SpaceX Partners with Nvidia to Scale AI Compute from Earth to Orbit
Elon Musk taps Nvidia's Vera Rubin systems to bypass terrestrial power limits and launch the Starmind orbital compute program.
SpaceX has entered into an exclusive partnership with Nvidia to power its artificial intelligence infrastructure across both Earth and orbit. The move signals a strategic shift toward a hybrid compute model designed to bypass the physical limitations of terrestrial data centers.
Elon Musk announced that SpaceX will exclusively use Nvidia GPUs for its AI workloads, stating the decision was made "because they are the best." Central to this deployment is the Vera Rubin NVL72 rackscale system, codenamed Kyber, which will be installed in ground-based data centers and launched into space. The orbital component is part of the Starmind satellite program, with launches scheduled to begin next year. For these space-based workloads, Nvidia's Space-1 Vera Rubin module delivers up to 25 times the AI compute capability of an H100 GPU.
Solving the Terrestrial Bottleneck
The transition to orbital computing addresses critical constraints currently facing the AI industry. Terrestrial data centers are increasingly hampered by scarcity of land, limited power grid capacity, and the immense energy costs associated with cooling high-density GPU clusters. By moving compute into orbit, SpaceX can leverage the natural vacuum of space for more efficient thermal management and utilize abundant solar energy to power its hardware.
This approach aligns with a growing trend toward orbital computing, which is essential for real-time geospatial intelligence and the autonomous operations required for deep-space exploration and satellite management. By shifting the heavy lifting of AI training and inference off-planet, SpaceX aims to decouple high-performance computing from Earth-bound resource limits.
Scaling the Compute Frontier
The scale of the ambition is reflected in SpaceX's aggressive power targets. The company aims to have 2GW of compute capacity online by the end of 2025, with plans to scale that figure to 10GW by the end of 2027. This massive expansion represents a significant strategic victory for Nvidia, securing a lock-in with one of the world's most influential aerospace and technology firms.
Beyond the immediate hardware win, the partnership establishes a new paradigm for high-performance computing (HPC). SpaceX is attempting to create a scalable, off-planet cloud infrastructure that could redefine how large-scale models are processed, effectively turning the orbital environment into a massive, solar-powered data center.
What to Watch
As the Starmind program moves toward its launch window next year, the industry will be watching for the first performance benchmarks of the Space-1 Vera Rubin modules in a microgravity environment. While the hardware specifications are promising, the practical challenges of maintaining rackscale systems in orbit—including radiation shielding and hardware longevity—remain the primary hurdles to achieving the 10GW target by 2027.