Anthropic taps SpaceX Colossus 1 to raise Claude limits

Anthropic announced in a May 6 blog post that it has signed an agreement with SpaceX to use the full compute capacity of the Colossus 1 data center, adding more than 300 megawatts of new capacity and access to over 220,000 NVIDIA GPUs within the month (Anthropic blog post; Coindesk). Anthropic's announcement states the capacity increase allows three immediate changes: it is doubling Claude Code's five-hour rate limits for Pro, Max, Team, and seat-based Enterprise plans, removing the peak-hours limit reduction on Claude Code for Pro and Max accounts, and raising API rate limits for Claude Opus models, effective the same day (Anthropic blog post). The company also said it has expressed interest in pursuing multiple gigawatts of orbital AI compute with SpaceX, though no orbital agreement was announced (Anthropic blog post; TheNextWeb).

Large-scale additions of inference capacity generally boost throughput and reduce throttling for sustained workloads, rather than changing per-request model behavior. Observers note that adding several hundred megawatts and hundreds of thousands of accelerators primarily affects concurrency limits, sustained QPS, and the ability to serve enterprise SLAs at scale. QZ reports that Colossus 1 deploys dense arrays of NVIDIA H100, H200, and GB200 accelerators, according to SpaceX, which aligns with common hardware choices for high-throughput inference (QZ).

Reporting frames this Colossus 1 deal as one element of Anthropic's wider compute stack, which includes multi-gigawatt agreements with Amazon and Google, a strategic Microsoft-NVIDIA arrangement for Azure capacity, and a large-scale U.S. infrastructure investment with Fluidstack (Anthropic blog post; Coindesk; TheNextWeb). Coindesk and other outlets emphasize the timing of the announcement ahead of SpaceX's planned IPO as commercially notable, because named compute customers can strengthen a cloud/infra provider's pitch to investors (Coindesk).

For developers and enterprise practitioners, the reported changes translate into materially higher sustained usage ceilings for Claude Code workflows and larger API throughput for Claude Opus models. Anthropic says these changes are aimed at improving the experience of using Claude for its most dedicated customers (Anthropic blog post). Industry observers will watch whether the additional capacity also lowers queuing latency or yields faster autoscaling during demand spikes, though Anthropic has not published latency or error-rate targets tied to the Colossus capacity announcement (Anthropic blog post; Engadget).

Reporting by QZ cites local residents and activists raising pollution concerns about Colossus 1's buildout and notes coverage in CNBC about gas-burning turbines xAI used to supply power; those community and environmental issues are part of the broader public record around Colossus 1 (QZ; CNBC via QZ). The Anthropic announcement also highlights plans to add in-region infrastructure for regulated enterprise customers in some markets, but specifics about timelines and geographies were not included in the blog post (Anthropic blog post; TheNextWeb).

monitor published rate-limit ceilings and any changes to per-request latency or error rates in the weeks after capacity comes online. Industry context: watch SpaceX's public filings and investor materials for mention of commercial compute customers, and track whether other AI providers follow similar large-scale third-party hardware deals. Editorial analysis: track community and regulatory pushback around Colossus 1's power and emissions footprint, since local objections could affect capacity timing and operational constraints (Coindesk; QZ).