2026-05-07
Anthropic signs SpaceX Colossus deal — 220,000 GPUs, Claude Code rate limits removed
Anthropic taps SpaceX Colossus 1 in Memphis for 220K+ NVIDIA GPUs going live this month, and immediately lifts peak-hours throttling on Claude Code Pro and Max.
Announced at the Code with Claude SF conference on May 6, Anthropic signed a deal with SpaceX to access the full capacity of Colossus 1 — the Memphis, Tennessee data center housing 220,000+ NVIDIA H100/H200/GB200 GPUs across 300+ megawatts of power. Anthropic said the capacity goes live within the month.
The immediate effect: rate limits lifted
As a direct downstream consequence, Anthropic removed peak-hours throttling on Claude Code for Pro and Max users on the same day. Opus API rate limits were raised significantly. This addresses one of the most consistent complaints from heavy Claude Code users: agentic sessions being interrupted mid-task during US business hours.
What Colossus 1 is
Colossus 1 is the same cluster xAI uses to train and serve Grok. It was built in approximately 122 days in 2024 and is one of the largest single-site GPU concentrations on the planet. Tapping it means Anthropic is expanding inference capacity faster than any purpose-built cluster they could have commissioned at this point.
The SpaceX relationship reflects a shift in how frontier AI labs source compute: rather than committing to 2–3 year hyperscaler contracts, they are acquiring capacity from whoever can deliver it fastest. SpaceX’s operational speed — Colossus 1 was activated at an unusually rapid pace even by data-center standards — makes it a realistic option for labs that need headroom now.
What stays the same
The deal is for inference capacity, not training. Anthropic’s primary training infrastructure remains on AWS and Google Cloud. Colossus 1 adds to the serving layer, specifically targeting the consumer and API surface where rate-limit complaints have been loudest.
Practitioner note
If you have been hitting rate walls during agentic Claude Code sessions or heavy Opus API calls during peak hours, the mitigation should be measurable within a few weeks. The practical test: run a parallel multi-agent job you previously had to schedule for off-peak hours and check whether throttling still triggers. If your team is on a Max or Team/Enterprise plan, the rate-limit raise is the more relevant number — review the updated limits in the Anthropic console before repricing your API budget.
Sources
- Anthropic inks computing deal with SpaceX — Bloomberg ↗
- Anthropic SpaceX data center capacity — CNBC ↗
- Higher limits and SpaceX compute — Anthropic blog ↗