SAN FRANCISCO — Elon Musk's xAI is preparing one of the largest single expansions of computing power the artificial intelligence industry has ever seen, with plans to roughly double the number of Nvidia chips inside its Colossus 2 supercomputer before the end of the year.
Musk said this week that the Memphis, Tennessee cluster — already described by the company as the most powerful AI training machine on the planet — will bring another 220,000 Nvidia GB300 processors online within days. A second wave of roughly 220,000 chips is slated for November, with a possible third batch in late December. The pace reflects how central raw compute has become to the race to build ever more capable models like Grok.
Toward 1.44 million GPUs
Colossus 2 currently runs on about 110,000 GB200 processors and 440,000 of the newer GB300 chips. If the deployment schedule holds, the facility could house roughly 1.44 million GPUs by year end — a scale that dwarfs nearly every other commercial AI installation and pushes xAI to the front of the global compute race.
"$SPCX will have ~1 million GPUs online next week as another 220K GB300s come online," Musk wrote, referring to the combined SpaceX-xAI company that now trades publicly. Reaching seven figures of active accelerators would be a milestone no rival has publicly claimed.
Built for Grok, opened to the world
xAI originally constructed the first Colossus in 2024 to train its Grok family of models, standing up an initial cluster in a matter of months in a feat that stunned data center veterans. Colossus 2 extends that approach, and the additional GB300 silicon will feed both the next generation of Grok and the fast-growing roster of products built on it, from coding assistants to the voice tools showcased in Grok Voice Transcribe 2.0.
The company has also begun renting spare capacity to outside developers, turning a training asset into a revenue engine. That dual role — powering xAI's own frontier research while selling compute to others — helps justify the enormous capital outlay and positions Colossus 2 as one of the most strategically valuable pieces of infrastructure in Musk's portfolio.
Power, cooling and a Memphis bet
Deploying more than a million GPUs is as much an energy and engineering challenge as a chip-buying exercise. The Memphis campus has expanded rapidly, and xAI has leaned on on-site power generation and aggressive liquid cooling to keep the machine fed. Each new tranche of GB300 chips raises the bar on electricity, water and networking, and xAI's willingness to solve those problems at speed has become a competitive advantage in its own right.
For a company that did not exist a few years ago, the trajectory is striking. As Musk detailed in remarks reported by Bloomberg, the goal is not simply to match competitors but to out-build them. If Colossus 2 crosses the million-GPU threshold on schedule, xAI will enter 2027 with a compute foundation few can rival — and Grok's next leap may follow close behind.