Anthropic: SpaceX becomes compute partner with 300 MW and doubled Claude Code limits

On May 6, 2026, Anthropic announced a new compute agreement with SpaceX that grants access to capacity in the Colossus 1 data center. This involves more than 300 MW of new capacity and over 220,000 NVIDIA GPUs, with Anthropic stating the entire package comes online “within a month.”

Alongside the partnership announcement, the company introduced concrete improvements for end users: five-hour rate limits for Claude Code have been doubled, and API rate limits for the Opus model have been significantly raised.

What does SpaceX bring to Anthropic?

SpaceX’s Colossus 1 is the data center to which Anthropic is migrating part of its inference and training growth. At 300+ MW and 220,000+ NVIDIA GPUs, this agreement ranks among the largest single compute engagements in the industry in 2026.

The partnership fits into Anthropic’s broader strategy, which includes up to 5 GW with Amazon (nearly 1 GW by end of 2026), 5 GW with Google and Broadcom (starting 2027), $30 billion in Azure capacity with Microsoft and NVIDIA, and $50 billion in US infrastructure investment with Fluidstack.

Anthropic also noted a long-term interest in developing “multiple gigawatts of orbital AI compute capacity” with SpaceX — a hint that the partnership has a phase beyond terrestrial data centers, with no concrete date attached.

How are Claude Code rate limits changing?

The rate limit (the ceiling on requests or tokens within a time window) for Claude Code is changing in two ways from May 6, 2026:

• Five-hour rate limits doubled for Pro, Max, Team and Enterprise plans
• Peak-hour limit reduction removed for Pro and Max accounts

This eliminates one of the most common frustrations among power users — occasional session blocking during busy periods of the day. For Pro and Max accounts, there is no longer a “silent penalty” that activated during European afternoon or US morning peak hours.

What about Opus API limits?

The announcement also includes a table of raised rate limits for Claude Opus models on the API. Anthropic does not state specific numbers in the announcement text, but the table explicitly covers new ceilings for all usage tier categories.

This is a direct consequence of the new capacity: when physical GPUs come online, Anthropic has room to ease the rationing that applied while the fleet was under load. Developer teams that encountered 429 rate limit errors on Opus should see relief from May 6 onward.

Why does this matter?

Anthropic has been positioning itself in recent weeks as a player with a diversified compute backbone — no single hyperscaler represents its own single point of failure. SpaceX entering that portfolio brings a fourth major compute partner (alongside AWS, Google/Broadcom and Microsoft/NVIDIA) and potentially the fastest time-to-online deployment phase (one month).

For Claude Code and Opus API users, the practical consequence is immediate: higher ceilings and less peak-hour throttling. For the industry, the news signals that compute partnerships in 2026 are being measured in hundreds of megawatts per contract, no longer individual clusters.