CCID Research Institute: China's AI inference data volume exceeds training data volume for the first time, with average daily token calls exceeding 175 trillion in June
English text is being prepared.
The "Top Ten Trends in Innovative Development of New Domains and New Quality" released by the CCID Research Institute shows that China's AI inference data volume has exceeded training data volume for the first time, marking a transition from the model training and R&D stage to the stage of large-scale industrial application. As of June 2026, average daily token calls exceeded 175 trillion, an increase of more than 1,000 times compared with the beginning of 2024. The report also mentions that more than 500 excellence-level smart factories have been built cumulatively, covering 31 provinces (autonomous regions and municipalities) and more than 80% of major manufacturing industry categories.
Why read this
AI score 65
The CCID Research Institute report presents the judgment that inference data volume has surpassed training data volume for the first time, along with figures such as average daily token calls, which can be used to observe the pace at which AI is shifting from training to application deployment.
Topics
Body
工信部赛迪研究院：我国人工智能推理数据量首超训练数据量，截至 6 月日均词元调用量超 175 万亿
2026-11-01 18:46 IT之家 - 归泷
感谢IT之家网友 不一样的体验 的线索投递！
IT之家 11 月 1 日消息，据央视新闻今日报道，中国电子信息产业发展研究院（简称“赛迪研究院”，英文缩写 CCID）昨日发布的《新域新质创新发展十大趋势》显示，我国人工智能推理数据量首次超过训练数据量，标志着我国人工智能从模型训练研发阶段迈入规模化产业应用阶段。
数据显示，截至 2026 年 6 月，日均词元调用量超过 175 万亿，相比 2024 年年初增长 1000 多倍。顶尖大模型的迭代周期也从过去的几个月压缩到几周甚至几天，人工智能正从“喂模型”的实验室阶段，全面转入“用模型”的产业落地阶段。
报道提到，随着“人工智能 + 制造”行动深入实施，我国已累计建成 500 多家卓越级智能工厂，覆盖 31 个省（区、市），涵盖超过 80% 的制造业行业大类。国产算力、大模型已在钢铁、汽车等 20 个行业落地 1000 多个应用场景，推动个性化定制、全生命周期服务等新模式不断拓展。
据IT之家此前报道，在今年 9 月的 2026（第七届）中国互联网基础资源大会上，中国互联网络信息中心政策与国际合作所发布《生成式人工智能应用发展报告（2026）》。报告显示，截至 2026 年上半年，我国生成式人工智能用户规模已超过 7 亿人，普及率突破 50%。