China’s DeepSeek AI chatbot suffered its longest service disruption on record this Monday, staying down for seven hours and 13 minutes before the company confirmed a full fix at 10:33 a.m. Beijing time (2:33 a.m. GMT).
The trouble started Sunday evening, when users began reporting problems on outage-tracking platform Downdetector. DeepSeek’s status page confirmed a fault at 9:35 p.m. and marked it resolved roughly two hours later, but the problems returned. It took until Monday morning to fully clear them.
The company has not explained what caused the disruption. That is consistent with its standard protocol, though it offers little comfort to users who lost hours of access. Outages like this can trace back to anything: a server failure, a bad update rollout, or traffic that overwhelmed infrastructure at the wrong time.
Why This Outage Is Different
What sets Monday’s incident apart is context. According to DeepSeek’s own status history, the chatbot’s main interface had never gone dark for more than two hours before this week. Seven hours is a different category of problem.
The company had maintained close to 99% uptime since its R1 reasoning model launched in early 2025 and went viral, turning DeepSeek from a relatively obscure Chinese startup into a global AI contender overnight. That reliability record has quietly been part of its appeal, especially for developers building on top of the platform.
DeepSeek’s API service, which developers use to integrate the chatbot into their own applications, did experience back-to-back day-long outages in late January 2025 at the height of R1’s viral peak. But those disruptions hit the developer layer, not the main interface that everyday users depend on. Monday was the first time the general public felt it at this scale.
Bad Timing on Multiple Fronts
The outage arrives at a difficult moment for DeepSeek beyond just the technical side.
Anthropic, the US company behind Claude, recently accused DeepSeek and two other Chinese AI firms of using approximately 24,000 fake accounts and over 16 million unauthorised exchanges to extract capabilities from Claude’s models. DeepSeek has not responded publicly to those allegations.
There is also the question of what comes next for the platform. The AI industry has been watching for DeepSeek’s follow-up to V3, and recent job postings from the company have pointed toward work in agentic AI. No release timeline has been shared.
For now, services are restored. Whether DeepSeek will offer any explanation for what actually happened remains to be seen.
and then