SpaceX commits exclusively to NVIDIA Vera Rubin GPUs; plans space-based AI cluster next year
SpaceX and xAI will use NVIDIA GPUs exclusively, with Elon Musk praising the Vera Rubin NVL72 architecture as the best AI computer available. Musk announced on SpaceX's earnings call that the company plans 2 GW of compute capacity by year-end and 10 GW by end of 2027, deployed both on the ground and in orbit via a "Starmind" AI satellite program launching next year.
The NVL72 is NVIDIA's flagship rack-scale system featuring 72 Rubin GPUs and 36 Vera CPUs in a single liquid-cooled rack with fanless, tubeless, cableless design. Musk cited the cable-free compute trays as eliminating a major source of installation errors and failures, and improving both serviceability and assembly speed.
SpaceX appears poised to become a significant NVIDIA customer — Musk indicated SpaceX will receive a substantial percentage of NVIDIA's Vera Rubin GPU allocation next year. The move cements NVIDIA's dominance in frontier AI compute while signaling that orbital data centers are shifting from research concept to near-term commercial deployment under space launch constraints.
For architects planning large-scale AI infrastructure: SpaceX's exclusive adoption of Vera Rubin validates the platform for production workloads and suggests orbital compute is becoming a serious capacity play, not a distant concept. Vera Rubin's 5x inference performance over Blackwell and sub-$0.10 per token costs make it the reference architecture for 200B+ parameter deployments in 2026–2027.