Musk targets 13GW of AI compute by 2027, plans 440,000-GPU Vera Rubin training cluster
rickasaurus · x · 2026-10-21
On the earnings call, Musk laid out SpaceX's AI compute expansion: 2.5-3GW online by end of 2026, plus another 10GW in 2027 (at least 13GW cumulative by end of next year). Deployment runs at roughly 200,000 NVIDIA GB300 GPUs per month for the rest of 2026, shifting mainly to Vera Rubin. A single 440,000-GPU Vera Rubin training cluster is targeted for end of Q2 2027, backed by an additional 20GW of power and cooling. He called >20GW of 'superintelligence' compute next year a non-zero possibility. On the merger, he stressed it is far from certain and would need 50% of all Tesla shareholders.
Related event: Musk maps out SpaceX AI compute roadmap: 13GW by 2027, $500B ARR target(6 posts)→
More from Companies & People
- Un_prompted summit: 200+ founders and CMOs converge on 'agents are the new customer' — amasad · 2026-10-21
- Replit opens Tokyo office, teams with Microsoft and Endeavor Japan for launch events — amasad · 2026-10-21
- NVIDIA partners with EBU and Adobe on sovereign AI for broadcasters and faster video search — NVIDIAAI · 2026-10-21
- Aiden Bai on 10 months of dead side projects: agents breed shiny object syndrome — aidenybai · 2026-10-21
- Nine Papua New Guinea schools have students make one-day AI short films, then a festival — lmoroney · 2026-10-21
- Gary Marcus mocks Anthropic: wearing a fancy suit is the best way to bypass its biosecurity guardrails — GaryMarcus · 2026-10-21