Elon Musk plans to roughly double Colossus 2’s already enormous collection of advanced Nvidia AI chips before the end of 2026.
According to Reuters via Investing.com Musk said the Memphis, Tennessee, AI cluster currently contains 110,000 Nvidia GB200 processors and 440,000 GB300s. That gives Colossus 2 a present tally of 550,000 chips.
Another 220,000 GB300s should be fully operational next week, followed by another 220,000 in November. If everything behaves itself, Musk thinks another batch could squeeze in before New Year.
“If we get lucky, yet another 220k GB300 by late December,” Musk said.
That schedule would take Colossus 2 to 1.21 million GB200 and GB300 processors by the end of December. Even without the final 220,000, the cluster would reach 990,000 chips. The numbers are substantially larger than SpaceXAI’s previously advertised target of equipping its Memphis operation with one million GPUs during 2026. The company says Colossus provides the computing muscle behind its Grok AI models.
Musk also gave figures for the older Colossus 1 cluster. It contains 150,000 Nvidia H100S, 50,000 H200S and 30,000 GB200 processors, bringing that machine’s tally to 230,000.
The build speed has been almost as bonkers as its scale. SpaceX disclosures say the first Colossus 2 cluster brought roughly 110,000 GB200 processors and about 210MW of compute power online in 91 days. A second cluster containing 110,000 GB300 processors and 220MW of compute power followed in 64 days. SpaceX said the next expansion phase would add at least another 220,000 GB300 processors and more than 400MW.
Nvidia previously said Colossus 2 would house more than half a million of its GPUs. Musk’s latest numbers suggest that milestone has already been passed and the accelerator population could more than double again within three months.
All that silicon needs a suitably ridiculous amount of electricity. Independent research group Epoch AI estimates Colossus 2 currently supports about 946MW of IT power, although Musk did not say whether sufficient power and supporting infrastructure are ready for every processor in his year-end timetable.
Colossus is no longer being kept entirely for Musk’s own AI toys. The complex has made computing capacity available to outside AI operations, while SpaceXAI continues expanding the Memphis site toward the scale needed to train and run increasingly hungry frontier models.
