Musk Disbands xAI, Rents 220,000 GPUs to Anthropic, Halving Claude Prices

Deal Overview

Anthropic signed a lease for the entire compute capacity of SpaceX’s Colossus 1 data centre. The facility provides over 300 MW of power and houses approximately 220,000 NVIDIA GPUs. The hardware is expected to be brought online within the month for running Claude models.

Impact on Claude Service

Pro, Max and Team subscription tiers previously limited the number of requests per five‑hour window; the limit is now doubled.

API call throughput has been increased because the previous GPU shortage has been eliminated.

Colossus 1 Background

Location: Memphis, Tennessee.

Originally built as the super‑computer for training xAI’s Grok model.

SpaceX acquired xAI on 2 February 2024, valuing the combined entity at $1.25 trillion; Colossus 1 became a SpaceX asset.

In January 2024 xAI launched Colossus 2, a gigawatt‑scale system, relegating Colossus 1 to a secondary role.

Strategic Context

SpaceX IPO preparation – a prospectus seen by Reuters indicates a target valuation of $1.75 trillion and a raise of $750 billion, with AI compute assets highlighted as a core narrative.

Anthropic’s compute‑buying spree – in April 2024 Anthropic signed three large‑scale compute agreements (Amazon, Google/Broadcom, and SpaceX), demonstrating high demand for GPU capacity.

xAI pivot – after acknowledging in March 2024 that Grok underperforms on programming tasks, xAI began renting compute from Colossus 1 to Cursor and is exploring deeper partnership or acquisition possibilities.

Orbital Compute Prospects

SpaceX has applied to the FCC for permission to launch up to one million satellites that could serve as orbital data centres. Anthropic expressed interest in jointly developing multi‑gigawatt orbital AI compute, implying future Claude instances might run on low‑Earth‑orbit platforms.

Code example

来源丨
经授权转自
JackCui（ID：JackCui-AI）
作者丨
JackCui