From the end of 2025 to the beginning of 2026, China’s AI market is undergoing a crucial transition from technological enthusiasm to value rationality. The policy-driven “Artificial Intelligence +” initiative continues to deepen, technological breakthroughs in multimodality and intelligent agents are accelerating, and the listing of the first AGI enterprise on the capital market has opened a new chapter. Multiple forces are jointly driving the industry into a new stage of “high-quality development”. Based on data from authoritative institutions such as the Ministry of Industry and Information Technology, CCID Research Institute, and IDC, combined with recent core industrial events, this article analyzes the structural changes in China’s AI market from a third-party perspective.

I. Capital Milestone: Intensive Listing of AGI Enterprises, Commercialization Enters Value Assessment Period

In January 2026, China’s AI capital market witnessed a landmark event: on January 8th, Zhipu AI was listed on the Main Board of the Hong Kong Stock Exchange, becoming the world’s first listed company with general artificial intelligence (AGI) foundation model as its core business; the next day, MiniMax followed suit to complete its listing, forming a pattern of “dual giants competing for speed”. This phenomenon marks that China’s AI enterprises have stepped from the “financing and burning money” stage to the “capitalization verification” stage, and the capital’s evaluation logic for AI has shifted from technological potential to commercial closed-loop capability.

In terms of core data, Zhipu AI’s commercialization process is quite representative: as of September 2025, it has served 12,000 enterprise customers worldwide, over 80 million terminal devices, and 45 million developers. Nine out of the top ten Internet companies in China have adopted its GLM large model, and its revenue has doubled for three consecutive years from 2022 to 2024. However, it should be noted that both listed enterprises still have large-scale losses, reflecting the high-investment nature of large model research and development—even leading enterprises need to find a balance between “technological iteration” and “profit balance”.

Another notable trend on the capital side is that investment is tilting towards value realization. According to incomplete statistics, the total financing of China’s AI-related enterprises reached 107.07 billion yuan in 2025, with an average of 2.6 enterprises receiving financing every day. However, the embodied intelligence sector accounted for more than 50% of the total, with cumulative financing of 25.7 billion yuan, becoming the most popular track. JPMorgan Chase predicts that private capital investment in China’s AI track will increase from 1.9 billion US dollars in 2025 to 2.5 billion US dollars in 2026, focusing on areas with clear return expectations such as industrial applications and intelligent terminals.

II. Technological Leap: From Multimodality to Intelligent Agents, Core Capabilities March Towards the Physical World

Since 2025, China’s AI technological breakthroughs have shifted from a single parameter competition to a dual main line of “multimodal fusion” and “physical world interaction”, with significantly improved technological maturity. Data from CCID Research Institute shows that China’s large models have increased their language understanding and multimodal capabilities by 30% and 50% year-on-year respectively, and core capabilities such as reasoning and programming have achieved leapfrog development.

In the field of general technology, native multimodality has become the mainstream direction: enterprises such as Alibaba and Baidu continue to increase investment in multimodal large models, realizing the integration of understanding and generation of text, images, videos and other data; the SALMONN audio-visual large model developed by Tsinghua University has surpassed GPT-4o and Google Gemini in multiple authoritative evaluations, performing prominently in tasks such as video description and question answering. Tencent’s open-source world model Hunyuan Voyager also ranked first in comprehensive capabilities in Stanford University’s WorldScore benchmark test. Its breakthrough in 3D space and time perception reasoning capabilities has laid a technical foundation for scenarios such as autonomous driving and robots.

Intelligent agents (AI Agents) have become the core carrier of technological landing, accelerating the transition from concept to commercial use. In the C-end market, ByteDance launched the AI headset Ola Friend equipped with Doubao large model, realizing functions such as personalized companionship and scenario-based question answering; in the B-end field, Baidu’s “Zhijin” financial intelligent agent is deeply integrated into scenarios such as wealth management and compliance review, and Zhipu AutoGLM intelligent agent supports cross-terminal operation of mainstream APPs. IDC predicts that by 2026, 50% of the data teams of China’s Top 500 enterprises will use intelligent agents to complete data preparation and analysis, and the collaborative efficiency of intelligent agents will become the core competitiveness of enterprises.

The supporting capacity of elements has been upgraded simultaneously: in terms of computing power, joint calculations by IDC and Inspur Information show that China’s intelligent computing power scale reached 1037.3 EFLOPS in 2025, and 10,000-card-level clusters have become the mainstream carrier for large model training; in terms of data, China’s total data production exceeded 50 ZB in 2025, the annotation scale of 7 major annotation bases including Hefei and Chengdu exceeded 29 PB, and 524 high-quality industry datasets were built, providing core support for model iteration.

III. Industrial Penetration: From Pilot to Large-Scale, “AI + Manufacturing” Becomes the Core Engine

Driven by the “Opinions on Further Implementing the ‘Artificial Intelligence +’ Initiative”, China’s AI applications have shifted from local pilots to comprehensive penetration in 2025, among which the manufacturing industry has become the most deeply integrated and effective field. Data from CCID Research Institute shows that as of the end of 2025, the number of intelligent factories in China exceeded 30,000, AI technology drove production efficiency up by 22.3%, and the R&D cycle was shortened by nearly 30%.

The landing of vertical industry large models has become a key starting point for industrial upgrading. TCL’s Star Intelligence Model 3.0 focuses on the semiconductor display field, increasing product problem analysis efficiency by 20% and material development efficiency by 30%, which confirms the commercial value of vertical large models. From the perspective of value chain distribution, the penetration of AI in the industrial field presents a trend of “elevation of core links”: the proportion of application cases in the production and manufacturing link has increased from 19.9% to 25.9%, breaking the previous smile curve pattern of “high investment in R&D design and marketing services at both ends”.

The continuous expansion of industrial scale confirms market vitality: data from the Ministry of Industry and Information Technology shows that the scale of China’s core AI industry exceeded 1 trillion yuan in the first 11 months of 2025, and the number of artificial intelligence enterprises exceeded 5,300 as of September, among which more than 400 were national-level specialized, refined, characteristic and innovative “Little Giant” enterprises, and the overall strength ranked among the world’s first echelon. JPMorgan Chase predicts that the scale of China’s core AI industry will reach 1.2 trillion yuan in 2026, a year-on-year increase of 30%, among which the export growth rate of AI servers is expected to exceed 40%, becoming a new growth point for high-tech exports.

IV. Challenges and Prospects: Building Ecological Barriers in Competition and Cooperation

Despite the rapid development of China’s AI market, core challenges still exist. At the technical level, Tang Jie, founder of Zhipu AI, admitted at the AGI-Next Frontier Summit that the gap between China’s large models and the world’s top level has not narrowed, and core problems such as hierarchical memory structure and autonomous learning still need to be broken through; at the industrial level, problems such as difficulty in obtaining industrial data and insufficient encapsulation of process knowledge restrict the penetration speed of AI into core production links; at the commercial level, most enterprises still face the dilemma that “revenue growth cannot cover R&D costs”, and the profit model needs to be further optimized.

Looking forward to 2026, the industry will present three major trends: first, technological competition will shift from parameter competition to “intelligent efficiency” improvement, and technologies such as sparse attention and mixture of experts models will achieve a balance between computing power costs and performance; second, the ecological pattern will shift from “thousands of models competing” to “ecological survival”, with leading enterprises building barriers through the open-source foundation + industry fine-tuning model, and small and medium-sized enterprises focusing on breakthroughs in vertical scenarios; third, the value of data elements will be released at an accelerated pace, the scale of trusted data spaces will expand, and cross-domain data sharing mechanisms will be gradually improved, providing high-quality support for model iteration.

Overall, China’s AI market in early 2026 is standing at the intersection of “technological maturity” and “commercial feasibility”. The triple role of policy guidance, technological breakthroughs and capital rationality will drive the industry to shift from “scale expansion” to “quality improvement”, and gradually build an industrial ecosystem with technological originality, commercial closed-loop capabilities and global competitiveness.