At the 2026 Zhongguancun Forum, the vanguard of China’s artificial intelligence sector gathered to signal a profound shift in the industry’s trajectory. Moving beyond the era of mere chatbots, leaders from Moonshot AI, Zhipu AI, and Xiaomi articulated a vision where 'Agents'—autonomous digital workers capable of multi-step reasoning—become the primary drivers of economic value. This transition marks the end of the speculative 'training' phase and the beginning of a massive 'inference' boom that is reshaping the physical and economic landscape of Chinese tech.
The centerpiece of the discussion was the emergence of OpenClaw, an agentic framework that has effectively turned large language models from passive answer-engines into active problem-solvers. Zhang Peng, CEO of Zhipu AI, noted that as these models transition from 'chatting' to 'working,' the consumption of tokens is skyrocketing. A single complex task performed by an Agent can consume 10 to 100 times the data of a standard query, necessitating a re-evaluation of pricing strategies and compute efficiency to ensure the industry remains sustainable.
This explosion in demand is giving birth to a new industrial concept: China as the 'World’s Token Factory.' Xia Lixue of Everly AI argued that just as China once leveraged its manufacturing prowess to export physical goods, it is now positioning itself to export 'intelligence' in the form of low-cost, high-efficiency tokens. By integrating domestic chipsets with software-defined infrastructure, China aims to create a self-sustaining ecosystem where energy and silicon are converted into the cognitive labor that will power the global AI economy.
Looking toward the next twelve months, the panel identified 'self-evolution' as the defining theme. Luo Fuli, head of Xiaomi’s MiMo model team, suggested that we are entering an age where Agents will not only assist humans but will autonomously conduct scientific research and optimize their own architectures. As software moves from human-centric graphical interfaces (GUI) to agent-native command lines, the barrier between digital planning and physical execution is expected to dissolve, fundamentally altering the role of the white-collar worker.
