On May 6, 2026, Anthropic announced a deal with SpaceX for exclusive use of the Colossus 1 data center's compute capacity — over 300 megawatts and more than 220,000 NVIDIA GPUs available within the month. Simultaneously, the company immediately raised Claude Code and Claude Opus API usage limits, responding to longstanding user frustration.
Key highlights
- SpaceX Colossus 1 data center: over 300 MW and 220,000+ NVIDIA GPUs for Anthropic
- Claude Code five-hour rate limits doubled for Pro, Max, Team, and seat-based Enterprise plans
- Peak-hours limit reduction removed for Claude Code on Pro and Max accounts
- Significant API rate limit increases for Claude Opus models (details in platform documentation)
- Anthropic's total contracted compute: up to 5 GW from Amazon, 5 GW from Google/Broadcom, $50 billion with Fluidstack, $30 billion Azure capacity (Microsoft/NVIDIA)
The SpaceX deal and the scale of compute capacity
Anthropic has signed an agreement with SpaceX for full use of the Colossus 1 data center — the same facility SpaceX originally built for its own AI projects and Starlink operations. Over 300 megawatts of capacity with more than 220,000 NVIDIA GPUs is to be available to Anthropic within one month of the announcement. These additional resources will go directly to the infrastructure serving Claude Pro and Claude Max subscribers.
Under the same agreement, Anthropic expressed interest in co-developing orbital AI compute capacity with SpaceX targeting multiple gigawatts — a concept at an early stage, but one that signals strategic thinking about compute as global infrastructure.
The SpaceX deal joins a growing portfolio of Anthropic compute contracts: an agreement with Amazon for up to 5 GW (approximately 1 GW of new capacity by end of 2026), an agreement with Google and Broadcom for 5 GW coming online from 2027, a partnership with Microsoft and NVIDIA covering $30 billion of Azure capacity, and a $50 billion investment in AI infrastructure with Fluidstack.
Usage limit changes — what is changing today
First change: doubling Claude Code's five-hour rate limits for Pro, Max, Team, and seat-based Enterprise plans. This is a direct response to complaints from users heavily using Claude Code for programming work, who were hitting limits mid-session.
Second change: removing the peak-hours limit reduction for Claude Code on Pro and Max accounts. Previously, limits were automatically reduced during periods of high server load — a source of unpredictability for users working across different time zones.
Third change: significant increases to API rate limits for Claude Opus models. Specific values are available in Anthropic's platform documentation. This affects both enterprise users accessing the API directly and developers building applications on the model.
International expansion and energy policy
Anthropic has announced that part of its compute expansion will be built outside the US — particularly in response to compliance requirements from enterprise customers in financial services, healthcare, and government, who need regional infrastructure meeting local data residency rules. The Amazon agreement already includes additional inference infrastructure in Asia and Europe.
The company states that in selecting partners and locations it applies geopolitical criteria: prioritizing democratic countries with appropriate legal frameworks and secure hardware supply chains. Anthropic had previously committed to covering any consumer electricity price increases caused by its US data centers, and has announced plans to extend that commitment to new jurisdictions.
Why this matters
Doubling Claude Code limits is a change that users relying heavily on Claude as a programming tool will feel immediately. Rate limits have been among the most frequently cited complaints in developer communities — especially in the context of the growing market for agentic AI coding tools, where Claude Code competes directly with GitHub Copilot and Cursor.
The compute context matters more than the limit changes themselves. Anthropic is building a compute portfolio expected to reach tens of gigawatts by mid-decade — an ambition comparable to the largest players: Microsoft/OpenAI and Google/DeepMind. The SpaceX deal is both a compute transaction and a strategic signal: Anthropic is diversifying its compute providers beyond its primary cloud partners (Amazon and Google), reducing dependence on any single ecosystem. Orbital compute remains a distant vision, but the fact of its public announcement indicates that Anthropic thinks about compute as global infrastructure, not merely as a resource rented from cloud providers.
What's next
- By end of 2026, approximately 1 GW of new Amazon capacity is expected to become available for Anthropic — which should translate into further rounds of limit increases or API price reductions
- Google/Broadcom compute infrastructure under the 5 GW agreement will begin coming online from 2027
- Anthropic has announced regional infrastructure deployment in Asia and Europe under the Amazon agreement — without specifying exact dates or countries
Sources
- Anthropic — Higher usage limits for Claude and a compute deal with SpaceX
- Anthropic — Anthropic and Amazon compute partnership
- Anthropic — Google and Broadcom partnership for compute
