DeepSeek suffered a roughly five-hour outage that took key AI models and chat services offline, issuing partial refunds as questions linger about capacity constraints.

DeepSeek, a fast-rising Chinese artificial intelligence company, experienced a service disruption on Thursday that lasted roughly five hours, temporarily taking some of its core products offline.

According to the report, the outage began around 8 a.m. and continued until about 1 p.m., during which users were unable to access DeepSeek’s V4-Pro and V4.1-Flash models, as well as the company’s chat-based AI services. The interruption effectively prevented customers from using the affected tools for routine tasks such as querying the models or running AI-assisted workflows.

DeepSeek has not publicly specified what triggered the failure. However, the incident comes as Chinese AI developers navigate tightening constraints on advanced computing infrastructure. The report notes that the problems may be linked to insufficient computing capacity, a pressure point that has intensified amid U.S. restrictions on the export of high-end chips used for training and running large-scale AI systems.

In response to the disruption, DeepSeek issued partial refunds to paying subscribers whose services were impacted. While that step provides some immediate relief to customers, it does not resolve the broader uncertainty around the platform’s operational resilience—particularly as demand for AI services grows and compute resources remain a critical bottleneck.

The episode highlights the challenges facing AI firms that must deliver continuous, high-availability services while contending with limited access to cutting-edge hardware. For companies operating at the frontier of model performance, even short interruptions can ripple outward, affecting businesses and developers that rely on these systems for production workloads.

Why it matters

For customers using DeepSeek’s models in day-to-day products and services, multi-hour outages can halt work, disrupt applications, and erode confidence in reliability. The possibility that compute shortages are contributing reflects a wider issue for China’s AI sector under U.S. chip export controls. What to watch next is whether DeepSeek provides a clearer explanation and what steps it takes to strengthen capacity and redundancy to prevent similar disruptions.