2 min read
Elon Musk’ xAI finally turned on its Colossus supercomputer located in Memphis, Tennessee. Driven by 150 MW of Tesla Megapack batteries and a recently operational substation, Phase 1 is now live using 200,000 Nvidia H100 and H200 GPUs. A second substation this fall will quadruple Colossus’s grid power use to 300 MW, enough to run 300,000 homes. It currently consumes 150 MW right now.
Rapid Deployment Achieved in Record Time
Construction on the Colossus cluster began in early 2024 at a repurposed Electrolux factory site and in just 122 days, xAI scaled from zero to 200,000 GPUs, a build speed that Nvidia CEO Jensen Huang says typically takes years .
The site initially launched with 100,000 H100 GPUs in July 2024, drawing about 70 MW, and then doubled its GPU count by February 2025 to reach full Phase 1 capacity .
Tesla Megapacks Stabilize Power Supply
To smooth demand spikes and reduce reliance on backup generators, xAI installed 150 MW of Tesla Megapack battery banks alongside the new substation.
These batteries handle peak loads and outages until the second TVA/MLGW substation comes online, at which point portable gas turbines will be relegated to emergency standby .
Strategic Implications for AI Competition
Colossus underpins xAI’s Grok chatbot and next-generation language models, giving Musk a scale advantage over rivals like OpenAI and Google by a great margin. By vertically integrating Tesla energy storage, Supermicro liquid-cooled racks, and rapid infrastructure buildout, Musk aims to lower training costs and accelerate AI development under Musk’s “compute factory” vision.
Environmental and Regulatory Concerns
Despite the substation upgrade, xAI still operates dozens of natural-gas turbine generators without full Clean Air Act permits. Advocacy groups and local officials report up to 35 on-site turbines, far exceeding the 15 permitted, prompting Shelby County regulators to schedule a public hearing over air-quality and health impacts .
Musk’s Phase 2 and Beyond
Phase 2 of the project, scheduled for late 2025 operation, will oversee another 150 MW substation to reach a 300 MW draw, sufficient to power 200,000 + GPUs under load with room to expand.
Longer-term plans call for scaling Colossus to one million GPUs, which would require an estimated 1.55 GW of power and further grid enhancements: efforts already in talks with MLGW, TVA, and hardware partners Nvidia, Dell, and Supermicro.
