Colossus
Large-scale GPU training cluster
Colossus is xAI's dedicated training supercomputer — one of the largest GPU clusters ever assembled, built specifically to train successive generations of Grok rather than repurposed from general-purpose cloud capacity. Its scale is what lets xAI ship new model generations on a faster cadence than competitors renting shared infrastructure.
Purpose-built, not rented
Colossus was designed and built specifically for training Grok, not leased from a third-party cloud — giving xAI full control over scheduling and scaling.
Grid-scale power
Running at 300+ megawatts, Colossus draws power on the scale of a small city, with dedicated on-site generation supplementing grid supply.
Fast iteration
The scale of the cluster compresses training time for each new Grok generation from months to weeks.
Compute
Power
Location
01 / 02
02 / 02