xAI will add 110k Nvidia GB200 GPUs to its Supercomputer

xAI founder, Elon Musk, has revealed the plan to install 110k new Nvidia GB200 GPUs at the Colossus supercomputer facility in Memphis, Tennessee. This large expansion is targeted at gaining the immense training capabilities for the next generation of Grok AI models.

In October 2024, xAI announced the completion of its Colossus supercomputer cluster, featuring 100,000 Nvidia Hopper GPUs tagged with the Spectrum-X ethernet networking platform. Phase 1 of the facility was built and went online in just 122 days, and the second phase was completed in just 92 days, adding another 100k GPUS.

The AI firm revealed this expansion during the Grok 3 series launch in February this year. However, it also promised to take the platform further with more GPUs.

Source – xAI

Musk revealed the current generation of GPUs used in the Colossus computer. That includes 150,000 H100s, 50,000 H200s and 30,000 GB200s. In the next phase, xAI will add 110,000 GB200s, totaling 340,000 GPUs.

The xAI chief said the new cluster will be online soon, but refrained from providing any date. This scale of GPU is massive, and the company may have spent a lot of money to buy the hardware and technologies to expand the existing cluster.

The AI company has a launch pending for Grok 3.5, and we could expect to get more information about the new GPU addition at the unveiling live stream soon.