Anthropic x SpaceX Deal: 7 Claude Code Limit Changes You Can Use Right Now
Anthropic's 300 MW SpaceX compute deal just doubled Claude Code session limits and removed peak-hour throttling. Here's what changed.
Anthropic Just Doubled Claude Code’s Session Limits — Here’s What Changed
Anthropic announced three concrete changes to Claude Code rate limits on the same day it held its first “Code with Claude” developer conference in San Francisco. The Claude Code 5-hour rate limit has been doubled for Pro, Max, and Team plans, effective immediately. If you’ve been hitting walls mid-session, the ceiling just moved.
This isn’t a vague “we’re improving infrastructure” announcement. The numbers are specific, the timing is tied to a real compute deal, and the changes affect how you should be building right now.
The SpaceX Deal Is Why This Is Happening Now
Anthropic signed a partnership with SpaceX that delivers 300 megawatts of capacity and access to more than 220,000 Nvidia GPUs. That’s the direct reason Anthropic could make these changes effective immediately rather than promising them on a roadmap.
The compute acquisition didn’t start with SpaceX. Anthropic has been building out infrastructure agreements across the industry — Amazon, Google, Broadcom, Microsoft, Nvidia, and Fluid Stack are all part of the picture. The day before the San Francisco conference, Anthropic also announced a joint venture with Goldman Sachs and Blackstone. SpaceX is the headline, but it’s one piece of a broader compute buying spree.
Other agents start typing. Remy starts asking.
Scoping, trade-offs, edge cases — the real work. Before a line of code.
The underlying problem was simple: demand outpaced supply. Over the past quarter, Claude experienced repeated outages. Anthropic briefly blocked new Pro plan signups from accessing Claude Code. They tested restricting API usage for tools like OpenClaw. They introduced peak-hour throttling that made your session limit burn faster during weekday mornings. These weren’t product decisions — they were triage.
The SpaceX deal is what ends the triage phase.
Three Changes, Broken Down
1. The 5-hour session limit is doubled across all paid tiers.
Pro, Max, and Team plans all get twice the Claude Code usage within each 5-hour window. This is the most direct change for anyone who builds with Claude Code daily. If you were hitting your limit after two hours of heavy agentic work, you now have roughly four hours before the same wall appears.
2. Peak-hour throttling is gone for Pro and Max.
About a month before this announcement, Anthropic introduced a system where your session limit depleted faster during peak hours — roughly weekday mornings. The logic was capacity management. That restriction has been removed for Pro and Max accounts. Your session limit now behaves consistently regardless of when you’re working.
3. Opus API rate limits increased significantly.
This one is for API users, not Claude Code subscribers. The Opus input token rate limit went from 30,000 tokens per minute to approximately 348,000 tokens per minute at tier 1 — roughly a 16x increase. The output token rate limit went from 8,000 tokens per minute to 80,000 tokens per minute, a 10x increase.
To put the output number in context: 8,000 output tokens per minute is a real constraint when you’re running parallel agents or building production pipelines. At 80,000 tokens per minute, multi-agent architectures that were previously impractical become viable. Five sub-agents each reading 50,000 tokens of context simultaneously is now a reasonable design choice rather than a rate-limit gamble.
Why This Matters for What You’re Building
The session limit change sounds like a quality-of-life improvement. It’s actually a structural shift in what you can build.
Before this change, if you wanted to run Claude Code automations as scheduled routines — things that fire overnight or on a weekly cadence — those automations competed with your daily knowledge-work sessions for the same 5-hour budget. You were choosing between autonomous workflows and interactive development. With double the limit, you can push more of those routine workflows into the background without gutting your interactive session time.
The 1 million token context window that Claude supports has been technically available for a while. But in practice, if you were hitting rate limits mid-session, loading a million-token context was a liability — you’d burn through your allowance faster and get cut off at a worse moment. With the new limits, that context window is actually usable in production without the interruption risk that made it impractical before.
For API builders specifically, the Opus rate limit increase changes the math on production agents. If you tried building an Opus-backed production system six months ago and gave up because you kept hitting the 8,000 output token ceiling, that constraint is gone. The wall you hit may not exist anymore.
What’s Buried in the Announcement
Built like a system. Not vibe-coded.
Remy manages the project — every layer architected, not stitched together at the last second.
The most interesting line in the SpaceX announcement isn’t about the 300 megawatts or the 220,000 GPUs. It’s this: Anthropic and SpaceX have expressed interest in developing multiple gigawatts of orbital AI compute capacity.
GPUs in orbit. That’s the long-term direction they’re signaling.
Anthropic’s stated reasoning is that terrestrial compute has a real ceiling — not just cost, but physical constraints around power, water, cooling, and community opposition to data center expansion. Putting compute in space sidesteps the land-use and utility problems that are already slowing down data center buildouts in certain regions. This isn’t happening in 2025, but the fact that it’s in the announcement at all tells you something about how Anthropic is thinking about the next five-plus years of infrastructure.
The “Code with Claude” conference itself is another signal. It was held in San Francisco, London, and Tokyo — and demand was high enough that they extended it, adding an extra day in each location. Anthropic also announced managed agents with webhooks, auto-dreaming, and multi-agent orchestration at the same event. Claude Code is clearly the product they’re betting on as the developer-facing flagship, not just a feature.
The Goldman Sachs/Blackstone JV announcement the day before the conference wasn’t coincidental. Anthropic is building enterprise-scale infrastructure and signaling to enterprise buyers that the compute is there to back it up.
The Token Management Problem Doesn’t Go Away
Doubled limits don’t mean unlimited. Context management still matters, and the habits that made you efficient under the old limits will make you more effective under the new ones.
If you’ve been defaulting to Sonnet or Haiku to stretch your session, you now have more room to use Opus on tasks that actually benefit from it. But that doesn’t mean using Opus indiscriminately. The 18 Claude Code token management hacks that applied before still apply — you’re just working with a larger budget, not an infinite one.
One pattern worth revisiting: the /model opus plan mode approach, where you use Opus for planning and a cheaper model for execution. With the new limits, you can afford to run Opus on more of the planning phase without immediately burning through your session. The Opus plan mode token-saving approach is still worth understanding even if the pressure to use it has eased.
For teams building multi-agent workflows, the API rate limit increases are the more significant change. Platforms like MindStudio handle this kind of orchestration across 200+ models and 1,000+ integrations — but if you’re building directly against the Anthropic API, the new Opus limits mean you can run parallel agent architectures that would have been throttled into uselessness under the old 8,000 output token ceiling.
What to Actually Do Differently
Retest workflows you abandoned. If you tried building an Opus-backed agent six months ago and gave up because of rate limits, the constraint may be gone. The specific number to check: were you hitting the 8,000 output token per minute ceiling? That’s now 80,000. Were you hitting the 30,000 input token ceiling? That’s now roughly 348,000 at tier 1.
Day one: idea. Day one: app.
Not a sprint plan. Not a quarterly OKR. A finished product by end of day.
Reconsider your model routing. If you built routing logic that pushed work to Haiku or Sonnet specifically to preserve session time, audit whether that logic still makes sense. Some of it will — Haiku is still faster and cheaper for tasks that don’t need Opus-level reasoning. But routing decisions made under scarcity constraints aren’t always the right decisions under normal conditions.
Move automations off your interactive session. The doubled limit makes it more practical to run Claude Code routines — scheduled workflows, background agents — without those automations eating into the time you need for active development. This is the structural change that matters most for anyone building production systems on top of Claude Code. The Claude Code agentic workflow patterns that previously required careful session budgeting now have more room to breathe.
Build the multi-agent architecture you’ve been deferring. Five sub-agents each reading 50,000 tokens of context simultaneously is now a reasonable design. The API output token limit was the binding constraint on parallel agent designs. At 80,000 tokens per minute, you have enough headroom to run real parallel workloads. If you’re thinking about how to structure that kind of system, the multi-agent setup with Paperclip and Claude Code is a practical reference for how agent orchestration actually gets wired together.
Think about what you’re building toward. For teams moving from prototypes to production infrastructure, the rate limit changes matter less than the signal they send. Anthropic is investing heavily in compute, signing enterprise deals, and positioning Claude Code as a production tool. If you’ve been treating Claude Code as a prototyping environment, the infrastructure is now there to support production use cases.
When you’re building full-stack applications on top of this kind of AI infrastructure, the spec-to-deployment question becomes relevant. Tools like Remy take a different approach to that problem: you write an annotated markdown spec, and it compiles into a complete TypeScript backend, SQLite database, auth, and deployment. The spec is the source of truth; the generated code is derived output. It’s a different layer of abstraction than Claude Code, but the two can complement each other — Claude Code for agentic development work, Remy for compiling a spec into a deployable stack.
The Bigger Picture
Anthropic spent the last quarter in a difficult position: a model that developers wanted to use heavily, and infrastructure that couldn’t keep up. The peak-hour throttling, the Pro plan signup restrictions, the OpenClaw API crackdown — these were all symptoms of the same problem.
The SpaceX deal, combined with the Amazon, Google, Broadcom, Microsoft, Nvidia, and Fluid Stack agreements, represents Anthropic’s answer to that problem. The rate limit changes are the first tangible output.
My read: Anthropic is treating Claude Code as its primary developer relationship, not just a product feature. The “Code with Claude” conference, the managed agents announcement, the infrastructure investment — it all points in the same direction. The question for builders isn’t whether to take Claude Code seriously as production infrastructure. It’s whether you’ve updated your assumptions about what’s possible now that the compute is actually there.
The 5-hour limit doubling is the most visible change. The removal of peak-hour throttling is the most immediately practical one. But the Opus API rate limit increases — 16x on input, 10x on output — are what make the architectural conversations worth having again.
Not a coding agent. A product manager.
Remy doesn't type the next file. Remy runs the project — manages the agents, coordinates the layers, ships the app.
If you want to understand how tokens work under the hood before you redesign your workflows around the new limits, the Claude Code source code leak analysis has some useful context on how the system actually manages context and session state.
The walls moved. Time to find out where the new ones are.
