A concept picture of AI city File photo: VCG

The year of 2026 is a pivotal year for the leap of artificial intelligence (AI) from cognitive intelligence to embodied intelligence, with more high-efficiency large AI models expected to be rolled out in China, Wang Jin, executive vice dean of the School of Future Science and Engineering of Soochow University in Suzhou, East China's Jiangsu Province, told the Global Times in an interview on Sunday on the sidelines of the Third China Innovation Challenge on Artificial Intelligence Application Scene.

The event attracted the attendance of a total of 113 teams, comprising more than 350 participants from across the country to showcase their latest achievements in AI applications.

Wang forecast three major trends in China's AI industry in 2026. First, the focus of competition among large AI models is shifting from scale to efficiency and capability, focusing on reasoning improvement and efficient construction of intelligent agents, driving AI's evolution from generation to planning and execution. 

Within this year, more high-efficiency large AI models are expected to emerge, and more enterprise-level applications will be equipped with task-oriented AI agents, serving as "digital employees" to enhance the efficiency of basic tasks, he said.

In addition, spatial intelligence has become a research frontier, and large AI models are breaking through spatial understanding and empowering fields such as unmanned systems and digital twin. Moreover, the integration of AI and industry has entered deep waters, with AI being deeply embedded into the core processes of various industries and driving the reconstruction of industrial paradigms, according to Wang.

In terms of new achievements, embodied intelligence will see small-scale commercial use in scenarios such as industrial inspection and home services. The development of scientific large AI models will accelerate, reconstructing scientific research processes and shortening R&D cycles. AI innovations using green energy will be more valued, with new computing architecture and green data centers to be deployed to alleviate energy pressure, becoming new supports for the large-scale application of AI, Wang said.

China had more than 6,000 AI enterprises in 2025, while the scale of the country's core AI industry was expected to exceed 1.2 trillion yuan ($171.39 billion) in 2025. Meanwhile, AI applications have expanded to cover key industries including steel, non-ferrous metals, power and telecommunications. They are increasingly being applied in product research and development, quality inspection and customer services, according to the latest data from the Ministry of Industry and Information Technology.

However, the development of AI across the world still faces challenges and bottlenecks such as the contradiction between computing power demand and resource supply, the contradiction between the capability shortcomings of large AI models and the requirements for precise use, and the contradiction between rapid technological development and lagging ethical governance, Wang said.

In order to solve the problem, efforts are needed to improve energy utilization efficiency of computing power training by restructuring training frameworks and optimizing underlying chip design, incorporating low-cost green energy into the computing power planning system, and deploying edge computing power and advancing cross-domain scheduling of computing resources to alleviate uneven resource distribution, the expert said.

Moreover, more efforts are needed to improve the training structure and data logic of large models, strengthen the causal reasoning capabilities of models, improve the accuracy of generated content, and simultaneously overcome technical bottlenecks in model interpretability to enhance decision-making reliability. Third, efforts should be made to improve ethical norms and laws and regulations in the AI field, while strengthening international coordination and rule alignment in AI governance, Wang said.