Musk Commits SpaceX AI Data Centers to NVIDIA Vera Rubin, Targets 10GW by 2027
rohanpaul_ai · x · 2026-08-06
Elon Musk announced that SpaceX's expanding AI data centers will exclusively use NVIDIA's upcoming Vera Rubin chips, with plans to reach nearly 10GW of AI compute capacity by the end of 2027.
SpaceX reportedly spent around $15.8 billion on AI infrastructure in Q2, claiming that new compute deployments currently repay their capital costs within a year. NVIDIA stated that during Kimi-K2-Thinking tests, the NVL72 architecture delivers up to 10 times more tokens per megawatt compared to GB200 NVL72, at one-tenth the cost. Additionally, SpaceX plans to launch "Starmind," a simplified, space-optimized NVL72 computer, starting next year.
More from Companies & People
- Cambridge CAIRD free one-day AI research workshop opens applications, due Aug 30 — StephenLCasper · 2026-08-25
- KubeCon China reveals first keynote speakers lineup — PyTorch · 2026-08-25
- When and how will specialization in model APIs create platform splits like Apple/Samsung? — ethanniser · 2026-08-25
- Reflecting on late 2010s: OpenAI and Neuralink as glimpses of the future — flowersslop · 2026-08-25
- Situational Awareness AI hedge fund probed by SEC — TechCrunch AI · 2026-08-25
- ResNet Co-author Ren Shaoqing Launches Embodied AI Startup, Nio to Invest — 量子位 · 2026-08-25