Elon Musk said on Friday that SpaceX’s xAI unit will roughly double the Nvidia graphics processors inside its Colossus 2 supercomputer by year-end, a commitment that locks in billions of fresh revenue for the chipmaker and cements the rocket company as its single largest anchor customer. The post on X, dated September 25, outlines a delivery schedule that would push the cluster past the one-million-GPU target SpaceX set for the end of 2026.
The numbers
Colossus 2 currently runs 110,000 GB200 chips and 440,000 GB300s. Another 220,000 GB300s are slated to come online next week, a matching batch in November, and Musk added that a final tranche of 220,000 GB300s could arrive in late December “if we get lucky.” The initial 550,000 GPUs cost xAI roughly $18 billion; the new allotment, heavier on the newer GB300 silicon, will likely carry a higher price tag.
The partnership deepens
Earlier this month Musk declared that SpaceX would standardize on Nvidia hardware going forward, calling the Vera Rubin architecture the best available and valuing the partnership. He projected the company would finish 2026 with about two gigawatts of compute capacity and aims to reach closer to ten gigawatts than five by the end of 2027. At an estimated $50 billion to $60 billion per gigawatt, of which $35 billion is chips and other Nvidia components, the spend trajectory implies a multi-hundred-billion-dollar procurement pipeline.
The orbital bet
Nvidia and SpaceX are also co-designing a Vera Rubin NVL72 system hardened for orbit, where cosmic radiation flips bits in standard chips and there is no practical way to correct the errors without redundant checks. The plan is to launch the first unit toward the end of 2027 and begin scaling the orbital data-center fleet in 2028.
What to watch
Nvidia’s dominance in training infrastructure remains intact, and the Groq acquisition has given it a specialized language-processing unit now woven into the CUDA stack for inference workloads. Vera-branded CPUs are gaining traction with agentic AI, SpaceX confirmed in August it will use them for those tasks, but the CPU market is still a fraction of the GPU opportunity. The real signal is the scale of SpaceX’s commitment: a single customer building a compute budget that rivals the GDP of a mid-sized economy.