Elon Musk to Expand Colossus 2 AI Cluster with 660K GPUs by Year-End

Elon Musk is set to expand the Colossus 2 AI training cluster by adding 660,000 NVIDIA GB300 GPUs by late December. This ambitious move will increase the facility's total GPU count to over 1 million, solidifying its status as the largest AI data center globally in terms of IT power consumption.

The current configuration of Colossus 2 already includes approximately 330,000 B300 GPUs and 110,000 B200 GPUs, generating around 1.112 million H100-equivalent compute units. The expansion is projected to elevate this capacity to about 1.825 million H100-equivalent units.

The Memphis complex, which also includes Colossus 1 and future developments, is expected to exceed 1 million GPUs and achieve a total power capacity of roughly 2 GW. Currently, the facility relies on temporary gas turbines for power, which has raised environmental concerns. A permanent 1.2 GW power plant is planned to be operational by mid-2027.

NVIDIA’s Blackwell architecture, chosen for this expansion, offers a significant improvement in AI training efficiency compared to previous generations, making it a crucial element in the facility's growth.