Higher Usage Limits for Claude and a Compute Deal with SpaceX: Everything You Need to Know
You know that feeling.
It’s 10:30pm. You’re in flow. Claude Code is humming along, refactoring a gnarly authentication module that you’ve been putting off for weeks. And then, bam — the usage limit hits. “Try again in 2 hours.” The flow dies. You stare at the wall. You contemplate learning woodworking instead.
For months, this was the Claude user experience in a nutshell: incredibly capable AI that you couldn’t quite rely on because the limits kept showing up at the worst possible moment.
That just changed, substantially.
On May 6, 2026, Anthropic announced a double announcement: higher usage limits across the Claude platform and a landmark compute deal with SpaceX that makes those higher limits possible. The company is doubling Claude Code’s five-hour rate limits, removing peak-hour restrictions for Pro and Max users, and significantly increasing API rate limits for the Claude Opus model family.
And the reason they can do all of this? They’ve secured access to all of the computing capacity at SpaceX’s Colossus 1 data center — more than 300 megawatts of power and over 220,000 NVIDIA GPUs.
Let’s break down exactly what changed, why it matters, and what it tells us about where AI is heading.
What Anthropic Actually Announced
Three specific changes took effect immediately on May 6, 2026. Not “rolling out gradually.” Not “coming soon.” Immediate.
1. Claude Code’s five-hour rate limits just doubled
This is the headline change and it affects every paid user: Pro, Max, Team, and seat-based Enterprise plans all get double the Claude Code usage within a five-hour window.
Why five hours? Anthropic uses a rolling usage window for Claude Code. Think of it like a “conversational budget” that resets as time passes, not at a specific hour. Previously, heavy users would hit the ceiling quickly during intense coding sessions. Now the ceiling is literally twice as high.
(Side note: I’ve always found the five-hour window a bit of a head-scratcher, it’s not a calendar day, not a work session, just… five hours. Anthropic hasn’t explained the specific logic, but it likely reflects how they provision GPU capacity across their user base.)
2. Peak hours restrictions on Claude Code are gone (Pro & Max)
You might not have even known this was a thing. Back in March 2025, Anthropic introduced “peak hours” limit reductions, basically, during the busiest times of day (roughly Pacific Time morning hours), Claude Code users on Pro and Max plans would hit their limit even faster than usual.
That’s now completely removed. No more getting penalized because everyone in California also wants to code at 10am.
(Honestly, this might be the most underrated part of the announcement. There’s something particularly frustrating about paying for a tool and then being told “sorry, it’s too crowded right now.” Removing this is a big quality-of-life improvement.)
3. Claude Opus API rate limits raised significantly
For developers and businesses calling the Claude Opus model via API, the rate ceilings, how many requests you can make per minute, have been raised “considerably.” Anthropic published updated rate limit tables showing higher allowed throughput across usage tiers.
If you’re a startup building on top of Claude or an enterprise running Opus for complex reasoning tasks, this is probably the most impactful change. Earlier in 2025, some API users were reporting conservative default limits, especially on the Opus tier, which is Anthropic’s most capable (and most compute-intensive) model.
The SpaceX Deal: Colossus 1 and 220,000 GPUs
So how is Anthropic paying for all this extra capacity? They’re renting the entire thing.
Anthropic signed an agreement with SpaceX to use all of the compute capacity at the Colossus 1 data center in Memphis, Tennessee. This single facility houses:
- More than 300 megawatts of power capacity (enough to power roughly 300,000 homes)
- Over 220,000 NVIDIA GPUs, including H100, H200, and next-generation GB200 accelerators
- Full access within the month of the announcement
Let that sink in. Anthropic didn’t just buy some extra capacity, they took over an entire world-class AI supercomputer, all at once. The company explicitly stated this additional capacity will “directly improve capacity for Claude Pro and Claude Max subscribers.”
Wait, isn’t Elon Musk meant to be Anthropic’s rival?
It’s a fair question. Elon Musk runs xAI (which builds Grok), has previously called Anthropic “misanthropic and evil,” and has been publicly critical of CEO Dario Amodei.
And yet here we are. Musk wrote on X that he “spent a lot of time last week with senior members of the Anthropic team” and was “impressed”, adding, with characteristic flair, that “nobody set off my evil detector.”
What actually happened? In early 2026, Musk merged xAI with SpaceX. xAI’s Colossus 1, which was underutilized for Grok, became available for lease. Meanwhile, SpaceX is heading toward a massive IPO (targeting a valuation between $1.75 trillion and $2 trillion), and they need marquee customers for their AI infrastructure business. Anthropic, desperate for compute, was the perfect tenant.
(Business, as they say, makes for strange bedfellows. The moral of this story: when the compute is there, principles get flexible.)
Why This Matters (More Than You Think)
You might be thinking: “Okay, so limits are higher. That’s nice, I guess.”
But this matters for a deeper reason: AI is hitting a physical compute wall, and whoever solves it first wins.
The AI compute bottleneck, explained simply
Imagine you’re running a restaurant. The kitchen can make 100 meals per hour, that’s your compute capacity. Every customer is an API call or a Claude Code session. When more customers show up than the kitchen can handle, you have to ration, that’s usage limits.
Now imagine the waitlist is growing 10x per year. You can either:
- Build more kitchens (data centers), which takes years,
- Raise prices to reduce demand, or
- Limit how much each customer can order.
Anthropic did all three in 2025, introducing weekly limits, peak-hour throttling, and maintaining premium pricing. The SpaceX deal is them aggressively choosing door #1: just get more kitchens, as fast as possible.
In a white paper from July 2025, Anthropic projected that the U.S. AI sector would need at least 50 gigawatts of electric capacity by 2028 to maintain global AI leadership, roughly the output of 50 large nuclear reactors.
300 megawatts from Colossus 1 isn’t 50 gigawatts. But combined with recent deals announced with Amazon (up to 5 GW), Google and Broadcom (5 GW), and Microsoft/NVIDIA ($30 billion in Azure capacity), Anthropic is quietly assembling one of the largest compute footprints of any AI company on the planet.
Who Benefits Most (and Who Doesn’t)
Power users of Claude Code: You’re the biggest winner here. Doubled limits plus no peak-hour throttling means freelancers, indie hackers, and engineering teams can rely on Claude for longer, uninterrupted sessions.
API developers on Opus: Higher rate limits unlock more ambitious application architectures, think real-time agent loops, multi-step reasoning pipelines, and customer-facing experiences that won’t break because you hit a ceiling.
Pro and Max subscribers: The removal of peak-hour limits alone makes the subscription more predictable. You pay monthly, you get consistent access regardless of what time you work.
Free tier users: This announcement doesn’t directly change anything for you. Free Claude still has relatively tight session quotas, and Anthropic has made no indication that’s changing.
Enterprise and Teams: You get the doubled Claude Code limits, but, importantly, the Opus API changes affect you too, depending on your contract structure. Check with your Anthropic account team about updated rate limit ceilings.
AI’s Infrastructure Arms Race
The Anthropic-SpaceX deal isn’t happening in a vacuum. It’s part of an unprecedented infrastructure buildout:
- OpenAI is heavily dependent on Microsoft Azure, a single-provider strategy that some analysts see as a risk.
- Anthropic, by contrast, now has multi-vendor compute across Amazon (AWS Trainium), Google (TPUs), Microsoft/NVIDIA (Azure/GPUs), and now SpaceX (Colossus 1).
- Global AI capital expenditure is expected to reach $480 billion in 2026, up 33% from 2025.
The companies that secure compute today will define the AI capabilities of tomorrow. Usage limits aren’t just a customer experience issue, they’re a strategic signal. When a company increases limits, it means their infrastructure is scaling. When they tighten them, it means they’re struggling to keep up.
(Here’s a thought: watch what happens to ChatGPT Plus limits or Gemini Advanced limits over the next quarter. If they stay flat or tighten while Claude expands, that tells you something about who’s winning the compute war.)
What Comes Next: Orbital Data Centers
Buried at the bottom of Anthropic’s announcement is a line that sounds like science fiction: the company has “expressed interest in partnering with SpaceX to develop multiple gigawatts of orbital AI compute capacity.”
Yes, AI data centers in space.
The logic is surprisingly practical. Data centers on the ground face three problems that get worse as they scale: power availability, cooling, and land. In orbit, you get:
- Unlimited solar power
- Passive radiative cooling (space is cold!)
- No real estate constraints
SpaceX has already filed paperwork with the FCC to explore this concept at scale, and they’re positioning themselves as the only organization with “the launch cadence, mass-to-orbit economics, and constellation operations experience to make orbital compute a near-term engineering program rather than a research concept.”
This is years away, realistically. But it’s a clear signal that Anthropic is thinking about compute not in terms of “next quarter” but “next decade.”
Frequently Asked Questions
Q: When did these changes take effect?
A: All changes went live on May 6, 2026, the same day as the announcement.
Q: Do I need to update anything to get the higher limits?
A: No. The doubled Claude Code limits, removed peak-hour restrictions, and Opus API rate limit increases are automatic for eligible plans.
Q: Which plans get the doubled Claude Code limits?
A: Pro, Max, Team, and seat-based Enterprise plans. Free plans are not included.
Q: Is Claude usage now unlimited?
A: No. Limits still exist, they’re just significantly higher. Anthropic has not removed limits entirely.
Q: What’s Colossus 1?
A: It’s a massive AI data center in Memphis, Tennessee, originally built by xAI (now part of SpaceX). It houses over 220,000 NVIDIA GPUs and 300+ MW of power capacity.
From Frustration to Flow
Here’s how I’d frame this whole thing, honestly:
AI usage limits are one of the quiet frustrations of 2025-2026. You invest time learning a tool. You build workflows around it. And then, just when it’s delivering value, a limit yanks it away. It’s like having a car that drives beautifully… but runs out of gas after 30 miles, every single time.
This announcement doesn’t make Claude unlimited. But it makes the “gas tank” significantly bigger, removes the speed limit during rush hour, and opens up more roads for developers to build on.
And it’s backed by something real: physical infrastructure. Not marketing promises. Not “we hear you, and we’re working on it.” Actual GPU clusters, coming online within weeks.
Is Claude worth trying (or upgrading) now? If you’re a coder or power user who was held back by the limits before, this is the best time to revisit it. The limits won’t disappear, but they just became a lot less likely to interrupt you when you’re in flow.
And isn’t that what we all want?
Comments
Post a Comment