Claude.ai and Claude Code Down Three Hours After Database Maintenance
Summary
- • Claude.ai and Claude Code suffered a ~3-hour outage on March 11, 2026
- • Root cause: primary application database I/O degradation following routine maintenance
- • Claude API remained fully operational throughout the incident
- • Outage coincided with Anthropic's fastest growth period, with DAUs reportedly up 180%
Details
Claude.ai and Claude Code outage lasted ~3 hours (14:17–17:11 UTC)
The incident affected login, session management, and general performance for Claude.ai and Claude Code users. The Claude API was explicitly called out as unaffected, meaning programmatic integrations continued working while end-user products were down.
Primary database suffered severely degraded I/O after routine maintenance
Official status attributed the outage to database I/O degradation triggered by a routine maintenance event — a classic post-maintenance regression that is difficult to catch in staging due to differing load profiles.
Recovery compounded by thundering herd and cache cold-start issues
Quick fixes risk cascading failures; restarting under production load triggers thundering herd and cache pre-warming problems; staggered rollouts extend the window before full recovery can be confirmed.
Anthropic DAUs reportedly up 180%, paid subscriptions doubling at time of outage
SimilarWeb data cited in HN discussion attributes growth to a Streisand Effect from Pentagon-related news. Higher user volume means each hour of downtime affected a significantly larger base than prior incidents.
Developers set up dual-provider API failover in response to outage
Some developers configured a primary and secondary AI API provider to eliminate single-point-of-failure risk — a practical infrastructure response to the reliability concern the outage exposed.
API and consumer product infrastructure paths are separate
Keeping the API unaffected while consumer products were down highlights that Anthropic's API tier doesn't share the same database dependency as Claude.ai and Claude Code — relevant for enterprise reliability assessments.
Infrastructure = outage/ops event, Tech Info = technical detail or root cause, Stat = quantitative data point, Strategy = developer/vendor response, Insight = analytical observation
What This Means
A routine database maintenance task took down Claude.ai and Claude Code for three hours, exposing how a single shared dependency can knock out consumer-facing products even when the underlying API stays healthy. The incident landed at an unusually sensitive moment: Anthropic is in the middle of its fastest-ever user growth, meaning the outage touched more users than any prior event. For developers, the practical signal is clear — single-provider AI dependencies are a reliability risk, and dual-provider failover is becoming standard practice for production workloads.