China's AI models lead global race with record token usage

According to a report by CCTV, China's daily average token usage in artificial intelligence applications has surpassed 500 trillion as of June 2026. This surge in token demand reflects the intensifying global competition among large language models, with a significant increase in computational power requirements. CCTV Finance noted that the global race for large models is at its peak, with rapid iterations and fierce competition. Chinese models have established themselves as a top-tier force in this arena. Repeatedly, industry insiders emphasized that the competition has shifted from merely measuring model intelligence to a broader contest involving intelligent agents and ecosystem development. Feng Wen, Chief Architect at Xiyu Technology's Open Platform, stated that by the second half of 2025, model updates were expected every three months, but now new models are released every four to six weeks. Companies are continuously increasing their investment in training, leading to larger model parameters and expanded context windows. IT Home observed that AI is evolving from simple conversation to practical task execution, with intelligent agents beginning to be deployed on a large scale. Each task involves multiple steps, including repeated retrieval, context reading, tool calls, and multi-round feedback, which has led to a dramatic rise in inference computing power needs. Liu Feng, General Manager of Tencent's Smart Industry Division, mentioned that the official release of Hunyuan 3 saw a 68-fold increase in token usage compared to its predecessor, Hunyuan 2. Zheng Zihao, General Manager of the AI Computing Power Center at Envision Group, highlighted the rapid development of AI computing centers, driven by long-term and stable orders from users. Domestic demand for computing power is surging, with many regions accelerating the construction of domestic token factories and intelligent computing complexes, pushing the domestic computing power supply chain from product validation to mass delivery.