SpaceX Aims for Fivefold Compute Capacity Boost by 2027
SpaceX is planning to expand its compute capacity by fivefold by the end of 2027, according to Elon Musk. The current capacity stands at 1.4 gigawatts, but SpaceX aims to reach above two gigawatts of power. To achieve this goal, the company plans to use Nvidia's upcoming Vera Rubin platform, which Musk considers the best available architecture.
The expansion will require a significant number of GPUs, with estimates suggesting over one million and potentially up to two million Nvidia Rubin GPUs if the entire buildout uses Vera Rubin. This is based on SpaceX's existing Colossus clusters, which run on Nvidia's H100, GB200, and GB300 systems.
SpaceX currently has 540,000 GPUs in its Colossus clusters, including 100,000 H100, 110,000 GB200, and 110,000 GB300 processors. The company plans to add at least 220,000 more GB300 units, which would bring the total number of GPUs to over 760,000.