Musk to Add 660,000 GB300 GPUs to Colossus by Year-End

Elon Musk said SpaceXAI will activate 660,000 Nvidia GB300 GPUs in the Colossus 2 datacenter by the end of December 2026, according to remarks he posted on X.

The assertion frames Colossus 2 as a GB300-first expansion that, Musk said, will push xAI's total processing footprint above one million GPUs. He placed the comment amid a broader discussion about competing with Anthropic and OpenAI and the operational challenges of bringing large GPU volumes online quickly.

Construction of the Colossus project began in 2024 and the first phase, Colossus 1, is already operational in Memphis. Musk has said Colossus 1 relied primarily on Nvidia's Hopper generation and that the complex includes 200,000 H100 GPUs plus 30,000 GB200 GPUs, bringing the total GPU count for that installation to 230,000 devices.

Hardware choices and activation schedule

Musk described Colossus 2 as chiefly built around Nvidia's Blackwell GB300 chips. He noted that GB300 is a more advanced evolution of GB200; both GB200 and GB300 pair with Nvidia's Grace CPU family, while GB300 includes a newer GPU variant named Blackwell Ultra.

In his posts on X, Musk stated that the datacenter currently operates with 110,000 GB200 GPUs and 440,000 GB300 GPUs. He outlined a stepwise activation plan for GB300 capacity: the number of active GB300 devices should reach 220,000 by the end of this week, another 220,000 by the end of November, and a final 220,000 by the end of December. Taken together, Musk said, those last additions will bring Colossus 2 to 660,000 GB300 GPUs and move xAI's overall GPU scale beyond one million.

Musk also acknowledged the difficulty of rapid scale-up, saying 'Activating a huge volume of compute quickly is very difficult.' He added that SpaceXAI's AI efforts are only three years old compared with six years for Anthropic and ten years for OpenAI, and that 'If our growth rate remains strong, SpaceX will be number one in about six months.' He previously said SpaceXAI aims to compete with Anthropic's Fable and OpenAI's GPT-6, which he suggested could be available in two to three months.

Alongside the timeline and chip details, Musk positioned Colossus as a gigafactory-style supply of compute for xAI, modeled on Tesla's approach to in-house production and scale.

Beyond the activation schedule, the remaining salient detail is Musk's projection that the completed Colossus expansion would raise xAI's aggregate GPU inventory to more than one million devices.