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.
More from Companies & People
- Proton CTO's leaked AI manifesto sparks employee revolt, CEO stands firm — garrytan · 2026-10-30
- Vercel founder Rauch: trust is the deepest moat for a post-AGI business — soleio · 2026-10-30
- Cyberiad Labs partners with DevCareer to train Nigeria's developers in AI, 250 free spots — jbarbier · 2026-10-30
- How Oxide Computer uses LLMs internally — FAQ Friday #66 — Sethwinterroth · 2026-10-30
- NODES 2026 goes free and virtual Nov 12 with 100+ sessions and Ng-Eifrem fireside — JeremyCMorgan · 2026-10-30
- ELLIS x UniReps talk: why distinct learning processes yield similar representations — ClementineDomi6 · 2026-10-30