Elon Musk plans Memphis AI cluster expansion to 1.1M GPUs by 2026
In brief
- Musk commits to doubling Colossus 2 cluster's Nvidia chip count by end of 2026
- Expansion targets 1.1M+ H100-equivalent accelerators across Memphis-area facilities
- 1.2 GW gas-fired power plant under construction to replace temporary turbines
- Google and Anthropic secure capacity at expanding facility
Scaling the AI Compute Footprint
The Memphis-area cluster had already hit roughly 1 GW of AI compute capacity by mid-2026, and the expansion signals a dramatic acceleration in infrastructure buildout. The facility now integrates multiple sites, including the original Colossus 1 facility alongside Colossus 2 and newer additions. The full Colossus 2 buildout is modeled for an IT load of around 946 MW, according to facility specifications.
Major AI labs are already moving in. Google signed a contract in June 2026 for approximately 110,000 GPUs, paying a rate of $920 million per month from October 2026 through June 2029. Anthropic has also secured capacity at the facility.
Power Infrastructure Race
The expansion hinges on solving a critical constraint: energy. The facility is primarily powered by temporary gas turbines and Tesla Megapacks. But regulators have ordered the phase-out of 69 temporary turbines by mid-2027, forcing a permanent solution.
A permanent 1.2 GW gas-fired power plant is under construction in Southaven to replace the temporary generation equipment. The timeline matters — the plant must be operational before the regulatory deadline hits. This mirrors the broader AI infrastructure race, where power availability has become the hard constraint limiting GPU deployment worldwide.
