Elon Musk Announces AI Supercomputer xAI Colossus 2 To Be Launched in the Coming Weeks
Credit: X/@elonmusk
After Sam Altman announced OpenAI’s plans to increase the number of GPUs to more than 1 million by the end of 2025, Elon Musk joined the computing power race. The billionaire reported that 230 thousand GBUs are currently running on the Colossus supercomputer, 30 thousand of which are the new NVIDIA GB200. And in the coming weeks, the first stage of Colossus 2 with 550 thousand GB200 and GB300 will be put into operation.
The company has two campuses in the city: Colossus 1 and Colossus 2. The first is located on the territory of the former Electrolux plant and includes 230 thousand accelerators, including 30 thousand NVIDIA GB200. The system is used only for training, inference is carried out in the clouds of xAI partners. The second campus, Colossus 2 on the Tulane Road site, will receive 110 thousand GB200 and GB300 in the first phase, which will require 170 MW of power. It should start working within a few weeks. The deployment dates for the remaining 440 thousand accelerators are not determined, since the supply of NVIDIA GB300 is delayed.
The input capacity will be used to train Grok 4 and other xAI models. Let me remind you that in August the company plans to release a special version of Grok 4 Coder for programming, and in September the release of the basic model V7 is scheduled – an updated Grok 4 trained on a higher-quality dataset. Also by the end of the year the company will launch a new version of image generation and the first-ever AI for video creation.
