Elon Musk targets 1.2 million Blackwell GPUs for xAI Colossus 2 mega-cluster Other

Elon Musk stated that xAI plans to assemble approximately 1.2 million Nvidia Blackwell GPUs across its Colossus 2 computing campus in Memphis by the end of the year.

The expansion aims to provide two to three times more training compute for Grok 5 than current frontier models, scaling across four interconnected data center facilities.

What it is

• Four-facility Memphis footprint. Two facilities housing 220,000 GB300 accelerators each are currently active for training, with a third coming online in November and a fourth dedicated to inference scheduled for December.
• Leading global compute density. Independent data from Epoch AI ranks the campus as the highest-capacity AI compute site globally at roughly 1.1 million H100-equivalents, outpacing dedicated clusters from Amazon, Microsoft, and Meta.
• Compute disparity. The planned scale would provide xAI with significantly higher training capacity than was utilized to train preceding frontier releases such as Claude Opus 5.5.
• Colossus 1 sublease. xAI leased the entirety of its initial Colossus cluster to Anthropic starting in May, deploying subsequent models on newly built capacity.
• Schedule feasibility. Industry analysts note that reaching 1.2 million Blackwell accelerators by December 31 represents an aggressive deployment timeline subject to electrical interconnect and supply constraints.

Global AI campus comparison (Epoch AI)

Sources

• Epoch AI: Tracking Frontier AI Compute Clusters
• Elon Musk on X