Elon Musk said in a post on X that his xAI supercomputer, Colossus, will have 220,000 Nvidia GB300 chips running within the next week, according to Tom’s Hardware. He added that a second batch of 220,000 GB300s will come online in November, with a third batch of the same size following in late December, a target he qualified with the phrase “if we get lucky.”
Taken together, the three tranches would push Colossus toward roughly 1.44 million GPUs in operation, a total Tom’s Hardware framed as the goal in sight rather than a number Musk stated outright. Every figure here traces back to Musk’s own account of his own company, not a filing or an outside audit.
The pace matters because GPU supply, not model design, has become the binding constraint on how fast xAI can train against OpenAI and Anthropic. A slip in the conditional December tranche would push full deployment into 2027 and widen the gap with rivals who have locked in steadier chip deliveries this quarter.
Reported by Tom’s Hardware, citing Elon Musk’s own post on X, on 25 September 2026.
