Simon Willison · 2026-05-07 · notable
Simon Willison: Notes on the xAI/Anthropic Data Center Deal
Simon picks apart Anthropic's new SpaceX/xAI Colossus 1 deal: environmental record, Elon's stated right to reclaim compute, and the awkward optics of choosing a controversial Memphis facility.
Simon Willison reads the small print on Anthropic's new SpaceX/xAI Colossus 1 deal and finds three load-bearing risks.
What is it?
A short take on the May 6 announcement that Anthropic has signed for all of the capacity of SpaceX/xAI's Colossus 1 data center in Memphis, while xAI keeps Colossus 2 for its own work. Simon reads the deal as the biggest item from the Code with Claude conference and works through what he sees as three uncomfortable risks.
How does it work?
He stitches together public reporting on the Memphis facility — gas turbines that initially ran without Clean Air Act permits, reports linking the site to local air-quality and hospital admissions data — with the deal's stated provision that Elon Musk reserves the right to reclaim the compute if Anthropic's AI 'engages in actions that harm humanity.' He also flags the timing: xAI deprecated several Grok models with two weeks' notice the night before the announcement.
Why does it matter?
Anthropic has been compute-starved for months; the deal unlocks more than 300 MW (about 220,000 NVIDIA GPUs) inside a month. Simon argues that the choice of this specific data center is a 'really bad look' as AI siting becomes a political flashpoint, and that letting Elon define 'harm humanity' is a non-trivial supply-chain risk for a frontier lab.
Who is it for?
AI policy watchers, infra engineers, anyone tracking the Anthropic compute story.
Try it
Read at simonwillison.net/2026/May/7/xai-anthropic/