By Ramish Zafar
Publication Date: 2026-09-25 14:46:00
SpaceXAI CEO Elon Musk plans to bring more than half a million NVIDIA AI GPUs online as part of his firm’s Colossus data center project. Colossus started construction in 2024, and the first phase of the project, called Colossus 1, is already operational. Musk’s comments covered the second phase of the project, Colossus 2, which, like the first project, is also being developed in Memphis. According to Musk, while Colossus 1 relied primarily on NVIDIA’s Hopper GPU generation, Colossus 2 will rely primarily on NVIDIA’s Blackwell GB300 GPUs.
SpaceXAI’s Colossus 2 Data Center In Memphis Aims To Add 660,000 NVIDIA Blackwell GPUs Online By 2026 End, Says Musk
With GPUs being one of the scarcest commodities for the AI buildout, Elon Musk’s Colossus 2 data center is one of the largest in the world. Borrowing a leaf from Tesla’s book, xAI calls Colossus its ‘Gigafactory’ of compute that relies on 200,000 NVIDIA H100 GPUs. Interestingly, Musk’s latest comments for the data center bring the total number of GPUs present at the site to 230,000 as he outlined that Colossus 1 also relies on 30,000 NVIDIA GB200 AI GPUs.
For Colossus 2, Musk and xAI aim to focus primarily on NVIDIA’s Blackwell GB300 GPUs. The GB300 is an advanced variant of GB200. While they feature the same Grace CPU, the GPU is a newer variant dubbed the Blackwell Ultra.
The executive’s latest remarks came as part of a broader set of comments on X related to his claim that…
