Elon Musk said SpaceX plans to use about 220,000 GB300 GPUs at Gun Range for training by late November and about 200,000 at Spam for inference in December. The former is intended for high-bandwidth training connections, while the latter would use a comparatively simpler inference network. He also presented a target of 1.2 million Blackwell GPUs in total by year-end. The report adds facility-level configurations and schedules, but does not verify completed power and network infrastructure, operational reliability or throughput. Procurement and installation plans should not be counted immediately as compute capacity ready for production services.
Back to Tech
Musk outlines year-end plans for two SpaceX AI compute clusters
The Gun Range and Spam GPU counts and dates are targets, not verified operational capacity or performance.