Musk says Colossus 2 will field 660,000 Nvidia Blackwell GPUs by end of 2026
Musk said on September 27 that xAI’s Memphis Colossus 2 data center will run 660,000 Nvidia Blackwell GPUs by the end of 2026, currently running 110,000 GB200 and 440,000 GB300 units.
On September 27, Kuaikeji reported that xAI is pushing ahead with its Memphis Colossus 2 data center, whose main chip is the Nvidia Blackwell GB300. Musk said Colossus 2 will have 660,000 Nvidia Blackwell GPUs online by the end of 2026, with more than 500,000 AI GPUs deployed overall, and that the site is among the largest data centers in the world. xAI borrows the Tesla naming convention and calls Colossus its compute "Gigafactory."
Musk gave a current-operating breakdown: Colossus 2 is currently running 110,000 GB200 GPUs and 440,000 GB300 GPUs, with a phased plan to bring 220,000 more GB300s online by the weekend, another 220,000 by the end of November, and the final 220,000 by the end of December 2026. Colossus 1 was originally built around 200,000 Nvidia H100 GPUs and now runs 230,000 GPUs, including 30,000 Nvidia GB200 AI GPUs.
Alongside the capacity expansion, Musk made a rare model-comparison promise, claiming xAI should reach the capability level of Anthropic’s current models and OpenAI’s GPT-6 within two to three months, with a realistic shot at industry leadership within about six months. GB300 shares the Grace CPU with GB200 but ships the newer Blackwell Ultra GPU. Musk also stressed that bringing large-scale compute online fast is "extremely difficult," framing it as xAI’s core engineering challenge.