Claude Code users got way more room to work without hitting a wall. On May 6, 2026, Anthropic doubled the usage caps on Claude Code (its AI coding tool), scrapped a daytime slowdown, and raised the limits on its most powerful model. Three changes for paying customers, all aimed at one complaint: people kept running out of headroom mid-task. The reason Anthropic could move this fast? A fresh deal with SpaceX for a huge data center.
Rate limits are just caps on how much you can use a tool in a set window of time. Here's what changed for paying Claude Code users:
- The five-hour limit doubled for Pro, Max, Team, and seat-based Enterprise plans.
- Pro and Max lose the peak-hours slowdown, the 8 AM to 2 PM ET window that had made afternoon work feel rationed.
- API limits for Claude Opus, Anthropic's heaviest model, jumped a lot, with the biggest gains on higher-tier accounts that kept hitting walls.
Together these roll back most of the squeeze that landed earlier in 2026, when tighter caps had Pro and Max users counting tokens with one eye open. TechWalrus covered that pinch in The Claude Code Subscription Squeeze Is Coming for Everyone. That pressure hasn't vanished. Anthropic just bought room to push it off.
Anthropic Higher Usage Limits Claude Compute Deal
The deal behind it
What unlocked it is a deal with SpaceX that hands Anthropic all the compute (the raw computer power that runs AI models) at SpaceX's Colossus 1 data center in Memphis, Tennessee. By Anthropic's own numbers, that's over 300 megawatts of new capacity and more than 220,000 NVIDIA graphics chips, with the bulk coming online within a month of the announcement.
If Colossus 1 sounds like an xAI project, that's because it was. Elon Musk merged xAI into SpaceX earlier in 2026, so the building Musk's AI lab poured, wired, and named now belongs to his rocket company. Yeah, it's weird. The practical bit for Anthropic? The thing is already built.
Sites like this don't go up in a few weeks. By xAI's own account, the first Memphis cluster took 122 days starting in 2024, then doubled to roughly 200,000 chips in another 92 days. Signing with SpaceX lets Anthropic skip to a finished cluster instead of waiting on a fresh build. So caps could move in early May 2026 instead of next year.
Not just SpaceX
The SpaceX deal stacks on top of every other compute commitment Anthropic signed in the months before:
- Up to 5 gigawatts with Amazon, nearly 1 gigawatt of it due by the end of 2026, with more than $100 billion of long-term spend committed.
- 5 gigawatts with Google and Broadcom for next-generation chips, starting in 2027.
- A Microsoft and NVIDIA partnership covering $30 billion of Azure capacity.
- A $50 billion infrastructure investment with Fluidstack inside the US.
Add Colossus 1 and you get a company that decided in early 2026 it would not run short on chips again. That 300-megawatt figure is what makes the new caps possible right away; the other deals are what make the next few years believable. Big difference. One pays off right away, the rest in 2027 and beyond.
Anthropic Higher Usage Limits Claude Compute Deal
What it means for users
If you pay for Claude Code on Pro, Max, Team, or seat-based Enterprise, the five-hour rolling limit just doubled. On Pro and Max, the afternoon slowdown is gone, so that early-afternoon dead zone should stop showing up. Hit the API directly on an Opus model and your limits are now a lot higher, biggest at the top tiers.
Teams running several AI agents at once, or doing long Opus runs over big codebases, get the change that actually moves the needle. Honestly, the rationing loosening up matters more day-to-day than any pricing-page headline.
Now the fine print hasn't budged. Top-tier models still cost real money per token, and Anthropic still nudges heavy enterprise users toward metered API billing instead of flat subscriptions. Push usage well past the new caps and the squeeze TechWalrus wrote about is still your future. Everyone else gets a real upgrade.
The space footnote
Buried at the bottom of the announcement is a line that got screenshotted fast: the company "expressed interest in partnering with SpaceX to develop multiple gigawatts of orbital AI compute capacity." Data centers in space, basically.
That's not a 2026 plan or a 2027 plan. Power and cooling are already the hardest limits on data centers down here, the sun never sets in orbit, and SpaceX is one of the few outfits with the launch rate to make space data centers more than a daydream. So the math isn't crazy. Wait, scratch that, the idea isn't crazy; making it real on any near timeline still is. Read it as a hint of where compute talk lands a decade out, not a roadmap.
CNBC Anthropic, Spacex Announce Compute Deal Includes
The bottom line
For a Claude Code subscriber, the move is simple. Open the editor, run that long task you kept putting off because of caps, and see if the wall still shows up. For everyone watching the AI race, Anthropic just grabbed 300 megawatts of headstart and signaled it won't fall behind OpenAI or Google.
The bills here are going to be enormous, and the per-token economics of frontier AI are still rough. No way around that. But in early May 2026, the meter on Claude Code finally moved in the user's favor.
