Anthropic Secures SpaceX Compute Deal, Doubles Claude Limits

Anthropic announced a landmark compute partnership with SpaceX that will grant the AI lab access to the entire Colossus 1 super‑computer in Memphis. The deal, valued at roughly $5 billion per year, is expected to add more than 300 MW of power and about 220 000 NVIDIA GPUs to Anthropic’s infrastructure. Within days, the company plans to start ramping Claude inference on the new hardware, a move that will immediately lift the 5‑hour rate limits for Pro, Max, Team, and seat‑based Enterprise accounts and eliminate the peak‑hour throttling that had previously applied to Pro and Max users. The Opus API will also see a substantial increase in rate limits, although weekly limits remain unchanged for now because only a small fraction of users hit them.

The announcement was made during Anthropic’s second annual developer event, “Code with Claude,” where the company showcased new managed‑agent capabilities such as Dreaming—cross‑session memory—and Outcomes, a rubric‑based grading system. The event highlighted the company’s commitment to safety and governance while also revealing an 80‑fold growth in Claude usage, prompting the need for the compute boost. Anthropic’s CTO Tom Brown emphasized that the partnership would enable faster model inference and higher throughput for enterprise workloads, and that limits would be adjusted further as more compute becomes available.

The partnership has drawn attention to environmental and supply‑chain concerns surrounding the Colossus 1 data centre. The facility has operated gas turbines without Clean Air Act permits, contributing to poor air quality and increased hospital admissions in the region. Elon Musk’s tweets framed the lease as a partnership that “ensures Claude is good for humanity,” but also warned that compute could be reclaimed if the AI causes harm, adding a new layer of risk for Anthropic. The deal underscores the growing scrutiny of AI data‑center practices and the political pressure on large‑scale compute providers.

For developers and enterprises that rely on Claude, the expanded capacity means higher throughput and fewer throttling constraints, but it also requires vigilance over model dependencies. The partnership signals a broader trend of frontier labs renting data‑center capacity from competitors to meet demand, and it positions Claude as a direct competitor to OpenAI’s offerings. As Anthropic continues to scale, the company will monitor usage patterns and adjust limits accordingly, ensuring that the platform remains robust and responsive to its growing user base.