SpaceX will build its AI infrastructure exclusively on NVIDIA hardware, Elon Musk told investors on the company’s first earnings call as a public company. “Going forward, we’ve decided to build exclusively on Nvidia,” Musk said, calling the Vera Rubin architecture the best AI computer and adding that he expects SpaceX to receive a significant share of NVIDIA’s GPU supply next year.
The plan centers on the Vera Rubin NVL72 rack-scale system, codenamed Kyber, which pairs 72 Rubin GPUs with 36 Vera CPUs. SpaceX intends to deploy it on the ground and in orbit, with about 2GW of AI compute installed by the end of this year and 10GW targeted by the end of 2027.
For space, each Starmind satellite will carry a system optimized from the NVL72 design, with launches planned for 2027. NVIDIA already offers a space-certified module with four Rubin GPUs and two Vera CPUs that it says delivers up to 25 times the AI compute of an H100 for orbital workloads such as geospatial intelligence and on-orbit analytics.
The commitment marks a sharp turn from two weeks ago, when Musk dismissed a report that SpaceX had ordered $52B in NVIDIA GPUs through Foxconn as fake news. Now the company is publicly locked into NVIDIA for both terrestrial data centers and orbital AI, a win for the chipmaker as it races to supply the next wave of compute demand.
