Colossus 2 (first gigawatt-scale training cluster)
Elon Musk says Colossus 2 is operational as the world's first gigawatt-scale AI training cluster, with a planned upgrade to 1.5 GW in April.
Second Memphis-area cluster, announced in a Musk post on X. xAI had said in late 2025 it was buying a third building to reach nearly 2 GW and over 555,000 Nvidia GPUs (about $18B). Combined with Colossus 1, press reports it above one million H100-equivalents.
- Date
- Saturday, 17 January 2026
- Lab
- xAI
- Kind
- infra
- Access
- closed API
Figures
| Measure | Value | Measured by |
|---|---|---|
| Power | 1 GW operational, 1.5 GW planned for April 2026 Musk's own post as quoted by press | company |
| Planned GPUs | 555,000+ Nvidia GPUs, about $18B Colossus 2 target announced 2025-12-31 | company |
Capacity claims come from Musk and have not been independently audited. Wikipedia's summary also reports that Anthropic leased essentially all Colossus 1 compute in May 2026 and Google agreed to rent about 110,000 GPUs in June 2026. I could not corroborate either with a news or primary source, so they are not treated as facts here. 'access' is not meaningful for infra.
Sources
- interestingengineering.com/ai-robotics/elon-musk-xai-gigawatt-scale-ai-training-cluster
- www.techzine.eu/news/infrastructure/137578/xai-expands-colossus-megadata-center-to-2-gigaw
- en.wikipedia.org/wiki/Colossus_(data_center)
This record was checked against its sources on 6 October 2026. How we check