SpaceXAI intends to activate approximately 420,000 Nvidia GPUs in November to fulfill new compute obligations. This expansion is part of a strategy to scale the Colossus supercomputer cluster in Memphis, Tennessee, which supports the development of the Grok AI model family, including version 4.7.
As of September 25, 2026, the Colossus infrastructure consists of two sites. Colossus 1 operates 230,000 chips, comprising 150,000 H100, 50,000 H200, and 30,000 GB200 units. Colossus 2 houses 550,000 chips, including 110,000 GB200 and 440,000 GB300 units.
The company plans to add 660,000 more GPUs by the end of 2026, aiming for a total fleet of roughly 1.44 million chips. To power this infrastructure, SpaceXAI is constructing a 1.2-gigawatt power plant.
SpaceXAI has begun leasing over 220,000 GPUs from the Colossus 1 site to Anthropic. This leasing agreement could reach a value of up to $84.5 billion by 2029, depending on future commitments.
Looking toward 2030, the company aims to build infrastructure equivalent to 50 million H100-class GPUs.
