We need to talk about the Claude Code rate limits

· media ai · Source ↗

Summary based on the YouTube transcript and episode description.

Theo (t3.gg) argues Anthropic’s Claude Code peak-hour rate cuts are economically justified but communications are a cultural failure, not a malicious money grab.

  • Claude Max ($200/mo) was delivering up to $5,000/month in compute — a 25x subsidy that Anthropic can no longer sustain.
  • ~7% of users will hit session limits they never hit before; one user reported a workflow consuming 100% of their 5-hour limit instead of the usual 10%.
  • Anthropic’s annualized 2026 revenue run-rate is ~$14B; 500+ customers now spend $1M+ annually, up from a dozen two years ago; 8 of Fortune 10 are customers.
  • Anthropic and Google are both behind OpenAI on GPU ownership — Google is now financing a data center leased to Anthropic to close the gap.
  • The February ‘spring break 2x’ promotion was a controlled experiment to see if cheaper off-peak pricing would shift usage out of the 5–11am Pacific crunch window; it didn’t move the needle enough.
  • The rate limit change was implemented before it was announced, and the only public communication came from a single individual-contributor employee (Thoric) on personal Twitter, two hours after the window had already fired.
  • Internally, Anthropic allocates GPUs across three competing groups — research, product, and enterprise/API users — with enterprise generating the most immediate revenue and therefore winning during peak hours.

2026-04-03 · Watch on YouTube