Via notateslaapp.com
The space-optimized NVL72 system will bring rack-scale AI computing to low Earth orbit, marking a new frontier for high-performance inference beyond the atmosphere
SpaceX and Nvidia are sending an AI supercomputer to space. The two companies have partnered to launch a space-optimized version of Nvidia’s Vera Rubin NVL72 system aboard the Starmind AI satellite, with the initial mission targeting Q4 2027 and broader orbital deployment planned for 2028.
What’s actually going up there #
The Vera Rubin NVL72 is Nvidia’s rack-scale AI supercomputer, packing 72 Rubin GPUs and 36 Vera CPUs connected through NVLink 6 interconnects. The system is purpose-built for agentic AI workloads, data processing, and inference tasks.
Elon Musk foreshadowed this move during SpaceX’s earnings call on August 4, 2026, when he confirmed the company would use Nvidia’s Vera Rubin platform exclusively. He called it the “best AI computer.”
Nvidia followed up on August 24, 2026, with announcements emphasizing the role of Vera CPUs in orchestrating agentic AI across a shared infrastructure that bridges Earth and space. The company framed the evolution as moving “from gigawatt AI factories to orbit,” maintaining a unified software ecosystem for SpaceXAI’s systems whether they’re running on the ground or 400 kilometers overhead.
The compute capacity ambitions #
SpaceXAI’s compute targets are staggering in scale. The company aims to deploy over 2 gigawatts of compute capacity by the end of 2026 and roughly 10 gigawatts by late 2027. That 10-gigawatt target will span both terrestrial and orbital deployments of the Vera Rubin NVL72 systems.
Why orbit makes sense for AI inference #
The Vera Rubin architecture’s improvements in tokens per watt and reduced inference costs compared to prior Nvidia generations are particularly relevant in space, where every watt matters. Solar panels on a satellite generate finite power, so computational efficiency isn’t just an economic consideration. It’s a hard physical constraint.
Nvidia’s emphasis on maintaining a unified software ecosystem across terrestrial and orbital deployments means developers writing applications for Nvidia’s CUDA platform wouldn’t need to maintain separate codebases for ground-based and space-based inference.
For Nvidia, the partnership locks in a major customer for its next-generation architecture in an entirely new market segment. SpaceX’s exclusive commitment to the Vera Rubin platform means Nvidia won’t be competing with AMD or custom silicon for orbital compute contracts, at least not within SpaceX’s ecosystem. Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our