Anthropic rents SpaceX's entire Colossus 1 capacity — usage limits doubled
On May 6, 2026, Anthropic announced a compute deal with SpaceX covering the full capacity of the Colossus 1 data center in Memphis. More than 300 megawatts and over 220,000 NVIDIA GPUs are coming online for Claude within the month. The direct effect for paying customers: doubled five-hour limits in Claude Code, raised Opus API rate limits, and the end of peak-hour throttling for Pro and Max.
What changed in concrete terms
- Pro and Max five-hour limits in Claude Code doubled.
- Peak-hour reduction removed for Pro and Max — limits now hold through US-West midday too.
- Opus API rate limits raised platform-wide (exact new values shown per account in the Anthropic pricing dashboard).
- Team and seat-based Enterprise limits also doubled.
What was true before
If you used Claude Code heavily, you knew the pattern: around US-West midday the effective limits dropped noticeably — a couple of sessions in a row and the five-hour counter was empty. Pro and Max plans were nominally provisioned with capacity, but load spikes meant the plan ceilings felt closer than the pricing tier suggested.
On the API side, Opus was the bottleneck model for many teams — rate limits hit faster than for Sonnet, and Anthropic had signaled multiple times in Q1 2026 that compute capacity was the main brake.
What’s true now
1. Doubled five-hour limits for Pro, Max, Team and Enterprise. The doubling takes effect immediately across all paid tiers. In practice that means: if you previously hit cooldown after three longer coding sessions per day, you now get roughly six sessions of comparable size before the limit kicks in.
2. No more peak-hour reduction for Pro and Max. The previously silent throttling between roughly 10:00 and 16:00 Pacific Time is gone. Anyone working from Europe whose schedule clipped into US-West windows now gets the full limit continuously.
3. Raised Opus API rate limits. Anthropic doesn’t quote a single percentage in the blog post — it points to the individual dashboard view per account. What’s observable: teams that fell back to Sonnet in Q1 because of Opus rate limits can run Opus again without hitting 429 errors.
4. Colossus 1 is Anthropic’s new primary capacity. The Memphis data center was originally built for xAI — Anthropic now rents its full AI compute capacity. The deal also includes a letter of intent for multi-gigawatt orbital compute, but that’s only operationally relevant years out.
Reading between the lines
Three things stand out about the deal — none of which is the headline itself. First: the Q1 2026 capacity squeeze was real, otherwise Anthropic wouldn’t have explicitly chained “capacity → limits” the way it did. Second: the peak-hour reduction wasn’t openly communicated before, and now shows up as “removed” in Anthropic’s own announcement — an indirect confirmation that it existed. Third: Anthropic taking the full capacity from an xAI-built Colossus cluster is a notable signal — Musk filed it under strategic, commenting on X simply with “No one set off my evil detector”.
For day-to-day work, the implication is straightforward: workflows we shifted into US-West evening hours because of limits can run during the regular workday again. One workaround leaves the routine.
What you can do now
If you’ve been hitting peak-hour throttling: Test your normal workflow between 18:00 and 22:00 European time (US-West midday). If the limit no longer kicks in — confirmed, the tier was throttled.
If you call the Opus API directly: Check your dashboard for raised rate limits. If they’re up, it’s worth revisiting Sonnet→Opus routing for tasks where quality matters more than latency.
If you’ve been considering a plan upgrade: Pro is enough again for many teams that needed Max before April. Wait two weeks under the new limits before upgrading — the scarcity that drove upgrade pressure is likely gone.
Entdecke mehr
Claude Code: How to enable Fast Mode for Opus 4.6 — and what it costs
Fast Mode for Opus 4.6 is now extra-usage only. Here's how to enable it — and what it really costs you.
GlossarAider
Aider is an open-source AI pair programmer for the command line that edits code directly in the Git repository and commits every change automatically.
LexikonComparing AI models — who builds what and how to choose
The major model families in 2026 at a glance. Who builds Claude, GPT, Gemini, Llama, Mistral, DeepSeek, Qwen — and which model to pick when.