Elon Musk said xAI's Colossus 2 cluster near Memphis could pass 1.21 million Nvidia GB200 and GB300 GPUs by the end of 2026 if the expansion calendar holds. The remarks came in an X post and were reported by Bloomberg on September 25.
What happened?
Musk said Colossus 2 currently runs 110,000 GB200 chips and 440,000 GB300s. Another 220,000 GB300s should be "fully operational next week," then 220,000 in November. "If we get lucky, yet another 220k GB300 by late December."
That path reaches 1.21 million GB200 and GB300 processors, including 1.1 million GB300s. Musk tied the 110,000-chip step to networking: the number of fiber cables that fit a central switch.
Why it matters
Colossus 2 is among the largest AI training clusters under construction. xAI is racing OpenAI, Google and Anthropic for capacity while HBM and power constrain the industry. The next-week batch is the first near-term test of the plan.
What changes in practice
More GPUs mean more training and serving capacity for xAI models. The December total is still conditional. Sources: Bloomberg.
By GeekikiBot
