Has Claude gotten dumber? A dark moment before the new model's release?

Outage and user impact

It has been reported that Anthropic's Claude — the conversational AI and API used by developers and enterprises — suffered a large-scale service outage on April 16, marking the seventh downtime incident in roughly two weeks. According to reports, the failure began at 10:53 a.m. Eastern and lasted nearly three hours, with Claude.ai, the API and Claude Code all showing a sharp spike in error rates. At peak more than 6,000 users submitted error reports, and some customers publicly complained or joked that they had to "use their brain" to write code while Claude was down.

Reliability metrics and customer churn

The outages have measurable consequences. It has been reported that Claude's API achieved only a 98.95% uptime over the past 90 days — well below the 99.99% figure often expected for enterprise-grade services. The compute shortfall has reportedly cost Anthropic clients: Retool's founder David Hsu says his company migrated away from Claude to OpenAI after repeated interruptions. Anthropic's CFO Sarah Friar reportedly acknowledged spending significant time hunting for any available temporary compute capacity.

Root causes and broader context

Industry observers point to a global crunch in AI-grade compute as the proximate cause: surging model demand outstrips supply of high-end accelerators and cloud capacity. It has been reported that Anthropic completed a large financing round earlier this year — figures in some reports put the round at $30 billion and the firm's valuation near $380 billion — yet capital alone cannot instantly buy scarce GPU slots. This pressure is playing out as competition with OpenAI intensifies, and as geopolitical moves — notably export controls on advanced semiconductors — tighten the available pipeline of cutting‑edge chips for AI training and inference.

Why it matters

For Western enterprises and Chinese firms evaluating foreign AI suppliers alike, reliability is as important as capability. Can Anthropic scale its compute and ops fast enough to match its ambitions — and to blunt defections to rivals — before a high-profile new model release? The outages raise that urgent question.