Operational Gas-fired (mobile generators & power plant)

xAI Gigascale Data Center (Tennessee/Mississippi)

Memphis, Eastern Tennessee / Northern Mississippi, United States

Capacity
1500MW
Operator
xAI
AI Focus
Training (foundation models for xAI)
Year
2024

xAI’s Memphis Supercluster, nicknamed “Colossus,” is one of the most consequential AI infrastructure stories of 2024. Built with breakneck speed by Elon Musk’s AI company, the facility assembled 100,000 NVIDIA H100 GPUs in an abandoned semiconductor factory (the former Electrolux plant in South Memphis, Tennessee) within just 19 days — an extraordinary engineering feat that shocked the industry.

The Memphis site began operation in mid-2024 to train Grok 2, xAI’s flagship AI model accessible through X (formerly Twitter). At full deployment, the facility targets 200,000 NVIDIA H100/H200 GPUs — making it the largest GPU cluster by unit count in the world. At approximately 1.5 gigawatts of power demand, Colossus would consume more electricity than the entire city of Memphis.

Power controversy: xAI initially powered the facility using hundreds of unpermitted mobile natural gas generators, bypassing MLGW (Memphis Light Gas and Water) utility approval processes and drawing significant regulatory and community pushback. Memphis residents and environmental groups noted that the gas generators were predominantly sited in majority-Black South Memphis neighborhoods with existing air quality challenges. Tennessee regulators opened investigations into the unpermitted generators, and Shelby County filed requests for xAI to cease and desist.

In response, xAI struck a deal with MLGW for permanent grid connection and acquired a Mississippi power plant (Panola County) to provide dedicated generation capacity for the expanded 1.5 GW campus. xAI also committed to renewable energy offset agreements, though critics noted these don’t address local air quality impacts from the natural gas generation.

Scale: When fully built out, the Mississippi/Tennessee cluster would house training infrastructure for xAI’s next-generation models, potentially with 1 million+ GPU-equivalent compute units. The facility underscores xAI’s strategy of owning generation assets rather than waiting years for utility power — a playbook likely to be replicated across the AI industry.

Coordinates
35.1495, -90.049