China’s generative AI market has moved from rapid growth to enormous scale. The China Internet Network Information Centre reports that the country reached 515 million users in the first half of 2025, more than double the total recorded at the end of December 2024.
The reported adoption rate is 36.5 percent. Behind that headline sits a market shaped by advanced infrastructure, active government encouragement, restricted access to leading American services, and domestic products that have become useful enough to fill the gap.
This is not simply another region adopting the same global platforms. More than 90 percent of surveyed users said a Chinese model was their first choice. DeepSeek, Alibaba Cloud’s Qwen, and ByteDance’s Doubao have become popular inside an ecosystem developing largely apart from OpenAI and Google DeepMind, whose principal services are officially blocked on the mainland.
Adoption Is Concentrated but Accelerating
CNNIC surveyed 30,000 people across 31 provinces. Users below the age of 40 accounted for 74.6 percent of the total, while 37.5 percent held a higher-education degree. Generative AI is therefore most established among younger, educated people positioned to incorporate it into professional and personal routines.
Another study, published by Microsoft, used a different measure: the share of the working-age population actively using AI tools. It estimated more than 195 million Chinese users in June 2024 and found that adoption more than doubled to 20 percent during the six months after DeepSeek launched R1 in January 2025. US adoption remained around 25 percent over the preceding year.
The studies are not directly interchangeable, but both identify a sharp acceleration. A capable, widely discussed domestic model appears to have turned general awareness into experimentation at national scale.
Domestic Models Are Becoming a Separate Centre of Gravity
The preference for local tools reflects more than availability. Beijing’s “AI Plus” initiative calls for AI to spread across society and the economy, while policy emphasises self-reliance from foundation models to computing hardware. Technology companies have an incentive to build competitive products, and users have a growing set of services designed for their language and market.
Innovation activity extends beyond consumer adoption. China had filed 1.576 million AI-related patent applications by April 2025, according to CNNIC, representing 38.58 percent of the global total and the largest share held by any country.
Zhang Xiao, CNNIC’s deputy director, expects further development in open-source AI, embodied intelligence, agents, and governance. Those areas suggest a broadening market: models will not only answer questions but increasingly enter physical systems, perform tasks, and encounter rules governing their behaviour.
Scale Could Divide the Global AI Landscape
A domestic audience of 515 million gives Chinese developers an extraordinary environment for feedback, distribution, and commercial experimentation. It can sustain products independently of Western platforms and help local technical approaches become standards within a vast market.
That independence may also deepen a global split. One sphere could centre on American models, cloud platforms, and governance assumptions, while another grows around Chinese alternatives. Different hardware constraints, regulatory priorities, and user expectations could influence how systems are trained, deployed, and evaluated.
The division would not necessarily stop ideas or open-source code crossing borders, but it could make interoperability and common safety standards more difficult. It also means global AI leadership cannot be judged only by the adoption of a handful of US services.
China’s six-month surge remains an early snapshot. Growth may slow as the easiest users are reached, and popularity does not prove that every use produces lasting value. Yet the milestone establishes that a parallel ecosystem is no longer hypothetical. Hundreds of millions of people are already choosing domestic AI first, giving Chinese platforms the scale to evolve on their own terms.
