Anthropic Adds SpaceX Compute, Raises Claude Usage Limits
Anthropic said it struck a compute partnership with SpaceX to expand Claude capacity and raise usage limits for several products. The company also said Claude Code limits are going up, while Anthropic leaders pointed to rapid growth and a larger push into managed agents.
Anthropic said it has signed a compute partnership with SpaceX that will expand the capacity behind Claude and immediately raise usage limits for several products. The announcement came during Anthropic’s second annual developer event, where the company focused less on a new model release and more on infrastructure, agent features, and product updates.
According to Anthropic, the deal will allow it to bring more compute online for Claude in the next few days. The company said the extra capacity will double Claude Code’s five-hour rate limits for Pro, Max, Team, and seat-based Enterprise users, remove peak-hour limit reductions for Pro and Max, and substantially increase Opus API rate limits.
xAI described the arrangement as SpaceXAI providing Anthropic access to Colossus 1 for additional Claude capacity. Anthropic CTO Tom Brown said Claude inference would be ramped up on Colossus “in the next few days,” signaling that the capacity increase is meant to show up quickly in customer-facing products.
The exact size of the deal was not formally spelled out in Anthropic’s main announcement, but multiple posts circulating around the event estimated the agreement at roughly 300 megawatts of capacity and about $5 billion per year. Those figures were presented as estimates in social posts, not as official numbers from Anthropic.
The compute announcement landed alongside Anthropic’s “Code with Claude” event, which included a keynote and sessions on Claude Code, GitHub-scale usage, and managed agents. Anthropic product lead Amol Avasare said weekly limits were not changing yet because only a small percentage of users hit them, while far more users were running into the shorter five-hour caps.
Anthropic executives also used the event to talk about the company’s growth and product direction. In a moderated session, Dario Amodei said the company had seen roughly 80x growth and discussed where he sees the market going next, including tiny teams, multiagent systems, and enterprise services.
Amodei said he still thinks 2026 could be the year of the one-person billion-dollar company. He argued that models are moving from writing code to helping people treat software engineering, and then business building, as tasks that can be done by very small teams.
He also described a future in which multiple agents work together, comparing it to “a country of geniuses in a datacenter.” On enterprise use, he said Claude Code can help individuals become more productive, but Anthropic increasingly wants to help whole teams and organizations work better together.
Another theme was bottlenecks. Amodei said he thinks about Amdahl’s Law, a principle that says the slowest part of a system limits the total speedup. In this context, he said the important problems are security, verifiability, and finding the bottlenecks in software engineering so they can be reduced.
The company’s public messaging framed the SpaceX deal as a direct answer to demand. Anthropic said the added compute is meant to translate into higher limits and more usable capacity for Claude Code and Opus, which are core tools for developers building with Anthropic’s models and agent products.
The event did not include a major new model release. Instead, Anthropic used the day to emphasize capacity, developer limits, and the platform direction for Claude as usage continues to rise.
