DeepSeek went down for almost five hours on Thursday, its second disruption this week. It has now done something few AI providers do: refunded a full week of usage. According to Caixin, the problems started at 8:11am Beijing time (01:11 UK time), partly affecting the DeepSeek-V4-Pro and DeepSeek-V4.1-Flash APIs and the chat service. Chat went fully offline at 9:08am. DeepSeek says it found the cause and had everything back by 1:04pm (06:04 UK time).
The South China Morning Post says users worldwide were affected, and many shared “server temporarily unavailable” screenshots. CGTN says “DeepSeek down” trended on Weibo, which is hardly surprising for an outage on a weekday morning in China.
The refund: a week of spend back as credit
In a “Service Restoration and Compensation Notice”, DeepSeek said it would refund whatever users spent between 8 and 14 October as credit. CGTN reports the details:
- the credit is deposited automatically;
- API billing comes out of the credit balance first;
- DeepSeek apologised and said it has started a “comprehensive investigation and corrective effort” on stability;
- it says it has reviewed and “eliminated single points of dependency”.
The notice doesn’t give a root cause. Note the window, too. The refund covers the week before the outage, not the outage itself. In effect, DeepSeek is paying back a full week of usage for a morning of downtime. That’s generous, and it suggests DeepSeek knows its reliability is now a competitive problem.
This isn’t a one-off
The incident history is what matters. CGTN says DeepSeek’s status page showed eight incidents in September and two more between August and October. Caixin notes a 15-minute interruption on 12 October and a near 12-hour outage in late March. For a company Caixin describes as raising money at a valuation of hundreds of billions of yuan and planning a 1 GW data centre in Ulanqab, that record will worry the developers building on its API.
CGTN points out that the API outage also took down third-party apps built on DeepSeek’s models. That’s the real cost, and it’s why this belongs on a search site and not just an AI one.
Why search and GEO people should care
DeepSeek is no longer just a chatbot. Other companies build products on its API, and as CGTN noted, those went down with it. We’ve covered DeepSeek’s push into agents and the browser and the funding round behind it. When the API goes down for five hours on a weekday morning, every tool that uses it for content briefs, answer monitoring or chat-based search stops too.
Three practical points:
- If you track “AI visibility” in DeepSeek, check your data for 15 October. Any tracker that polled DeepSeek during the outage will have gaps or errors. A drop in mentions that day means nothing.
- Build fallbacks. If your tools or client deliverables rely on one model API, this week is a reminder to have a second provider ready. In China, Qwen is the obvious one.
- Claim the credit. If you or your agency used the DeepSeek API between 8 and 14 October, check the balance. DeepSeek says the credit is automatic.
My take: the refund is a smart move, and it’ll get more goodwill than any status-page apology. But eight incidents in a month is a lot. DeepSeek has benefited from being the cheap, capable default for developers in China. Uptime is the next test, and right now it’s failing it more often than it should.