Musk Aims to More Than Double Colossus 2's Nvidia Chips by Year-End

Elon Musk's disclosure about Colossus 2 puts hard numbers on what has become the most conspicuous compute build in the industry. The Memphis cluster currently houses roughly 110,000 Nvidia GB200 and 440,000 GB300 chips. Musk detailed a delivery cadence that would roughly double that footprint: another 220,000 GB300s arriving imminently, 220,000 more in November, and a potential further 220,000 by late December — a trajectory that could push the cluster toward approximately 1.44 million GPUs.

The purpose is singular: feeding the training runs behind Grok's rapid version cadence, including the 4.7 API release this same week. The scale is a deliberate strategic wager that raw compute is the decisive lever on capability, and it makes xAI one of Nvidia's largest single customers at a moment when chip allocation is itself a competitive weapon.

Infrastructure specialists in the source community were divided. On one side, the ~780K-to-1.44M GPU scaling is framed as competitive necessity given how quickly OpenAI, Anthropic, and Google are provisioning. On the other, engineers questioned whether it is an unsustainable arms race, and whether pouring resources into raw compute rather than safety and alignment is the right bet — a debate sharpened by the same week's agent-containment failures at OpenAI and DeepSeek.

The build also intersects with the week's other capital story: compute deals and clusters (Anthropic-Akamai, Alibaba's 20GW plan) are the industry's real theme this week, with model launches almost a sideshow to the infrastructure land-grab underneath them.