The landscape of China's artificial intelligence (AI) market is rapidly shaking. Until the first quarter of this year, DeepSeek accounted for 99% of the total open-source AI model usage, but last month it plummeted to about 80%. It lost nearly 20 percentage points in just six months.
According to the Hong Kong South China Morning Post (SCMP) on the 28th (local time), the cloud platform PPIO noted that "since May, a wealth of excellent open-source models has emerged, expanding user choice and weakening DeepSeek's monopoly position." In particular, Alibaba's self-developed AI model Qwen surpassed DeepSeek, recording a maximum 56% share in PPIO as of the end of May.
Additionally, the Kimi-K2-Instruct model developed by the Alibaba-backed startup Moonshot AI has been added to PPIO since July. This model is rapidly spreading, backed by ease of use and learning efficiency, and is being quickly adopted in the global community.
DeepSeek was originally the first AI corporation to open its models to external companies on PPIO. In January, PPIO began offering DeepSeek's flagship models, V3 and R1, externally, which became a critical turning point in accelerating AI adoption in China.
However, the advance of latecomers has been strong. Alibaba significantly improved the functionality of its Qwen3 model group after April, and PPIO reported that Qwen3's market share rose from less than 1% to over 10% in June. Numerous startups, including Zhipu AI from Beijing, are also continuously releasing high-performance open-source models, diversifying the AI model selection market. Currently, the number of registered AI models in China has surpassed 1,500.
Competition is intensifying in the global market as well. The U.S. AI intermediary platform OpenRouter assessed DeepSeek and Qwen as the second (20%) and third (10.5%) models in the world, respectively. Both models recorded high usage following the models supported by Google and Amazon's Anthropic.
Although DeepSeek's market share is falling, brand preference remains high. In a global survey conducted by consulting firm Artificial Analysis, DeepSeek was chosen by more than half of the respondents as the "most preferred open-source model." This shows that DeepSeek's technological reliability remains strong despite the offensive from latecomers.
However, DeepSeek has remained tight-lipped about the development status of its next-generation models or future release schedule. Experts believe that as competition in open-source models intensifies, whether DeepSeek can maintain its technological leadership will determine the future direction of the Chinese AI market.