Chinese artificial intelligence startup DeepSeek experienced its most severe service disruption on record on Monday, March 30, 2026, with its chatbot offline for 7 hours and 13 minutes in a two-phase failure that began Sunday evening and was not fully resolved until mid-morning the following day.
DeepSeek’s own status page classified the incident as a “major outage,” with the disruption running from the early hours of Monday morning until 10:33 a.m. local time, when the incident was marked as resolved. Users first began reporting faults on Sunday evening, according to outage-tracking platform Downdetector, with the startup’s status page acknowledging an initial issue at 9:35 p.m. before marking it resolved about two hours later. A second wave of performance problems then emerged Monday morning and persisted until the 10:33 a.m. fix.
The extended downtime is unusual for DeepSeek, which has maintained a near-99 percent operational record since unveiling its R1 model in January 2025, with no previous major outages exceeding two hours. Its application programming interface (API) service, used primarily by developers to integrate the chatbot into third-party applications, had seen consecutive day-long outages in late January 2025 during the height of its viral moment, but its consumer-facing webpage had remained relatively stable until Monday.
The cause of the outage remains unclear. DeepSeek did not immediately respond to requests for comment and offered no specific explanation, in line with its standard company protocol. Such disruptions can stem from server overload, software bugs introduced during updates, or broader infrastructure failures.
The Hangzhou-based company’s chatbot is used by hundreds of millions of people globally, with Chinese social media platforms flooded with complaints during the breakdown. Developers and enterprise teams that rely on DeepSeek for coding assistance, customer support automation, and other business operations faced the most acute exposure during the disruption.
The global artificial intelligence industry is closely watching for the release of DeepSeek’s next-generation model, though the company has given no indication of a timeline. Anticipation remains high for DeepSeek’s multimodal V4 model, reported to be capable of generating images, video, and text. Monday’s incident, and the absence of a public explanation, is likely to intensify scrutiny of the company’s infrastructure as expectations around its next product milestone continue to build.
