On May 6, 2026, Anthropic announced it's doubling Claude Code rate limits for paid users and removing peak hour throttling entirely. The changes took effect immediately, backed by a new compute deal giving Anthropic access to SpaceX's entire Colossus 1 data center in Memphis, Tennessee.
TL;DR
- Claude Code's 5-hour rate limits doubled for Pro, Max, Team, and Enterprise plans
- Peak hour restrictions eliminated for Pro and Max users
- API rate limits for Claude Opus models increased substantially (Tier 1 saw 1500% input token increase)
- 300MW+ capacity from SpaceX's Colossus 1 (220,000+ NVIDIA GPUs) coming online within the month
- Weekly limits remain unchanged—watch your weekly usage carefully
What Changed for Claude Code Users
Rate Limit Increases by Plan
Pro Plan: 10 hours → 20 hours per month (5-hour sessions doubled) Max Plan: Proportional increase to Pro (exact hours not disclosed) Team Plan: 5-hour sessions doubled Enterprise (seat-based): 5-hour sessions doubled Free Plan: No changes
Important caveat: While 5-hour limits doubled, weekly limits did not increase. If you were previously using 10 hours spread across a week, you'll still hit the weekly ceiling even though individual sessions can now run longer.
Peak Hours Restrictions Removed
Previously, Pro and Max users faced reduced limits during peak usage hours (typically 9 AM - 5 PM Pacific). That throttling is now completely eliminated.
What this means in practice:
# Before May 6, 2026
# 9 AM PT: Hit rate limit after 3 hours (peak hours reduction)
# 6 PM PT: Full 5-hour limit available
# After May 6, 2026
# Anytime: Full 10-hour limit per session (Pro)
# No time-based throttling
For developers in US/CA/UK time zones who work during business hours, this is massive. You no longer need to schedule intensive coding sessions for evenings or weekends.
API Rate Limit Increases
Anthropic "substantially" raised Opus API limits. Published tier changes show:
Tier 1:
- Input tokens per minute: 1500% increase
- Output tokens per minute: 900% increase
import anthropic
client = anthropic.Anthropic()
# Before: Would hit 429 errors on high-throughput endpoints
# After: Higher ceilings for production workloads
response = client.messages.create(
model="claude-opus-4-7",
max_tokens=4096,
messages=[{"role": "user", "content": prompt}]
)
# Less need for complex rate limit handling
For production applications serving multiple users concurrently, this reduces the need for request queuing, caching layers, or switching to cheaper models for non-critical paths.
The SpaceX Colossus 1 Deal
Anthropic secured access to all compute capacity at SpaceX's Colossus 1 data center in Memphis, Tennessee. This is one of the largest AI compute facilities in the world.
By the Numbers
- 300+ megawatts of power capacity
- 220,000+ NVIDIA GPUs (H100, H200, GB200 accelerators)
- Within the month deployment timeline (announced May 6, online by early June)
- Full capacity lease (Anthropic gets everything, not shared)
To contextualize 300MW:
- Powers approximately 200,000 homes
- Runs millions of GPU hours monthly for AI inference and training
- Represents nearly half of SpaceX's ~500,000 GPU fleet
Why This Deal Matters
This isn't a 2027 capacity commitment. It's infrastructure that's already racked, cabled, and provisioning within weeks. That speed is unusual in data center deployments, which typically take 12-24 months from contract to production.
Anthropic explicitly stated the Colossus capacity will "directly improve capacity for Claude Pro and Claude Max subscribers." Translation: the rate limit increases are backed by actual silicon, not future promises.
The Broader Compute Stack
Colossus 1 joins Anthropic's other recent infrastructure deals:
- Amazon: Up to 5GW agreement, nearly 1GW online by end of 2026
- Google + Broadcom: 5GW agreement, coming online in 2027
- Microsoft + NVIDIA: $30 billion Azure capacity partnership
- Fluidstack: $50 billion American AI infrastructure investment
Total committed capacity across these deals: over 15 gigawatts. For perspective, a typical nuclear power plant produces about 1GW.
Orbital Compute (Future)
Anthropic "expressed interest" in partnering with SpaceX to develop multiple gigawatts of orbital AI compute capacity. SpaceX filed FCC paperwork in January 2026 for a million-satellite orbital data center constellation.
This is not part of the current deal—it's a future exploration. But it signals Anthropic is thinking beyond terrestrial infrastructure for long-term compute needs.
How to Maximize These New Limits
For Individual Developers
Audit current usage: Check your Claude dashboard to see how close you were to previous limits. If you regularly hit 80-90% of monthly hours, you now have room to expand.
Plan longer sessions: 20-hour monthly budgets support extended workflows:
- Multi-file refactoring across a codebase
- Full-day pair programming sessions
- Learning new frameworks with AI assistance
- Prototyping multiple architectural approaches before committing
Remove self-throttling: If you were rationing Claude Code usage to avoid hitting limits, stop. The ceiling moved.
For Teams
Reevaluate workflows: Tasks you previously couldn't justify (too many Claude hours) may now be economically viable.
Test API-heavy features: If you've been holding back on Opus calls due to rate limits, now's the time to load test:
// Example: Batch document processing
import Anthropic from '@anthropic-ai/sdk';
const client = new Anthropic();
async function processDocuments(docs: string[]) {
// Higher limits = batch more concurrently
const results = await Promise.all(
docs.map(doc =>
client.messages.create({
model: 'claude-opus-4-7',
max_tokens: 2048,
messages: [{role: 'user', content: `Analyze: ${doc}`}]
})
)
);
return results;
}
Monitor weekly limits: The 5-hour session limit doubled, but weekly caps didn't. If your team front-loads usage early in the week, you'll still hit weekly ceilings.
The Musk-Anthropic Tensions
This deal is surprising given the public history. Elon Musk previously called Anthropic "woke," "misanthropic," and "evil," and wrote in February 2026 that the company "hates Western Civilization."
After spending the past week with Anthropic leadership, Musk's tone shifted dramatically. He posted on X: "Everyone I met was highly competent and cared a great deal about doing the right thing. No one set off my evil detector. So long as they engage in critical self-examination, Claude will probably be good."
Musk also noted SpaceXAI (formerly xAI, merged with SpaceX earlier this year) had already moved training workloads to Colossus 2, freeing Colossus 1 for Anthropic. He added SpaceX reserves the right to reclaim compute if Anthropic's AI "engages in actions that harm humanity."
For context: Anthropic is currently suing the Trump administration after the Pentagon declared the company a supply chain risk and blacklisted it from military contracts in March 2026. That litigation is ongoing.
FAQ
Do existing Pro subscribers get the new limits automatically?
Yes. All changes took effect May 6, 2026 for active subscriptions. No action needed on your part.
Did pricing change with the rate limit increase?
No. Per-token API pricing remains unchanged. Rate limits increased, but you still pay the same per token consumed.
Are there still any peak hour restrictions?
No peak hour restrictions remain for Pro and Max plans. Team and Enterprise plans also benefit from this change.
When does the Colossus capacity go live?
Anthropic stated "within the month" in their May 6 announcement. Expect phased rollout through early June 2026 as infrastructure is provisioned and tested.
Will this affect Claude on claude.ai web chat?
These changes specifically target Claude Code (the coding assistant) and API usage. The web chat interface has separate usage policies, though infrastructure improvements benefit all Claude products.
What about the weekly limit?
Weekly limits did not increase. Only the 5-hour session limits doubled. Watch your weekly usage—if you double your session length, you'll hit weekly caps faster.
Final Thoughts
Doubling rate limits isn't just a nice-to-have—it fundamentally changes what's economically viable to build on Claude. Projects that required careful optimization from day one now have runway to prove value before hitting infrastructure constraints.
The speed of the Colossus deployment (weeks, not years) shows how seriously Anthropic is taking capacity constraints. When AI companies can secure 300MW and 220,000 GPUs with month-long deployment timelines, we're past the experimental phase of AI infrastructure.
For developers and teams building on Claude, this is the green light to expand usage. The ceiling just moved, and it's backed by real silicon coming online imminently.
If you're planning AI-powered features and want help architecting them to scale with these new limits, let's talk. I work with teams building production LLM integrations and can help you navigate the economics and technical trade-offs of scaling on Claude's API.
