For the past few weeks, Anthropic has continually upset devs with its “dumber” model, and by removing Claude Code access from some paid accounts. After securing lots of compute from SpaceX, could the reason have been to conceal capacity issues?

Pragmatic Engineer is a blog authored by Gergely Orosz, covering topics related to software engineering, career growth, and technology leadership. With a focus on practical advice and real-world experiences, Pragmatic Engineer offers  insights for developers and tech professionals. Developers can learn about software design principles, career advancement strategies, and industry trends through Pragmatic Engineer's articles and essays.

The Pragmatic Engineer

Anthropic has been frustrating developers over recent weeks by silently degrading Claude Code performance, removing access from paid accounts, and banning corporate users. A likely explanation is severe capacity constraints — Anthropic's CEO revealed 80x growth vs. a planned 10x. To address this, Anthropic struck a deal with SpaceX/xAI to lease the entire Colossus 1 data center (220,000+ NVIDIA GPUs). Following the deal, Anthropic announced doubled Claude Code usage limits and removed peak-hour restrictions. The arrangement is ironic: xAI builds Grok, a direct Claude competitor, and Anthropic previously banned xAI developers. The deal may signal that Grok is losing the AI model market race, with xAI sitting on idle capacity. The shared enemy — OpenAI — may be the underlying motivation for both parties.

The Pulse: Did capacity shortages turn Anthropic hostile to devs?