely more than Astra was trained on — suggests training/inference revenue-share deals with hyperscalers, and argues 3,000 staff is far more than needed to build a frontier lab.
Related event: Researcher Breaks Down Frontier Labs' Compute and Staffing Math(2 posts)→
More from Companies & People
- Honeycomb co-founder: using AI for emails and perf reviews is lazy and offensive — rseroter · 2026-09-14
- Oracle Cuts Double-Digit % of Some Teams While Hiring Aggressively for Data Centers and AI — mkheck · 2026-09-14
- Musk: AI will be 99% of SpaceX's value within four to five years — XFreeze · 2026-09-14
- Zuckerberg reportedly confirmed open-weights commitment twice in a month — qinzytech · 2026-09-14
- LMArena hiring engineers at up to $350K base as eval platform booms — jfiance · 2026-09-14
- Boltz partners with dsm-firmenich to design scent and flavor molecules with AI — GabriCorso · 2026-09-14