Elon Musk‘s artificial intelligence venture xAI has announced plans to expand the Colossus supercomputer in Memphis, Tennessee. The Colossus supercomputer, which is currently progressing towards 200,000 Nvidia-branded AI GPUs, will host 1 million GPUs in the future, making it the largest supercomputer in history. xAI aims for 1 million GPUs
The announcement was made during a lunch organized by the Greater Memphis Chamber, which helped build the xAI facility. Colossus was deployed in an incredible four months with 100,000 Nvidia H100 GPUs this summer. At that time, it was announced as the largest AI supercomputer. When scaled to 1 million GPUs, it will far surpass existing supercomputers. The current most powerful supercomputer, El Capitan, has about 43,000 AMD CPUs and 43,000 GPUs.
xAI plans to increase Colossus’ capacity to 200,000 GPUs in a short time and aims to reach 300,000 GPUs by next summer. Most of the GPUs are expected to be supplied by Nvidia. xAI has not disclosed when it will reach at least 1 million GPUs. Elon Musk said, “I’m just trying to complete 1% of a poor Kardashev Type I civilization!” Musk has been in talks with Huang for early access to the Nvidia Blackwell GB200 servers, which will cost $1.08 billion for the xAI supercomputer in the early next year. However, the cost of 1 million GPUs will be tens of billions of dollars.
xAI recently raised a $6 billion new fund, with some of the funding going to Colossus. The company offers a paid X users the Grok chat robot while developing artificial intelligence models under the same name.
On the other hand, other companies such as Meta, OpenAI, and Microsoft have also purchased large quantities of Nvidia’s latest GPUs. In January, Meta CEO Mark Zuckerberg stated that the company will have “about 600,000 H100 equivalent processors” by the end of this year. Meanwhile, the Greater Memphis Chamber stated that Nvidia, Dell, and Supermicro Computer, which helped establish Colossus, plan to operate in the city.
