Claude.ai went dark on Monday morning. Anthropic’s chatbot, used by more than 500 enterprise accounts that each spend over $1 million a year, locked users out of its web app starting at 11:49 UTC on March 2, 2026.
Anthropic’s Claude AI suffered a worldwide outage on March 2, 2026, after the company’s status page identified failures on Claude.ai’s login and logout paths. About 2,000 users flagged the disruption on Downdetector. The underlying API was working, the company said. It blamed the strain on surging demand from a wave of new sign-ups.
What broke on March 2
The web app’s login layer failed first. Anthropic posted the first incident notice at 11:49 UTC on the official Claude service status dashboard. Users hitting claude.ai were met with a banner reading, “Claude will return soon. Claude is currently experiencing a temporary service disruption.” Downdetector logged a spike of close to 2,000 reports. Roughly 75% pointed to Claude Chat, 13% to the mobile app, and 12% to Claude Code, the company’s developer tool that has become its fastest-growing revenue line.
Anthropic told customers the Claude API was “working as intended” and that the failures were tied to claude.ai and the login and logout paths. A first fix was deployed at 15:25 UTC. The issue then bounced back through the rest of the day, with elevated errors flagged on Claude Opus 4.6 at 16:50 UTC. Anthropic posted further fixes on March 3 at 08:39 and 10:27 UTC before resolving the broader incident.
Why the API stayed up while the app went down
Anthropic’s split status, healthy API and broken consumer app, points to where the company’s growth is now hitting the walls. Login and session services usually share infrastructure with identity, edge caching, and rate-limiting layers. Those layers see the brunt of any sudden user surge before model-serving clusters do. When millions of new visitors slam the front door at once, authentication queues back up first.
The split also matters for who notices. Enterprise customers wired into the API behind their own products kept running. Free and Pro users on the web app did not. In a market where consumer mindshare drives enterprise pipeline, that asymmetry is awkward.
The Pentagon fight that drove a sign-up spike
The outage trailed a public confrontation between Anthropic and the U.S. Department of Defense. On February 27, 2026, President Donald Trump directed federal agencies to halt purchases from the company after CEO Dario Amodei refused to lift restrictions that bar Claude from being used for fully autonomous weapons or domestic mass surveillance. Defense Secretary Pete Hegseth designated Anthropic a supply chain risk the same day, a move detailed in Mayer Brown’s March 2026 advisory on the Pentagon supply chain risk designation.
The standoff produced a counter-effect Anthropic did not advertise but clearly absorbed. Claude’s consumer app climbed to the top of the U.S. App Store charts in the days that followed, and download growth in several major markets pushed traffic well past forecast. By the time March 2 arrived, the front-end stack was running closer to its ceiling than usual.
A pattern, not a one-off
March 2 turned out to be the first of at least five disruptions Anthropic logged through the month. A second wave hit on March 11, with Downdetector reports topping 1,400 by 10:29 a.m. ET. Partial outages followed on March 17, March 25, and March 27. Across the cluster of incidents, Anthropic’s engineers traced the cause back to networking performance degradation inside its serving stack and resolved it by shifting affected workloads onto healthy infrastructure.
That is a different category of problem from a security incident or a data loss. It is the signature of a company growing faster than its plumbing.
The reliability gap behind a $380 billion valuation
Three weeks before the March 2 outage, Anthropic closed a $30 billion Series G at a $380 billion post-money valuation led by GIC and Coatue. The company reported run-rate revenue of $14 billion, customers paying over $1 million a year jumping from 12 to more than 500 in twelve months, and Claude Code revenue at a $2.5 billion run rate. Eight Fortune 10 firms are now paying customers.
“Claude is increasingly becoming critical to how businesses work.” Krishna Rao, chief financial officer, Anthropic.
Critical and reliable are different things. Anthropic’s published 99.9% API uptime target sounds airtight on paper. Real measurement tells a different story. Claude.ai’s logged uptime over a recent 30-day window dropped to 99.32%, which works out to roughly five hours of downtime in a month. For an enterprise running 24-hour pipelines, that is not a rounding error. AI workflows that fail mid-session leave engineers without their pair programmer, customer support bots silent, and content jobs stalled past deadline.
What customers are doing next
Anthropic’s enterprise base is large, sticky, and increasingly hedged. Procurement teams that signed Claude as their primary model in 2025 began routing fallback traffic through OpenAI, Google’s Gemini, or open models hosted on Amazon Bedrock through 2026. The March outage cluster fed those plans. Vendors selling cross-model orchestration layers reported incoming calls within hours of each disruption.
The company has taken some steps to absorb the load. Service tiers described in the Claude API service tiers documentation now include a Priority Tier with prioritized compute targeting 99.5% uptime for paying enterprise accounts. Free and Pro consumer users sit a tier below.
That stratification is a tell. The most expensive customers will keep working through the next outage. Everyone else gets the banner. Anthropic’s growth has not paused for its plumbing to catch up. Internal benchmarks from the company’s own funding announcement note that Claude Code now accounts for roughly 4% of all public GitHub commits worldwide, double the share from one month earlier. The demand curve is not slowing.
Anthropic’s next scheduled milestone is its Claude Opus 4.7 launch in mid-April with record coding benchmark scores, which will pull more developer traffic into a system that has already shown its seams. A separate cohort of users had complained earlier this year that the flagship model felt downgraded after February tuning changes, a frustration covered in reports of Claude users revolting over the Mythos rollout. Reliability slips compound brand slips.
For now, the company has done what its CFO promised investors it would do. It has made Claude critical. The work left is making it dependable.
Leave a Comment