Anthropic announced today (May 6, 2026) major capacity increases for Claude Code and the Claude API. The changes are effective immediately. The reason is the SpaceXAI Colossus 1 deal we covered yesterday - now translated directly into more tokens for paying users.
A Claude AI session - the developer experience that just got significantly more headroom. Photo: Wikimedia Commons.
1. What Changed for Claude Code
Effective immediately:
- 5-hour rate limits DOUBLED for Pro, Max, Team, and seat-based Enterprise plans
- Peak-hours limit reduction REMOVED for Pro and Max accounts
- Free tier unchanged
The peak-hours throttle was the change that frustrated developers most. Anthropic added it in late March 2026 in response to Claude Code's runaway adoption outpacing the company's compute supply. The Register reported users on the Pro and Max plans burning through their 5-hour quotas 30 to 60 percent into the window during US business hours.
That throttle is now gone for paying tiers.
2. What Changed for the Claude API
| Tier 1 API Limit | Before May 6 | After May 6 | Change |
| Max input tokens per minute | 30,000 | 500,000 | +1,567 percent |
| Max output tokens per minute | 8,000 | 80,000 | +900 percent |
Higher API tiers (2 through 5) received proportionally large increases as well. Anthropic published an updated rate-limit table on the Claude platform docs.
3. Why - The Compute Backstory
Elon Musk. The compute Anthropic just leased belongs to his newly merged SpaceXAI division. Photo: Wikimedia Commons, CC BY-SA.
The reason Anthropic can suddenly raise limits this dramatically is not better software or better caching. It is more raw GPUs.
| Colossus 1 (Memphis, TN) | Detail |
| GPUs | Over 220,000 NVIDIA H100, H200, and GB200 Blackwell |
| Power | Over 300 megawatts |
| Builder | xAI / SpaceXAI (built in 122 days originally) |
| Anthropic capacity | Full access to all of Colossus 1 |
| Status | Capacity available "within the month" |
The strategic significance: Anthropic - which competes head-on with Grok in the frontier-model market - has just become the largest customer of Elon Musk's AI infrastructure. The launch cost advantage that made SpaceX dominant in space is now being applied to AI compute supply. SpaceXAI is positioning itself as the picks-and-shovels supplier to the entire AI industry.
4. The Real-World Impact for Developers
If you build with Claude Code on a Pro or Max plan, today's changes are immediately visible:
| Workflow | Before | After |
| Long codebase refactor | Often hit 5-hour limit mid-task | 2x more headroom per window |
| Mid-day coding (peak hours) | Throttled since March 2026 | No throttle for Pro / Max |
| API integrations on Tier 1 | 30K input tokens / minute | 500K input tokens / minute |
| Long-running agentic tasks | Frequent rate-limit pauses | Substantially fewer pauses |
5. The Industry Picture
SpaceX Falcon Heavy. The same operational discipline that drove launch cost down by 10x is now being applied to AI compute supply. Photo: Wikimedia Commons, Public Domain.
This deal reshapes the AI compute landscape in three ways:
- Compute is being commoditized at scale. Anthropic does not own the GPUs that train Claude. SpaceXAI does. Anthropic is now a customer of compute - the same way Netflix is a customer of AWS. This is a structural shift in how frontier AI labs are organized.
- Musk has become a horizontal infrastructure player. SpaceXAI sells compute to Anthropic. xAI used Colossus to train Grok. SpaceXAI now does both - and rents capacity to its rivals.
- Orbital compute is the next frontier. The Anthropic-SpaceX agreement explicitly mentions "interest in working with SpaceX to develop multiple gigawatts of compute capacity in space." The orbital data center thesis is no longer hypothetical - it has its first paying customer.
SpaceX Starship. The vehicle that, in the next decade, may carry Anthropic's compute hardware to orbit. Photo: Wikimedia Commons, CC BY-SA.
6. The Numbers in One Frame
Claude Code: 5-hour limits 2x for Pro / Max / Team / Enterprise
Peak hours throttle: Removed for Pro and Max
API Tier 1 input: 30K to 500K tokens/minute (+1,567 percent)
API Tier 1 output: 8K to 80K tokens/minute (+900 percent)
Source of capacity: Colossus 1 (Memphis, 220K+ GPUs, 300MW)
Effective: May 6, 2026 (immediately)
7. What This Means for Frontier AI
For two years, the AI industry has been compute-supply-constrained. Frontier labs hoard GPUs. Customers wait in queues. Rate limits get tighter, not looser. Today's announcement is the first time in that two-year stretch that a major frontier lab has dramatically RAISED user limits because new compute genuinely arrived. That is what compute supply looks like when it actually shows up. And the source of that supply is, surprisingly, Elon Musk.
A Claude Opus session output. Long-context, high-token interactions like this are the workflows that most directly benefit from the new limits. Photo: Wikimedia Commons.
8. Action Items
- If you are a Claude Pro or Max user: The new limits are live. Re-run that long task that hit the wall yesterday.
- If you are an API customer on Tier 1: Check your dashboard - the rate limit should already be raised.
- If you are a Team or Enterprise admin: Notify your developers - no client-side changes required.
- If you have been considering Claude Max because of rate limits: The cost-per-token-of-headroom calculation just got significantly better.
Sources
- Anthropic Newsroom: Higher usage limits for Claude and a compute deal with SpaceX (May 6, 2026)
- PCWorld: Anthropic doubles Claude Code limits, thanks to a deal with SpaceX
- 9to5Google: Claude Code is getting higher usage limits, doubled for most users
- The Register: Anthropic admits Claude Code quotas running out too fast (March 2026)
- Jerry Cards: xAI Becomes SpaceXAI - The Merger Behind This Compute Deal