Anthropic has finalized a major agreement to take over the full computing capacity of SpaceX's Colossus 1 data center in Memphis. This infrastructure boasts over 220,000 NVIDIA GPUs and more than 300 megawatts of power. The facility is expected to become operational within a month, with initial rate limit changes for Claude users already rolling out.
This move directly addresses a long-standing issue for developers, who have frequently reported hitting Claude's usage ceilings, especially when utilizing Opus and Claude Code models.
Anthropic framed this announcement as an infrastructure development rather than merely a business deal, highlighting its provision of "higher limits, more capacity, and a path to orbital AI compute."
Key changes to rate limits include:
- Claude Code's 5-hour limits will double for Pro, Max, Team, and Enterprise plans within the next month.
- Peak-hour throttling will be entirely removed for Pro and Max subscribers.
- Claude Opus API limits are substantially increasing 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 |
This translates to a 10–16x increase depending on the tier. For developers running agents on Opus, this represents a significant breakthrough.
Compute constraints have historically capped AI product development. Rate limits are not just an inconvenience; they force architectural compromises, compelling developers to route tasks to cheaper, faster models, throttling agent loops mid-run, or necessitating enterprise contract negotiations just to secure functional quotas. This deal doesn't merely raise the ceiling; it effectively removes it.
The Colossus 1 cluster was originally constructed for xAI, Elon Musk's AI company, and reportedly assembled by SpaceX in 19 days. Following Musk's integration of xAI into SpaceX under the SpaceXAI brand, Anthropic—a company not typically aligned with Musk—has unexpectedly secured exclusive access to this powerful infrastructure.
Anthropic also indicated that it is exploring "orbital AI compute capacity" with SpaceX. While details remain vague, this suggests a deeper relationship beyond a one-time rental agreement.
A notable tension arises from Anthropic's stated policy of partnering only with "democratic countries" whose legal frameworks support large-scale investments. Partnering with a company controlled by Elon Musk, whose political trajectory has drawn criticism for leaning towards authoritarianism, presents a perceived contradiction that Anthropic did not directly address.
For developers currently using the Claude API (Opus), your rate limits may already be higher. It's advisable to check your current tier in the Anthropic console and adjust retry/backoff logic as needed. AI agents previously constrained by design can now operate more aggressively.
If you use Claude Code (Pro/Max/Team/Enterprise), the doubling of the 5-hour limit will roll out within a month. No immediate action is required, but developers who have been rationing sessions will soon have significantly more headroom.
For teams evaluating model providers, the compute capacity gap between major players is rapidly closing. Anthropic has existing compute deals with Amazon (5 GW), Google/Broadcom (5 GW), and Microsoft/NVIDIA.
