In 2026, China's AI industry officially entered the global first tier. The "2026 China AI Listed Enterprise Development Insights" report uses data to make one thing clear: domestic AI is not only highly competitive in the market but is also steadily becoming the preferred technological solution for developers at large.

Author | Fang Wensan

Image source | Internet

Usage surges, technology catches up

The most intuitive metric is Token consumption—that is, the scale of text processed by AI. In early 2024, China's average daily Token consumption stood at just 100 billion; by March 2026, it had exceeded 140 trillion, growing more than 1,000-fold in two years. Meanwhile, the performance gap between top Chinese and American models narrowed from 31.6% to 2.7%, open-source model downloads accounted for 41% of the global total, and AI patents accounted for as much as 60%. With a market strategy of "high cost-effectiveness plus limited-time free access," domestic models are becoming the core driving force behind global Token consumption growth.

Five major tracks, with the upstream most profitable

The report divides AI enterprises into five major categories, and revenue growth rates in the first half of 2026 show significant divergence:

l The intelligent computing chip and computing hardware sector (such as Cambricon and Hygon Information) grew fastest, with a median year-on-year increase of 16.1%; nearly 80% of enterprises achieved profitability, and the domestic substitution process continued to accelerate;

l The intelligent terminals and robotics sector (such as Xiaomi and Unitree Robotics) grew by 8.9%; embodied intelligence continued to gain momentum, but large-scale mass production still requires a certain cycle;

l The large model and AI platform sector (Tencent, Alibaba, Zhipu, etc.) grew by only 1.1%. Despite impressive technical performance, stock prices generally fell by 25%–35%, and capital expenditure weighed on short-term profits;

l The AI software and industry applications, and cloud/data foundation sectors achieved steady growth.

Landscape and trends

The large model field presents a "one superpower, multiple strong players" competitive landscape: DeepSeek's invocation volume is far ahead, with single-day Token consumption exceeding 40 trillion in September; native large model enterprises such as Zhipu and Minimax achieved doubled revenue, with R&D investment increasing significantly. In the future, as developers gradually adapt to the "Vibe Coding" (AI-assisted programming) working mode, highly cost-effective models will be more favored by the market. From a long-term perspective, AI self-training (RSI) and world model technologies will drive robots to achieve autonomous evolution in real-world scenarios.

Overall, China's AI upstream "water sellers" are highly profitable, midstream large model enterprises are deeply cultivating technology and honing internal strength, downstream robotics industries are poised for takeoff, and the entire industry is advancing from technological catch-up toward defining the future.

Part of the report content is as follows: