Anthropic Out of Compute: OpenAI's GPU Bet Was Right

My notes

Summary

Authority Hacker claims Anthropic has run out of compute capacity, forcing the recent Claude rate-limit cuts and blocking release of stronger (more token-hungry) models. OpenAI’s allegedly “overinvested” GPU buildout turned out to be the right call; Anthropic is now scrambling to sign Amazon and other compute deals that won’t come online until end of year.

Key Insight

• Rate limit reductions on Claude are not policy decisions - they’re capacity rationing
• Stronger frontier models cost more tokens per query; without spare compute headroom, releasing them would worsen the rate-limit pain
• OpenAI’s pre-buy of GPU capacity, widely criticized as overinvestment, is now a strategic moat - they can ship heavier models without throttling users
• Anthropic’s Amazon deal (and others) is reactive, not proactive - lead time on new compute is months, not days
• Implication: in 2026 the bottleneck for AI labs isn’t talent or model quality, it’s GPU contracts signed 12-18 months earlier
• Source caveat: this is a 31-second TikTok clip from a marketer, not insider reporting - treat as one perspective, not confirmed fact