Anthropic Partners With SpaceX's Colossus 1 and Raises Claude Usage Limits

A surprising deal between rival camps

Anthropic announced on May 6 that it has signed a compute agreement with SpaceX, giving it access to the full capacity of Colossus 1 - the xAI-built supercomputer cluster in Memphis, Tennessee. The deal is notable not just for its scale, but for who is involved. SpaceX acquired xAI earlier this year, and xAI builds Grok, a direct competitor to Claude. Elon Musk has publicly criticized Anthropic multiple times, calling it 'misanthropic' and 'evil.' That backdrop makes this a genuinely unusual business arrangement.

Musk appeared to soften his position after meeting with Anthropic's team. He posted on X that he was 'impressed' by the people he met and that 'no one set off my evil detector.' Whether that signals a lasting thaw or was purely transactional is an open question. What is clear is that the commercial logic won out.

What Anthropic is actually getting

The agreement gives Anthropic access to over 300 megawatts of compute capacity - more than 220,000 NVIDIA GPUs, including H100, H200, and GB200 accelerators - and that capacity is expected to be available within the month. Anthropic says this will directly benefit Claude Pro and Claude Max subscribers.

The Colossus 1 facility is xAI's largest data center to date, built in a former Electrolux factory in Memphis's Boxtown neighborhood. It has faced criticism from local civil rights groups over the use of natural gas turbines to power the site. Anthropic acknowledged that it is 'very intentional' about where it adds capacity, citing supply chain security and legal frameworks in democratic countries as key criteria - though it made no direct reference to the Memphis environmental concerns.

Beyond Colossus 1, Anthropic and SpaceX have expressed interest in jointly developing orbital AI compute capacity. SpaceX filed with the FCC in January to deploy a million-satellite orbital data center megaconstellation, though that project is still at an early stage.

What changes for users right now

Anthropic made three immediate changes to usage limits, effective May 6. Claude Code's five-hour rate limits are now doubled for Pro, Max, Team, and seat-based Enterprise plans. The peak hours limit reduction on Claude Code has been removed for Pro and Max users, meaning there is no longer a throttle during busy periods. And API rate limits for Claude Opus models have been raised considerably for developers building on the platform.

These changes address some of the most consistent complaints from Claude's heaviest users - particularly developers using Claude Code for extended sessions. The removal of peak-hour throttling is especially significant for teams in different time zones who hit limits during their normal working hours.

Part of a much larger compute buildout

The SpaceX deal is one piece of Anthropic's broader infrastructure strategy. The company has announced a 5 gigawatt compute agreement with Amazon that includes nearly 1 GW of new capacity by end of 2026, a 5 GW deal with Google and Broadcom that comes online in 2027, a Microsoft-NVIDIA strategic partnership covering $30 billion of Azure capacity, and a $50 billion investment in US AI infrastructure with Fluidstack. The Colossus 1 deal adds 300+ megawatts on a fast timeline - useful for near-term capacity, while the larger deals build out the longer-range foundation.

The timing also matters for SpaceX. The company is reportedly weeks away from filing its public S-1, targeting a valuation between $1.75 trillion and $2 trillion, with a roadshow expected in early June. Having Anthropic - a company approaching a $900 billion valuation in its current funding round - as a named compute customer strengthens SpaceX's pitch as an AI infrastructure business ahead of that listing.

The open question

Compute deals at this scale take time to fully absorb. The 300 megawatts from Colossus 1 is meaningful in the short term, but the real test is whether Anthropic can absorb capacity fast enough to keep pace with subscriber growth. Claude's usage has reportedly surged as xAI's Grok has seen declining engagement - a dynamic that makes access to Colossus 1, the very hardware built for a competitor, an almost cinematic reversal.

Anthropic also flagged international expansion as a growing priority, with regulated-industry customers in financial services, healthcare, and government needing in-region data residency. Its Amazon deal includes new inference capacity in Asia and Europe. As the capacity stack grows, the question shifts from whether Anthropic can get enough compute to whether it can deploy it coherently across a patchwork of partnerships.

Sources