Anthropic partners with SpaceX for computing power: secures the entire Colossus 1 facility—220k GPUs—and Claude lifts its usage limits

Anthropic announced on May 6 that it has reached a compute-power collaboration agreement with SpaceX, which will use all the compute capacity of SpaceX’s Colossus 1 data center—more than 300MW of added capacity, including over 220,000 Nvidia GPUs—with full deployment expected to be completed within a month. In a separate official announcement, Anthropic also announced that it is raising multiple Claude usage limits effective immediately: the Claude Code usage caps on the Pro/Max/Team/Enterprise plans are directly doubled every 5 hours, and the reductions in limits for Pro and Max users during peak hours are also removed.

Colossus 1: 220,000 Nvidia GPUs, including three generations of accelerators—H100, H200, and GB200

Core configurations of the Colossus 1 data center:

Over 220,000 Nvidia GPUs, with a mixed deployment of three generations of accelerators: H100, H200, and GB200

IT capacity of more than 300MW, expected to be fully available to Anthropic within a month

Use cases include large language model training, multimodal systems, scientific simulations, and generative AI workloads

“Full deployment within a month” is the most talked-about detail of this contract—compared with AI data centers built from scratch that often take 18–24 months, Anthropic is compressing the go-live timeline into units of months by directly leasing existing facilities. For Claude users, this means improvements in service experience brought by compute expansion can be felt almost immediately.

Claude usage limits are simultaneously loosened: Pro/Max/Team/Enterprise all benefit

The added compute power directly translates into the end-user Claude experience:

Claude Code: the usage cap every 5 hours is directly doubled on the Pro, Max, Team, and Enterprise plans

For Pro and Max users, the removal of reduced limits during peak hours means limits return to the level of non-peak hours

A significant increase in API rate limits for the Claude Opus model

Asia and Europe inference infrastructure expands in parallel, and regulated-industry enterprise users will gain access to local infrastructure

Over the past six months, Claude users have continued to report that “Claude Code gets rejected every 5 hours” and “throttling during peak hours” are core pain points—this adjustment amounts to a positive response to those reports. For Anthropic’s business strategy, loosening usage limits is also a key move to win enterprise users—Pro, Max, Team, and Enterprise are the company’s main paying cohorts.

Concurrent compute deployment and the future plan for “orbital AI compute”

The SpaceX contract is not an isolated case, but the latest step in Anthropic’s overall compute expansion. Here’s what the official announcements summarized:

Amazon: up to 5GW, with nearly 1GW expected online by end of 2026

Google + Broadcom: 5GW, expected to come online gradually starting in 2027

Microsoft + Nvidia: $30 billion Azure compute commitment

Fluidstack: $50 billion investment partnership

SpaceX: all of Colossus 1 compute capacity, plus an intent to cooperate on “orbital AI compute”

“Orbital AI compute” is the more future-looking part of the announcement—Anthropic said it has expressed interest in working with SpaceX to develop “multi-GW orbital AI compute capacity.” The original wording states that the compute required for next-generation systems has “exceeded the range of what can be supported by the power, land, and cooling capacity available on the surface.” This direction is still at the intent stage and has not yet moved into contract-level commitments.

What to watch next: Colossus 1’s compute is expected to be deployed within a month, and then changes in Claude clients’ throughput and peak-hour experience will be the direct validation indicators. This contract is also worth tracking as the first large-scale infrastructure-level cooperation announcement between Anthropic and SpaceX after heightened expectations from capital market IPO news—whether it will later expand into specific commitments for orbital compute is an event to watch.

This article, “Anthropic takes SpaceX compute: secures an entire Colossus 1 with 220,000 GPUs and Claude’s limits removed,” first appeared in ChainNews ABMedia.