Anthropic Suffers Global Outage: The Day Claude Went Down

Two days after announcing the largest fundraising in history, Anthropic faced the worst outage in its history. Claude was completely offline for 6 hours — from 09:34 to 15:42 UTC — affecting millions of users in 192 countries.

The Root Cause

The post-mortem published by Anthropic revealed an embarrassing cause for a company valued at $965B: a cascading failure in the inference orchestration system. A bug in a session database routing update caused the system to start allocating requests to nodes that were under maintenance. The domino effect took down all 24 inference clusters in 18 minutes.

Interestingly, the automatic backup system failed because it shared the same configuration database — a violation of the failure domain separation principle that any junior SRE engineer would know to avoid.

The Impact

• Lost revenue: Estimated at $48-52 million
• Users affected: 3.2 million requests lost or errored
• Enterprise clients: 47 contracts with violated SLAs, generating an estimated $120M in penalties
• Third-party impact: Companies like Jasper, Notion AI, and Replit that depend on Claude also had their products degraded

Anthropic's shares on the secondary market fell 3.8% on the day but recovered 24 hours later — the market appears to have a short memory for infrastructure incidents.

What This Means

This outage is a warning for the entire AI ecosystem. As companies build entire products depending on LLM APIs, the resilience of these providers becomes the resilience of your business. Anthropic has $65B fresh to fix the problem — but most companies have no plan B when their AI provider goes down. The incident accelerates two trends: the adoption of multi-provider strategies and investment in smaller models that run locally for fallback. Single-vendor dependency in AI is no longer acceptable.