Elon Musk plans to install 660,000 NVIDIA GB300 GPUs in his Colossus 2 AI training cluster by late December, bringing the Memphis facility’s total Blackwell-series chips well past the half-million mark. The current complex, already the world’s largest AI data center by power consumption, houses 440,000 GPUs generating 1.112 million H100-equivalent compute units across 946 megawatts of IT power. The capital cost so far stands at approximately $35.8 billion, with the full buildout projected at $58 billion. The site currently relies on temporary gas turbines while a permanent 1.2 GW power plant in Southaven, Mississippi is expected to reach full capacity by mid-2027. Leasing agreements reportedly include Anthropic at roughly $1.25 billion per month and Google utilizing approximately 110,000 GPUs.
Source: Read the original article
