Shanghai-based Chinese financial news site East Money recently reported that, around 9:35 PM on March 29th, well-known Chinese AI provider DeepSeek flagged a “Major Outage”. This sudden outage persisted until the following morning without being resolved.
During this extended period of no service, social media platforms across China were quickly filled with related discussions. DeepSeek officially announced several fixes, but in reality, the entire outage lasted at least half a day before the issue was truly resolved. During this extended 12-hour downtime, netizens’ reactions shifted from initial “understanding and support” to “give me back my productivity,” and finally to “have you guys run away?”
As the capabilities of large-scale models continue to improve, system load is also increasing accordingly. Taking 1M-Tokens long contexts as an example, while supporting more complex tasks, it also significantly increases computing power consumption. Infrastructure such as optical modules, liquid cooling systems, and GPU clusters still constitute the underlying support for current AI services. With the combined pressure of high-frequency calls and long text interactions, the continuous rise in computing power costs has become an industry consensus.
Publicly available data shows that in the first quarter of 2025 alone, DeepSeek’s daily active users surged from 120 million to 200 million, an increase of nearly 70 percent. However, sources familiar with the matter revealed that during the same period, DeepSeek’s computing power increased only by 8.3 percent.
Source: East Money, March 30, 2026
https://finance.eastmoney.com/a/202603303688405329.html