On May 6, 2026, Anthropic announced an agreement with SpaceX to use all of the compute capacity at SpaceX's Colossus 1 data center, which it said adds more than 300 megawatts and over 220,000 NVIDIA GPUs within a month. On the same day, Anthropic doubled Claude Code's five-hour rate limits on paid plans and raised API rate limits for its Claude Opus models. The deal is notable for its speed: unlike multi-year chip commitments that come online in 2027 and beyond, this capacity was available within weeks, and developers felt the effect immediately as higher limits.

Key Facts#

  • Announced: May 6, 2026, by Anthropic.
  • Capacity: all of the compute at SpaceX's Colossus 1 data center, which Anthropic describes as more than 300 MW of new capacity and over 220,000 NVIDIA GPUs, available within the month.
  • Beneficiaries: Anthropic said the capacity would directly improve availability for Claude Pro and Claude Max subscribers.
  • Claude Code limits: five-hour rate limits doubled for Pro, Max, Team and seat-based Enterprise plans, effective immediately.
  • Peak hours: the peak-hours limit reduction on Claude Code was removed for Pro and Max accounts.
  • API limits: rate limits for Claude Opus models on the API were raised considerably.
  • Future option: Anthropic said it had expressed interest in partnering with SpaceX to develop multiple gigawatts of orbital AI compute capacity.

What Happened#

Anthropic presented the SpaceX agreement as the reason it could loosen usage limits right away. The three limit changes all took effect on the day of the announcement and targeted the company's heaviest users: people running long Claude Code sessions on subscription plans, and developers calling the most capable Opus models through the API. Removing the peak-hours reduction mattered because it had meant that the same plan delivered less Claude Code usage at the busiest times of day.

The post placed the deal within Anthropic's wider compute portfolio. It listed an up-to-5 GW agreement with Amazon that includes nearly 1 GW of new capacity by the end of 2026, a 5 GW agreement with Google and Broadcom that begins coming online in 2027, a strategic partnership with Microsoft and NVIDIA that includes $30 billion of Azure capacity, and a $50 billion investment in U.S. data centers with Fluidstack. Anthropic repeated that it trains and runs Claude on AWS Trainium, Google TPUs and NVIDIA GPUs, and that it continues to look for more capacity.

Anthropic also addressed where future capacity will go. Enterprise customers in regulated industries such as financial services, healthcare and government increasingly need in-region infrastructure for compliance and data residency, so part of the expansion will be international, including additional inference in Asia and Europe under the Amazon agreement. Anthropic said it will add capacity in democratic countries whose legal frameworks support investments of this scale and where the supply chain for hardware, networking and facilities is secure.

The most speculative element was orbital compute. Anthropic said only that it had expressed interest in working with SpaceX on multiple gigawatts of AI compute in orbit. It gave no timeline, cost or technical detail, so this should be read as an exploratory statement rather than a plan.

Background#

The deal came about two weeks after Anthropic's agreement with Amazon for up to 5 GW, in which Anthropic acknowledged that fast consumer growth had hurt reliability and performance for Free, Pro, Max and Team users, especially at peak hours. Most of Anthropic's large agreements deliver capacity over years. Capacity that arrives within weeks addresses the near-term gap that those agreements cannot.

The partner was also unusual. SpaceX is better known for rockets and satellites than for AI infrastructure. For how it came to own large AI data centers, see our coverage of the SpaceX and xAI merger.

Energy was part of the context too. On February 11, 2026, Anthropic pledged to cover electricity price increases that consumers face from its U.S. data centers, including paying for 100% of the grid upgrades needed to connect them. That pledge applies directly to facilities developed with partners for its own workloads. For capacity leased from existing data centers, such as Colossus 1, Anthropic said it was exploring further ways to address its effect on prices. In the SpaceX post, it said it would also explore extending the commitment to new jurisdictions.

Why It Matters for Developers#

For developers, this announcement is less about GPUs than about limits, which directly shape how you build on Claude.

  • Revisit your throttling settings. If your application throttles itself to stay under older Opus rate limits, check your current tier and limits in the Claude Console and adjust concurrency. Keep client-side limits configurable rather than hard-coded, because providers change them often.
  • Keep retries and fallbacks anyway. Higher limits reduce rate-limit errors but do not remove them, and capacity can tighten again as demand grows. Retries with exponential backoff and jitter, circuit breakers and fallback models remain essential. Our resilience guide covers these patterns with Polly.
  • Budget for heavier agent sessions. Doubled Claude Code limits make longer autonomous sessions practical on subscription plans. Teams adopting agentic workflows should agree on review practices before usage grows, as described in our guide to AI-assisted .NET development.
  • Expect variation across a mixed fleet. Claude now runs on three chip platforms across several providers. Anthropic does not document routing, so measure latency and error rates for your own traffic with OpenTelemetry, as shown in our LLM observability and cost guide.

What's Next#

Anthropic's supply kept growing after this announcement. On May 28, its Series H announcement described the SpaceX agreement as providing access to GPU capacity in both Colossus 1 and Colossus 2. Usage limits rose again later: with the launch of Claude Opus 5.5 on September 22, Anthropic increased five-hour usage limits on Pro, Max, Team and seat-based Enterprise plans and gave subscribers a rate limit reset they can save and use when they choose.

The orbital compute idea remains the biggest open question. Anthropic has not said whether it will pursue it, and no timeline has been published.

Sources#