On May 6, Anthropic announced a change to Claude Code that developers had been requesting since the tool launched: doubled five-hour rate limits for Pro, Max, Team, and seat-based Enterprise subscribers. The peak-hours throttle, which had been quietly reducing available capacity during mornings and evenings in North America, was removed entirely for Pro and Max accounts. And for developers working through the API directly, the input token ceiling for Tier 1 accounts jumped from 30,000 tokens per minute to 500,000. The company also raised the output token ceiling from 8,000 to 80,000 tokens per minute.
The changes arrived because Anthropic had solved its compute problem. A week earlier, the company signed an agreement with SpaceX to lease the entire Colossus 1 data center in Memphis, Tennessee, taking on more than 300 megawatts of capacity and access to over 220,000 NVIDIA GPUs. Anthropic was ramping toward full utilization by the end of May, and that new supply translated directly into lifted ceilings for paying subscribers.
What Specifically Changed
Before May 6, a Claude Pro or Max subscriber running Claude Code through a long session would exhaust their five-hour allowance and then wait. During peak usage periods, which Anthropic had been managing by reducing effective limits for high-traffic hours, the wait came faster. Developers who wanted to run Claude Code continuously for extended tasks, or who were chaining multiple agent pipelines in parallel, regularly had to build manual pauses into their workflows.
After May 6, the five-hour limit itself did not disappear, but it doubled in effective capacity, and the peak-hours reduction was removed for Pro and Max. The practical result for most subscribers is that the kinds of sessions that previously required interruption and restart can now run to completion. For enterprise teams on seat-based plans, the doubling of the rate limit means that several engineers can run simultaneous coding sessions without collectively hitting the capacity ceiling as quickly as they did before.
Rate Limits Before and After
- Five-hour rate limit (Pro, Max, Team, Enterprise)Doubled
- Peak-hours throttleRemoved for Pro and Max
- API Tier 1 input (tokens/min)30,000 → 500,000
- API Tier 1 output (tokens/min)8,000 → 80,000
- New GPU capacity (Colossus 1)220,000+ NVIDIA GPUs
- New power capacity (Colossus 1)300+ megawatts
The SpaceX Deal Behind the Numbers
The Colossus 1 data center was built by xAI, Elon Musk's AI company, to train and run its Grok models. Anthropic's agreement with SpaceX gives Anthropic exclusive access to the entire facility, an arrangement that is unusual in the AI industry: one company's purpose-built infrastructure serving a rival's users at scale. Reports put the price at $1.25 billion per month, running through May 2029.
At that volume, Anthropic is committing more than $15 billion annually to a single data center. The bet is that the compute translates into capacity that subscribers notice, which drives retention and enables higher-tier plan sales. The rate limit increase is the most visible evidence, so far, that the Colossus capacity is being applied to the subscriber experience rather than held in reserve for training runs.
"Over the past few months, we've heard loud and clear that the most common thing you want is more usage." Anthropic, higher-limits announcement, May 6, 2026
What Developers Can Now Do Differently
The limit increases arrive as Anthropic has been pushing Claude Code toward more ambitious use cases. The claude -p command, used for headless automation and unattended agent pipelines, is popular with developers who want Claude running as a background process while they work on other things. With higher rate ceilings, those workflows can complete more before hitting a pause. Teams doing bulk code review across large repositories, continuous documentation generation, or automated test coverage expansion will notice the difference most.
One related change, announced separately and effective June 15, complicates the picture slightly. Starting on that date, the agent SDK credit pool moves to a dedicated monthly allowance rather than drawing from the same bucket as interactive conversations. The claude -p command, Claude Code GitHub Actions, and third-party tools authenticating via the Agent Client Protocol will consume from that separate pool. Pro subscribers get $20 of monthly credit, Max 5x users get $100, and Max 20x gets $200. The split is priced at full API rates, so heavy programmatic users may find the monthly credit exhausted before the billing cycle resets.
The combined picture is cleaner than it sounds. Interactive Claude Code sessions in the terminal benefit from the doubled rate limits. Background and automated pipelines move to a dedicated credit pool that gives Anthropic more flexibility to price those use cases independently going forward. For most developers whose primary workflow is interactive, the May 6 changes are straightforward improvements. For teams running heavy automation, the June 15 billing shift will require planning around how much monthly credit their pipelines actually consume.
Anthropic has said it expects to continue raising limits as Colossus capacity ramps further. Whether additional increases arrive before the end of the year will depend on whether utilization stays below the ceiling or whether subscriber growth from the IPO window creates new demand on the same infrastructure.
