A person uses DeepSeek app on a mobile phone on Feb. 17, 2025. (Xinhua/Huang Zongzhi)

An incident involving DeepSeek has highlighted the growing tension between user demand and the computing infrastructure that powers generative artificial intelligence (AI). As consumer and enterprise adoption continues to climb, operators face mounting pressure to expand resilient, locally controlled computing capacity to avoid recurring outages and sustain the next wave of AI applications, industry observers said.

The remarks came after DeepSeek, a fast-growing Chinese AI model application, suffered a prominent service outage from Sunday night to Monday morning, disrupting access to its web and mobile platforms, Chinanews.com reported on Monday.

According to the report, many users posted on social media saying that "DeepSeek is down." Both the web and app are having problems: users see a "Server Busy" message, can't start new conversations, and some have even lost conversation content.

DeepSeek said that as of 10 am on Monday, services had been fully restored after two emergency fixes, Jiemian News reported. Company records showed the app has experienced at least seven significant interruptions over the past 15 months, according to the report.

The incident has reignited debate over domestic AI computing capacity. Industry analysts attributed the outage to a sudden surge in user demand that strained available computing resources. During the incident, many users repeatedly retried loading pages, creating a short burst of abnormal traffic the report described as an "avalanche effect" that exacerbated server load. 

The inference and training workloads for large-language AI models have driven explosive demand for graphics processing units (GPUs) and data-center resources, Liu Gang, chief economist at the Chinese Institute of New Generation Artificial Intelligence Development Strategies, told the Global Times on Monday.

Liu said that the rapid growth in AI users has periodically strained computing resources. In recent years, generative AI has spread rapidly and AI applications have proliferated, driving explosive user demand; however, computing infrastructure expansion has failed to keep pace, exacerbating the supply-demand imbalance.

The episode echoes other recent outages in the sector — including a global disruption at Anthropic's Claude earlier in March and a high‑traffic interruption of Alibaba's Qwen AI during a promotional event — demonstrating that capacity bottlenecks are a systemic issue as generative AI services proliferate, the Economic Information Daily reported.

Zhao Lidong, CEO of Chinese chipmaker Enflame Technology, said that the key to solving the computing power problem lies in fully unlocking the potential of domestic computing power. The domestic computing power ecosystem has a solid foundation, but requires genuine collaboration across the industry chain, particularly among computing power providers, model developers, and application providers, the Science and Technology Daily reported.

China has accelerated policies to expand computing infrastructure. Official data shows that China now hosts 42 large GPU clusters with a combined intelligent computing capacity exceeding 1,590 EFLOPS (a measure of supercomputing performance), placing it among world leaders in aggregate scale, the Xinhua News Agency reported.

In a 2025 report, Morgan Stanley estimated that China's GPU self-sufficiency rate reached 34 percent in 2024 and would rise to 82 percent by 2027.