Anthropic has struck a deal to take over the full computing capacity of SpaceX's Colossus 1 data center in Memphis: 220,000+ NVIDIA GPUs and more than 300 megawatts of power. The infrastructure is expected to come online within a month — and the rate limit changes for Claude users are already rolling out.
This is a direct response to something developers have been complaining about for months: hitting Claude's ceilings constantly, especially on Opus and Claude Code.
"Higher limits, more capacity, and a path to orbital AI compute" — Anthropic's announcement framing it as infrastructure, not just a business deal.
What actually changed
- Claude Code 5-hour limits double for Pro, Max, Team, and Enterprise plans — within a month
- Peak-hour throttling removed entirely for Pro and Max subscribers
- Claude Opus API limits jump substantially across all tiers:
| Tier | Input tokens/min | Output tokens/min |
|---|---|---|
| 1 | 30k → 500k | 8k → 80k |
| 2 | 450k → 2M | 90k → 200k |
| 3 | 800k → 5M | 160k → 400k |
| 4 | 2M → 10M | 400k → 800k |
That's a 10–16× increase depending on tier. If you're running agents on Opus, this is a big deal.
Why it matters
Compute constraints have quietly been the ceiling on AI product development. Rate limits aren't just annoying — they force architectural compromises: developers route around Opus to cheaper, faster models; agent loops get throttled mid-run; teams negotiate enterprise contracts just to get workable quotas.
This deal doesn't just move the ceiling — it blows the roof off.
The Colossus 1 cluster was originally built for xAI (Elon Musk's AI company), assembled by SpaceX in a reported 19 days. Now Musk has folded xAI into SpaceX under the SpaceXAI brand, and in an unexpected move, Anthropic — not exactly Musk-aligned — has secured exclusive access to it.
Anthropic also noted it's exploring "orbital AI compute capacity" with SpaceX. That's still vague, but it suggests the relationship goes beyond a one-time rental.
One tension worth flagging: Anthropic stated it will only partner with "democratic countries" whose legal frameworks support this scale of investment. Partnering with a company controlled by Elon Musk, whose political trajectory has raised authoritarian flags, sits awkwardly against that framing. Anthropic didn't address the contradiction directly.
What to do
If you're on the Claude API (Opus): Your rate limits may already be higher — check your current tier in the Anthropic console and adjust your retry/backoff logic accordingly. Some agents that were throttled-by-design can now run more aggressively.
If you use Claude Code (Pro/Max/Team/Enterprise): The 5-hour limit doubling rolls out within a month. No action needed, but if you've been rationing sessions, you'll have more headroom soon.
If you're evaluating model providers: The compute gap between providers is narrowing fast. Anthropic now has deals with Amazon (5 GW), Google/Broadcom (5 GW), Microsoft/NVIDIA ($30B Azure capacity), Fluidstack ($50B), and now SpaceX's Colossus. The infrastructure story is becoming a competitive moat in its own right.
Source: The New Stack · Anthropic announcement · The Decoder
✏️ Drafted with KewBot (AI), edited and approved by Drew.
Top comments (0)