What Happened: Anthropic signed a deal to use every GPU at SpaceX’s Colossus 1 data center in Memphis. The immediate result for users: doubled Claude Code rate limits for all paid plans and removed peak-hour throttling for Pro and Max subscribers.
The Anthropic SpaceX deal announced on May 6, 2026 keeps getting more interesting as the specifics come into focus. The headline is straightforward: Anthropic now controls all available compute at SpaceX’s Colossus 1 data center in Memphis, Tennessee, gaining over 220,000 Nvidia GPUs and 300 megawatts of raw power.
The way I read it, the more telling part is why it happened. Dario Amodei admitted that Anthropic grew 80x in Q1 2026, obliterating their internal plan for 10x growth. When your growth outpaces your plan by 8x, you need compute yesterday, not next quarter.
What Did Anthropic and SpaceX Agree To
Anthropic secured exclusive access to one of the largest AI compute facilities on Earth. The deal covers all current capacity at Colossus 1, with onboarding of over 220,000 Nvidia processors within the month.
What is Colossus 1: SpaceX’s AI data center campus in Memphis, Tennessee, originally built by xAI. It houses over 220,000 Nvidia GPUs across 300+ megawatts of power capacity, making it one of the largest AI training sites in the world.
The GPU fleet spans three generations of Nvidia silicon: H100, H200, and next-generation GB200 accelerators. That 300 MW of power is roughly equivalent to powering 250,000 American homes, all dedicated to running Claude.
Anthropic’s annualized revenue run rate crossed $30 billion, up from $9 billion at the end of 2025. The demand for Claude API calls, Claude Code sessions, and enterprise deployments pushed their existing infrastructure past what anyone at the company expected.
| Detail | Number |
|---|---|
| GPUs gained | 220,000+ (H100, H200, GB200) |
| Power capacity | 300+ MW |
| Anthropic Q1 2026 growth | 80x (annualized, plan was 10x) |
| Revenue run rate | $30B+ (up from $9B end of 2025) |
| Valuation (secondary market) | $1T+ implied |
What I find most revealing is the scale mismatch between plan and reality. Anthropic planned for 10x growth and got 80x.
That is not a planning error you fix with incremental capacity additions. You fix it by signing a deal for an entire data center.
Why Is This a Bigger Deal Than a Compute Upgrade
The partnership between Anthropic and SpaceX represents the most unlikely AI alliance of 2026.
Elon Musk has been publicly combative toward both Anthropic and OpenAI for years, including an ongoing lawsuit against OpenAI. SpaceX leasing its flagship AI facility to a direct competitor of Musk’s own xAI would have been unthinkable six months ago.
Musk changed his tone. He wrote on X that he spent time with senior Anthropic staff and was “impressed.”
His exact words: “Everyone I met was highly competent and cared a great deal about doing the right thing.” From the person who once called Anthropic an “OpenAI clone,” that is a significant pivot.
From my perspective, the real story is not the GPU count. It is what the deal signals about the AI infrastructure race.
Anthropic’s Wall Street venture with Goldman Sachs and Blackstone was announced two days earlier. That was about financial backing.
This is about physical infrastructure. Anthropic is securing both the money and the silicon at the same time.
The space angle deserves attention too. The announcement mentioned that Anthropic has “expressed interest” in working with SpaceX to develop multi-gigawatt compute capacity in space. Whether that is marketing language or a real roadmap, the fact that it appeared in the press release tells you where the industry thinks the compute ceiling is headed.
For context, Anthropic’s valuation recently surpassed OpenAI’s on secondary markets, with shares trading at an implied $1 trillion or higher. The company is reportedly in talks to raise roughly $50 billion at a valuation above $900 billion. That kind of capital paired with Colossus-scale compute puts them in a position no other AI lab occupies right now.
What Does the SpaceX Deal Mean for Claude Code Users
Paid Claude Code users get doubled rate limits and zero peak-hour throttling, effective now. This is the most tangible, right-now outcome of the deal for anyone reading this.
Here is what changed for paid Claude Code subscribers:
- The 5-hour rate limits are doubled across Pro ($20/month), Max ($100 and $200/month), Team ($30/seat), and seat-based Enterprise plans
- Peak-hour usage cap reductions are removed for Pro and Max accounts, meaning no more silent throttling during high-traffic windows
- API token limits increased across all tiers, with Tier I input tokens jumping from 30,000 to 500,000 per minute
If you were hitting the wall halfway through a refactoring session, you now have twice the runway before throttling kicks in. The peak-hour fix is the one I would highlight, because before this deal, Claude Code would silently reduce your effective limits during busy windows. That friction is gone now.
Before: A Pro user running Claude Code during US business hours would hit rate limits roughly 40 minutes into a session, then wait for the 5-hour window to reset.
After: The same user gets double the limit with no peak-hour reductions, meaning uninterrupted sessions that last through a full coding block.
I would argue the API tier changes matter just as much for developers building on Claude. The Tier I jump alone is striking: input tokens per minute went from 30,000 to 500,000. That is a 16x increase for developers at the entry tier.
| API tier | Old input (tokens/min) | New input (tokens/min) | Old output (tokens/min) | New output (tokens/min) |
|---|---|---|---|---|
| Tier I | 30,000 | 500,000 | 8,000 | 80,000 |
| Tier II | 450,000 | 2,000,000 | 90,000 | 200,000 |
| Tier III | Not disclosed | 5,000,000 | Not disclosed | 400,000 |
| Tier IV | Not disclosed | 10,000,000 | Not disclosed | 800,000 |
If you are building multi-agent systems or running Claude Code for production work, those API limits were the real bottleneck. The SpaceX deal is the reason those numbers moved.
What Comes Next for Anthropic Compute
Anthropic is positioning itself as the first AI lab to solve the compute constraint at trillion-dollar scale. The SpaceX deal is one piece of a larger infrastructure push that includes financial partnerships announced the same week.
The $50 billion fundraise, if it closes at the reported $900 billion+ valuation, would give Anthropic the capital to replicate the Colossus playbook across multiple facilities. That is the pattern I would watch: secure massive compute first, then pass the capacity gains directly to users through higher limits and faster response times.
The space compute angle remains speculative. Multi-gigawatt capacity in orbit would solve the land, power, and cooling constraints that ground-based data centers face, but the engineering timeline is measured in years, not quarters. I would treat it as a signal of where Anthropic thinks the ceiling is, not something shipping anytime soon.
For Claude users, the immediate takeaway is simple. The limits that frustrated you last month are gone or dramatically expanded. Whether you are a Pro subscriber running Claude Code daily or an API developer building agent workflows, the SpaceX deal is already reflected in your account.
