In early 2025, China's DeepSeek-R1 emerged as a domestic challenger to the United States' ChatGPT, quickly capturing a peak market share of over 7.5 percent in January, according to international research firm SemiAnalysis.
However, data from SemiAnalysis also shows that DeepSeek's share had plummeted by nearly three percent within just a few months. Moreover, the share of tokens generated by DeepSeek models across the network from DeepSeek's own platform had dropped to only 16 percent by May.
But this decline in fact does not necessarily indicate the downfall of the company. Rather, when viewed alongside broader AI usage trends, it reflects the evolving dynamics of China's AI landscape and the increasingly sophisticated demands of its users.
The limitations of statistical methodologies are often underestimated. In explaining its report charts, SemiAnalysis explicitly states that it tracked the Chinese usage of DeepSeek-R1 poorly. The report primarily measures user engagement based on official website traffic, omitting third-party platforms such as Tencent Yuanbao, Baidu and Quark.
These platforms are where the DeepSeek model is extensively integrated and which, in reality, serve as the primary access points for users. This approach is also actively encouraged by DeepSeek itself.
As stated on its official website on May 28 following the release of the R1-0528 version: "After this R1 update, the context length of the model remains at 64K across the official website, mini programs, mobile Apps, and APIs. If users require longer context lengths, they may call the open-source version R1-0528 with a context length of 128K through other third-party platforms."
This circumstance is not a decline but rather a democratization of technology. DeepSeek focuses on foundational capabilities, driving industry progress through open-source strategies. By integrating DeepSeek's models, many applications allow users to access services without visiting DeepSeek's official site.
The results speak for themselves: Tencent Yuanbao saw its daily active users surge more than twentyfold between February and March, rising to become the third most popular AI-native mobile app in China by daily active user (DAU), while Quark recently topped usage charts. These developments confirm that the decline in direct website traffic paradoxically signals a deeper penetration of the technology.
In an interview with Chinese media outlet Waves, DeepSeek founder Liang Wenfeng emphasized that the company is currently in a phase of technological innovation rather than an application boom. Looking ahead, DeepSeek aims to establish an ecosystem that enables the industry to directly leverage its technologies and outcomes.
According to a recent KPMG report, the adoption of AI tools in Chinese workplaces has reached an impressive 93 percent, far exceeding the global average of 58 percent. Powered by DeepSeek's open-source models, AI tools are increasingly integrated across workplace ecosystems on multiple fronts: leading smartphone manufacturers have embedded DeepSeek models into their operating systems to enhance intelligent interactions, while major cloud platforms provide APIs with foundational support to industries such as finance, healthcare and e-commerce. This "device-cloud-ecosystem" collaborative model transforms AI from a technical hurdle into an accessible tool, marking China's AI industry shift from isolated breakthroughs to comprehensive system-level integration.
"Other companies develop B2B or B2C services based on our models, while we concentrate on fundamental research. Once the industry chain matures, there will be no need for us to develop applications ourselves. Of course, if necessary, we have the capability to do so — but research and innovation will always remain our foremost priority," Liang said.