DeepSeek outage jars China AI race

DeepSeek is investigating a prolonged disruption that knocked its chatbot offline for more than seven hours in China, in the most serious outage yet for the Hangzhou-based artificial intelligence company since its rise into the global mainstream. The company’s public status page showed web and app service problems began late on Sunday night, rolled into Monday morning, and were not fully marked resolved until 10:33 a. m. China time after a string of updates and monitoring notices.

DeepSeek hit by a seven-hour breakdown

Users first began encountering faults on Sunday evening, with the company opening one incident at 9:35 p. m. on March 29 and closing it at 11:23 p. m., before a second and more significant disruption was logged after midnight. DeepSeek’s status history shows repeated shifts from “investigating” to “monitoring” and back again before the final resolution notice, suggesting the fixes were incremental rather than immediate. Reuters, citing the same status data, reported the main outage lasted 7 hours and 13 minutes, making it the chatbot’s longest interruption since DeepSeek’s flagship R1 and V3 models propelled the company into prominence in early 2025.

DeepSeek did not publicly assign a cause to the breakdown, leaving room for several possibilities familiar to large-scale AI operators, including server instability, capacity stress or software issues linked to updates. That caution matters because the company’s own incident log shows that a fix was implemented more than once before service normalised, a pattern that can signal either a layered technical fault or an initial remedy that did not fully address the problem. For a consumer-facing chatbot used at scale, the length of the disruption is notable not only because of lost access, but because it interrupts the expectation that conversational AI tools should be available almost continuously, especially as they become embedded in work, study and software development routines.

The outage also lands at a sensitive moment for DeepSeek. The company remains one of the most closely watched AI groups in China, and Reuters said the wider industry is awaiting its next-generation model even though DeepSeek has not given a release timetable. That anticipation has increased scrutiny of operational resilience as much as model performance. A fast-rising AI company can tolerate occasional glitches in its early phase, but a service that has become a daily tool for ordinary users and developers is judged more like infrastructure. Reliability, latency and recovery speed start to matter alongside benchmark scores and token pricing.

There is also a wider pattern behind Monday’s incident. Reuters reported that DeepSeek’s API service suffered consecutive day-long outages in late January 2025 when demand around its breakout moment was surging, although its direct web chatbot had not previously recorded a major outage longer than two hours. A separate Reuters report from March 10, 2025 also documented a brief disruption affecting DeepSeek’s advanced R1 reasoning model across both API and web chat services before the company restored it. Taken together, those incidents suggest that while DeepSeek has usually recovered quickly, service continuity has become a growing test as the platform expands beyond a novelty product into a core player in the AI stack.