Musk Says Colossus 2 Will Pass 1.2M Nvidia Chips
Elon Musk has laid out the most detailed timetable yet for xAI's Memphis supercomputer: Colossus 2 runs about 550,000 Nvidia chips today and could more than double to over 1.2 million by the end of December, in three waves of 220,000 GB300s. He hedged the last wave on luck.
Elon Musk has put dates on the next stage of Colossus 2. In a post on X, he said the xAI supercomputer outside Memphis now runs 110,000 Nvidia GB200 chips and 440,000 GB300s, about 550,000 in all, and that it could more than double by the end of the year, Bloomberg reports. It is the most detailed timetable he has given for the build-out.
The plan comes in three waves of the same size. Another 220,000 GB300s are due to come online next week, and a further 220,000 in November. A third batch of 220,000 could follow in late December “if we get lucky,” Musk wrote, according to the Bloomberg report. If all three arrive, Colossus 2 would reach about 1.21 million Nvidia chips. Adding the original Colossus next door, the Memphis site would approach 1.44 million accelerators, WION notes, one of the largest concentrations of AI computing anywhere.
The customers are no longer only Musk’s own. Since xAI was folded into SpaceX this year, Colossus has been selling capacity to outside labs as well as training Grok. BitsMinds has covered Anthropic’s deal to run Claude on Colossus at about $1.25 billion a month, and a Google agreement to rent 110,000 GPUs there. That changes how the new chips should be read. Some of the extra capacity will go to Grok, but part of it is likely to end up serving xAI’s rivals, and each new wave makes the site more valuable as a landlord.
The timetable is a plan rather than a promise. Chip deliveries slip, and each wave of 220,000 GB300s needs power and cooling in a city where the site’s environmental impact has already drawn scrutiny. Musk’s own hedge on the December batch says as much. Even so, the first wave is due within days, which will soon show whether the schedule is holding. If it does, a cluster that held roughly half a million chips in September will end the year with more than twice that, added in a single quarter.
Want AI news before everyone else?
The morning's most important AI stories, straight to your inbox. No fluff.