We need to talk about the Claude Code rate limits

TL;DR

Anthropic secretly throttled Claude Code usage during peak hours (5–11am PT) because they're severely GPU-constrained, then communicated it poorly via a single employee's tweet.

Key Points

• 1.Claude Code subscribers were getting $5,000 of compute for $200/month. This 25x subsidization was unsustainable, and Anthropic is now restricting usage during weekday peak hours (5am–11am PT) while keeping weekly limits unchanged.
• 2.Around 7% of users will hit session limits they never hit before. One user reported their 5-hour limit was consumed entirely by a workflow that normally used only 10% — suggesting up to a 10x effective reduction during peak hours.
• 3.Anthropic is in a severe GPU shortage driven by three competing internal groups. Research, product teams, and paying users all fight over a fixed GPU pool, with enterprise customers prioritized during peak hours because they generate direct revenue.
• 4.Anthropic was relatively slow to buy compute compared to competitors. OpenAI reportedly bought all available compute aggressively; Anthropic is now relying on Amazon and Google-financed data centers to catch up, meaning much of their capacity isn't even owned by them.
• 5.The spring break '2x off-peak' promo was actually a data-gathering experiment. Anthropic tested whether doubling off-peak limits would shift usage away from the 5–11am window; when it didn't work, they moved to forced peak-hour restrictions.
• 6.Anthropic's infrastructure reliability is alarmingly poor. They've dropped to roughly 1-nine (90–95%) uptime, with near-daily outages — including a 6-hour reporting outage and a 4-hour login outage within consecutive days in February.
• 7.The announcement was made after the window ended and only on a personal Twitter account. Thoric, a single DevRel employee with no team, tweeted the change at 12:45pm PT — after the 11am window — with no official Anthropic account post, dashboard notice, or in-product notification.
• 8.Anthropic's core problem is cultural, not just computational. As a research-first company, they lack communications leadership — Thoric alone handles all public-facing crisis comms with no team, while OpenAI deploys 20+ people for similar situations, and Anthropic's lawyers reportedly block many of his posts.