Anthropic + SpaceX: The Compute Deal That's Doubling Claude Code Limits and Reshaping AI Infrastructure

This isn't just a headline. For Claude Code users who've been hitting rate limits constantly, this is immediate relief. And for Anthropic, it signals something bigger: the company is moving beyond infrastructure constraints and into competitive dominance.

Starting immediately on May 6, Anthropic made three concrete changes:

1. Claude Code limits doubled. Pro and Max subscribers now get 2x the five-hour window rate limits. If you hit limits in 5 hours before, you can now do roughly twice as much work before being rate-limited again.

2. Peak hours restrictions removed. Anthropic previously reduced Claude Code limits during peak hours (4 PM - 10 PM PT). That's gone. You now get your full rate limit 24/7, no time-of-day penalty.

3. API limits raised for Opus. If you're using Claude through the API (especially Opus), you got a boost too. Token limits per minute and requests per minute both went up.

Six months ago, you couldn't have predicted this. Elon Musk was publicly critical of Anthropic. In February, he tweeted "Anthropic hates Western Civilization." SpaceX was backing xAI. Anthropic was backing... well, mostly itself and its partners.

What changed? Compute. And economics.

Anthropic's demand for compute exploded. Claude Code adoption accelerated. Users shifted from single-agent tasks to multi-agent workflows (which burn through tokens faster). Anthropic's own infrastructure deals—with Amazon, Google, Microsoft, NVIDIA—take months or years to come online.

Meanwhile, Elon spent time with Anthropic's leadership and apparently came away convinced they're not evil. Last week, Musk tweeted: "I spent a lot of time last week with senior members of the Anthropic team to understand what they do to ensure Claude is good for humanity and was impressed. No one set off my evil detector."

SpaceX, on the other hand, had a massive capacity problem: Colossus 1 is enormous (300MW, 220k GPUs) but was built primarily for xAI's Grok. Grok isn't consuming all that capacity. So Colossus 1 was partially dark—expensive power running unused hardware.

For SpaceX, this deal is brilliant: They monetize idle capacity and turn a data center cost into revenue. For Anthropic, they get immediate compute without waiting. For xAI (Musk's AI company), it's awkward but fine—they still have access to Colossus and SpaceX keeps the revenue.

What's really significant here is Anthropic's infrastructure strategy. In the past 6 months, Anthropic has announced or finalized:

• SpaceX Colossus 1: 300MW, 220k GPUs, starting now • Amazon: Up to 5GW, nearly 1GW by end of 2026 • Google + Broadcom: 5GW, starting 2027 • Microsoft + NVIDIA: $30B Azure capacity • Fluidstack: $50B in American AI infrastructure

That's not just a lot of compute. That's control over the supply chain. By 2027, Anthropic will have more dedicated compute capacity than any other AI company except maybe Google (internally) or Microsoft (for Azure).

And Anthropic is doing something else smart: diversification. SpaceX (cooling, renewable energy adjacent), Amazon (cloud ecosystem), Google (TPUs, long-term), Microsoft (Azure, consumer reach). Anthropic isn't betting on one vendor or one hardware platform.

Here's the wildcard in the SpaceX deal: Anthropic and SpaceX expressed mutual interest in "partnering to develop multiple gigawatts of orbital AI compute capacity."

Translation: They're exploring satellite-based data centers. Sounds sci-fi? Maybe. But SpaceX has Starship. Anthropic has money. And the technical constraints (power delivery, thermal management, latency) aren't physics-breaking—they're engineering problems.

If orbital compute becomes real in the 2027-2028 timeframe, Anthropic could effectively escape terrestrial power and land constraints. That's not just infrastructure advantage—that's competitive dominance.

For users: The immediate win is higher limits and no more peak-hour restrictions. For Claude Code professionals, this is relief. For casual users, it doesn't change much—you weren't hitting limits anyway.

For developers: Anthropic just signaled that limits are temporary. Build aggressively with Claude. The constraints that were there in Q1 2026 are already loosening.

For competitors: This is a shot across the bow. OpenAI has Azure. Google has internal compute. Anthropic just went from "major player" to "infrastructure player." If the SpaceX partnership succeeds, OpenAI and Google need to respond or risk being outcomputed.

For the cloud giants: AWS, Google Cloud, and Azure suddenly have a customer (Anthropic) who's hedging by spreading compute across multiple providers. That's smart play but also signals that cloud lock-in isn't as sticky as it used to be.

If you use Claude Code: Update today. The limits have doubled. You might be able to do in 5 hours what you couldn't do in 10 hours yesterday.

If you use Claude through the API: Check your token/request limits in the console. You likely got a raise.

If you've been considering Claude for a project: Now's the time. Not because of Colossus 1 specifically, but because it signals Anthropic is investing heavily in staying competitive on capacity. That's good news for long-term reliability.

The SpaceX deal also signals something harder to measure: confidence. Anthropic locked in infrastructure deals with the biggest names in cloud and hardware. They're not panicking. They're planning for scale. That matters when you're choosing which AI platform to build on.