Anthropic’s Claude platform logged seven documented outages in March 2026 — the worst reliability month since the company launched its commercial API in 2023. The most severe incident, on March 27, took down the full Claude platform for approximately 2 hours and 55 minutes. Root causes across all incidents traced to capacity scaling failures and structural dependencies on third-party GPU providers CoreWeave and Fluidstack.
This reference compiles every March 2026 Claude outage with verified start times, durations, and affected services — drawn from status.anthropic.com incident reports and corroborated by Downdetector spike data.
March 2026 Anthropic Claude Outage Summary
| Date | Start (UTC) | Duration | Affected Services | Severity |
|---|---|---|---|---|
| Mar 6 | 14:22 | ~35 min | API | Minor |
| Mar 11 | 09:15 | ~2h 10m | Claude.ai | Moderate |
| Mar 17 | 16:40 | ~50 min | API | Minor |
| Mar 23 | 11:30 | ~4h | API, Claude.ai | Major |
| Mar 24 | 13:00 | ~6h | API, Claude.ai, Claude Code | Major |
| Mar 25 | 10:45 | ~3h 20m | API, Claude.ai | Major |
| Mar 27 | 08:12 | ~2h 55m | API, Claude.ai, Claude Code | Critical |
| Mar 29 | 19:30 | ~1h 15m | Claude Code | Moderate |
Incident-by-Incident Breakdown
March 6 — API Slowdown (Minor)
At 14:22 UTC, Anthropic’s API began returning elevated latency on inference requests, with p99 response times climbing above 30 seconds. The issue cleared within 35 minutes. Anthropic attributed the slowdown to a routine capacity rebalancing operation that briefly cut available GPU allocation.
March 11 — Claude.ai Two-Hour Outage (Moderate)
Claude.ai went fully offline at 09:15 UTC, returning 502 errors for 2 hours and 10 minutes. The API remained functional throughout. Anthropic’s post-incident report cited a misconfigured load balancer update that routed traffic to unhealthy backend nodes — a configuration error, not a capacity event.
March 17 — API Latency Spike (Minor)
A 50-minute latency event hit the API at 16:40 UTC. Claude.ai and Claude Code were unaffected. Anthropic flagged an upstream networking issue with Fluidstack in its European inference region — the first public acknowledgment of that provider dependency in a status update.
March 23–25 — Three-Day Capacity Crisis (Major)
The most protracted disruption of March 2026 ran across three consecutive days, accumulating more than 13 hours of degraded or unavailable service. This was a single cascading capacity failure, not three isolated incidents.
March 23: Beginning at 11:30 UTC, API throughput dropped by an estimated 60–70%. Claude.ai remained partially accessible but with high error rates. Duration: approximately 4 hours before partial stabilization.
March 24: The worst day of the three. Starting at 13:00 UTC, the incident expanded to cover all three services — API, Claude.ai, and Claude Code — for roughly 6 hours. Anthropic’s status page acknowledged “significant capacity constraints across multiple inference regions.”
March 25: A partial recovery on March 24 evening failed to hold overnight. By 10:45 UTC on March 25, API and Claude.ai were again degraded for approximately 3 hours and 20 minutes before full restoration at 14:07 UTC.
Root cause: a planned capacity expansion with CoreWeave fell behind schedule, leaving Anthropic under-provisioned during a demand surge driven by Claude 4 adoption. Anthropic acknowledged the CoreWeave dependency explicitly in its March 25 post-mortem — an unusually direct disclosure for the company.
March 27 — Full Platform Outage, Worst of the Month (Critical)
At 08:12 UTC on March 27, Anthropic suffered a complete platform failure. The API, Claude.ai, and Claude Code all went down simultaneously, remaining offline for 2 hours and 55 minutes before restoration at 11:07 UTC. This was the most severe single incident in Anthropic’s commercial history to that date.
The timing was damaging: it struck during peak North American morning hours, when enterprise and developer traffic is heaviest. Downdetector recorded more than 14,000 user reports within the first hour — the highest single-incident volume ever logged for Anthropic. Post-mortem analysis pointed to a cascading failure triggered by an automated traffic routing change that interacted with the degraded capacity environment left over from the March 23–25 crisis.
The architecture that failed had been partially exposed months earlier. A previous Anthropic source code disclosure revealed the complexity of its multi-provider GPU inference stack — the same stack that buckled under compounded load on March 27.
March 29 — Claude Code Timeouts (Moderate)
The month’s final incident hit Claude Code at 19:30 UTC on March 29. Widespread timeout errors persisted for approximately 1 hour and 15 minutes. The API and Claude.ai were unaffected. Anthropic listed the cause as “elevated queue depth in the coding infrastructure backend” — a localized failure consistent with the March 27 remediation still stabilizing.
Root Causes: CoreWeave, Fluidstack, and Structural Capacity Gaps
Anthropic’s post-mortems identified two structural drivers behind the March 2026 outage cluster:
- CoreWeave onboarding delays: A scheduled capacity expansion fell behind schedule, leaving Anthropic’s inference fleet under-provisioned exactly when Claude 4 demand was accelerating. The March 23–25 crisis and March 27 full outage both traced to this supply shortfall.
- Fluidstack networking instability: The March 17 latency event tied directly to Fluidstack instability in European inference regions. Multi-provider GPU sourcing reduces single-vendor risk but introduces correlated failure exposure when demand spikes across providers simultaneously.
The pattern reflects a structural problem across the AI infrastructure market. Providers are competing for the same limited pool of hyperscaler and specialty cloud GPU capacity. Nebius’s $10 billion owned data center build in Finland illustrates how aggressively well-capitalized players are moving to escape this third-party dependency risk — a bet Anthropic has not yet fully replicated at scale.
How to Monitor Claude Status in Real Time
Three sources provide reliable visibility into Claude outages:
- status.anthropic.com — Anthropic’s official status page. Updated during active incidents; post-mortems published within 48–72 hours of resolution. Subscribe via RSS for automated alerting.
- Downdetector — Crowd-sourced tracking at downdetector.com/status/anthropic. Typically surfaces incidents 5–10 minutes before Anthropic’s status page acknowledges them — useful as an early-warning signal.
- @AnthropicAI on X — Posts status updates during major outages. Less granular than the official status page but useful for broad confirmation during platform-wide events.
Enterprise API customers on Anthropic’s Business tier receive email notifications scoped to their service tier. Anthropic publishes no SLA-backed uptime commitment for its standard API or Claude.ai consumer plans.
FAQ: Anthropic Claude Outage March 2026
Was Claude down on March 27, 2026?
Yes. The full platform — API, Claude.ai, and Claude Code — was down from 08:12 UTC to 11:07 UTC on March 27, 2026. Total duration: 2 hours 55 minutes. This was the worst single Anthropic outage on record.
How long was the March 23 outage?
Approximately 4 hours. March 23 was the opening event in a three-day capacity crisis running March 23–25, which totaled more than 13 hours of degraded or unavailable service across API and Claude.ai.
Did Claude Code go down in March 2026?
Yes, on two separate occasions. Claude Code was fully affected during the March 24 incident (approximately 6 hours) and the March 27 full outage (2 hours 55 minutes), plus isolated timeout errors on March 29 lasting 75 minutes.
Is Claude stable now?
Anthropic reported no further multi-day outage events through April 2026. The March 27 post-mortem stated that CoreWeave capacity onboarding had been accelerated and automated traffic shedding implemented to prevent cascading failures. MegaOne AI tracks Claude availability as part of its monitoring of 139+ AI tools across 17 categories.
What is Anthropic’s uptime SLA?
Anthropic does not publish a formal uptime SLA for consumer Claude.ai plans or the standard API tier. Custom SLA terms are available under enterprise agreements negotiated directly with Anthropic’s sales team — but no public commitment exists.
Where do I check Claude’s current status?
Check status.anthropic.com for the authoritative real-time view. Downdetector provides a crowd-sourced secondary signal that often leads official acknowledgment by 5–10 minutes during fast-moving incidents.
Enterprises running Claude in production should treat March 2026 as the reference failure scenario: build with fallback inference providers, implement request queuing, and monitor status.anthropic.com via RSS or webhook. Waiting for an email notification during a 3-hour full-platform outage is not a recovery strategy.
