College students' favourite "external brain" malfunctions — DeepSeek outage leaves assignments undone

Outage and scale

DeepSeek (迪普西克), one of China's biggest generative-AI apps, suffered a near-12‑hour service outage starting March 29 at 21:35 and only returning to full service at 09:13 the next morning, it has been reported that the company’s status page shows. The interruption — the longest since DeepSeek released its R1 model in January 2025 — came as users across campuses and offices frantically searched for alternatives. DeepSeek ranks among the world's most-used web AI products (a16z placed it fourth globally in March 2026); QuestMobile data showed about 145 million monthly active users in Q3 2025, and daily active users reportedly jumped from roughly 120 million to 200 million in early 2026.

Students, workers and the "external brain" problem

For many college students the outage felt existential: AI has become an "external brain" for research, drafting and deadlines. Surveys cited during previous global outages (notably ChatGPT's long disruption in June 2025) found substantial numbers of students pausing study plans or asking for deadline extensions when models went dark. Gallup found in 2025 that 58% of employees were already using AI at work; for those who have baked these tools into daily routines, a sudden blackout translates directly into stalled coursework and lost productivity. When the crutch is gone, who still knows how to walk?

Why it happened, and what it means

DeepSeek's own post‑mortem points to capacity and architecture limits: daily token calls in China surged from about 100 billion in early 2024 to roughly 140 trillion by March 2026 — a growth of more than 1,000× — while DeepSeek's compute reserves rose only modestly. The app's Mixture‑of‑Experts (MoE) design saves training cost by activating subsets of "experts" on demand, but that same mechanism can cause resource contention under sudden load; user retry behaviour then creates "zombie requests" that amplify the spike. It has been reported that DDoS probes and other malicious traffic have hit DeepSeek in prior incidents, and there are rumours that the outage coincided with surging expectations for a delayed V4 model — reportedly fueling peak loads. Add global constraints on advanced chips and cloud capacity driven by trade policy and export controls, and the resilience question becomes geopolitical as well as technical. Are national AI champions building redundancy fast enough?

DeepSeek's blackout is a warning: generative AI is no longer an optional productivity accelerator but infrastructure. Students, developers and enterprises will increasingly demand fallbacks, multi‑vendor architectures and clearer contingency plans — or risk entire cohorts becoming dependent on services that can, and will, fail.