DeepSeek API Outage Lasts Nearly Five Hours, Firm Apologizes and Refunds Users

A five-hour outage at a major model provider is a reminder that AI infrastructure carries the same operational fragility as any cloud dependency, and that reliability, not capability, is often the binding constraint for teams building on top of it. The refund mechanism — bonus credits drawn down first — is notable because it compensates in future consumption rather than cash, keeping affected users inside the ecosystem. Whether the single-point-dependency fixes hold under load is the open question.