Overnight into March 30, DeepSeek experienced the largest outage since its launch. According to the company's status page, the first problems occurred at 21:35 Beijing time. They were marked as resolved roughly two hours later — but at 00:20 the situation recurred. The service was not fully restored until 10:33 the next morning.
So not seven hours, but at least thirteen — with four separate cycles of “investigating → fixing → investigating again.” That is the timeline on status.deepseek.com, which Bloomberg called the primary source.
Silence instead of explanations
The company did not provide any comment to Bloomberg or to other outlets. The causes of the outage have not been officially disclosed. This is atypical even for closed tech companies: most large platforms publish at least a short post-mortem after prolonged outages.
DeepSeek had maintained around 99% uptime since the launch of the R1 model in January 2025 — so the outage by itself is a statistical outlier. But the question isn't the uptime figure.
Context that matters more than the outage
In January 2025, the appearance of DeepSeek R1 triggered a sell-off in tech stocks totaling more than a trillion dollars — Nvidia lost about $600 billion in a single day. The startup claimed it trained the model for $5.6 million, while American competitors spend billions. Analysts at the Cloud Security Alliance later called that claim “significantly overstated.”
It also emerged that DeepSeek used Nvidia chips that were prohibited for export to China. In April 2025, a bipartisan committee of the U.S. Congress published a report calling DeepSeek “a deep threat to national security,” in part because user data is routed through infrastructure linked to China Mobile — a telecom operator that the U.S. banned from its market back in 2019.
“DeepSeek operates as a direct channel for foreign intelligence collection of private data on Americans.”
Report of the U.S. House Committee on China, April 2025
Researchers at Cisco found that the model did not block a single malicious request during security testing. Qualys recorded failures in more than half of the jailbreak tests. HiddenLayer explicitly stated that deploying DeepSeek on a company's own infrastructure carries risks due to “reliability and data security issues.”
What the outage says about the choice between price and reliability
DeepSeek appeals primarily because of cost: the model is free, the API is cheaper than OpenAI's, and local deployment is technically feasible. That's why companies around the world are beginning to integrate it into products — often without deep audits of the supplier's infrastructure.
The March 30 outage didn't break anything critical — it simply served as a reminder that after 13 hours of silence no company received an explanation from the provider it had entrusted with its requests. For a consumer chatbot that's an inconvenience. For an enterprise tool, it's an operational risk.
If DeepSeek truly expects to rely on the enterprise market outside China, the next major outage will be a test not of technology but of communications: whether a post-mortem with real causes will finally appear — or whether the company will once again choose silence.
