Elon Musk is personally overseeing the launch of supercomputer clusters in Memphis, spending up to four days a week on-site. The company intends to install 200,000 Nvidia GB300 accelerators every month through the end of 2026 and maintain this pace throughout the following year.
By the end of 2026, SpaceX expects to have 2.5 to 3 gigawatts of computing power operational. In 2027, the firm plans to add another 10 gigawatts, with Musk suggesting that total capacity could reach 13 gigawatts or potentially exceed 20 gigawatts under an ambitious scenario.
The company plans to shift its primary focus in 2027 toward computing systems based on the Vera Rubin architecture. Musk identified cooling systems and electrical infrastructure as the main constraints for scaling these operations.
To address these bottlenecks, SpaceX aims to build a 20-gigawatt cooling infrastructure by 2027. This capacity is intended to exceed the requirements of currently installed GPUs, allowing for faster expansion of artificial intelligence computing resources.
