The China Center for Information Industry Development (CCID) said AI inference data had surpassed training data in China for the first time in a report released on 31 October, Pandaily reported on 2 November. CCID puts daily token calls above 175 trillion as of June 2026, more than 1,000 times their level at the start of 2024. The report was released at the finale of the 2026 New Domain and New Quality Innovation Competition in Qingdao, Shandong province. CCID is a research institute under China’s Ministry of Industry and Information Technology.
Key points
- CCID says China’s AI inference data volume has exceeded training data volume for the first time.
- Its figure for daily token calls exceeds 175 trillion as of June 2026.
- A separate October report puts daily token consumption at about 500 trillion in June.
- CCID links the shift to wider industrial use of AI, including smart factories.
CCID’s 175 trillion daily token calls
CCID describes the crossover as a move away from a phase centred on model training and development towards AI use at industrial scale. Its report sets out ten trends in what it calls new-domain, new-quality innovation, with the change in inference data volume among its findings.
The institute also says leading large models now go through an iteration cycle of weeks or even days, down from several months, Bastillepost reported. That is a separate measure from daily token calls: one concerns how often models change, while the other counts their use.
Liu Wenqiang, CCID’s vice president, told the Qingdao ceremony that AI was becoming an instrument for changing industrial systems, rather than remaining a means of improving individual tools. His remarks place the reported crossover alongside changes in how factories and other businesses apply models.
CAICT puts June consumption at 500 trillion
The China Academy of Information and Communications Technology gave a different figure in a separate report released in October. It put daily token consumption at about 500 trillion in June, Pandaily reported. CCID’s figure is more than 175 trillion daily token calls for the same month.
Pandaily reported that estimates of token volume vary by source and method. It also said the reports on CCID’s findings did not explain how the institute measured the volumes of inference and training data. Those two volumes are the basis of CCID’s claim that inference has overtaken training.
China’s 500-plus excellence-level smart factories
CCID says China has built more than 500 excellence-level smart factories through its AI Plus Manufacturing initiative, spanning 31 provincial-level regions and more than 80% of major manufacturing categories. The factories are one part of the industrial activity the institute associates with the move towards inference.
Domestic computing resources and large models have also been applied in more than 1,000 scenarios across more than 20 industries, including steel and automotive, Bastillepost reported. The applications include personalised customisation and services spanning a product’s lifecycle.
Use extends beyond manufacturing. A September report from the China Internet Network Information Center put China’s generative AI users at more than 700 million by the first half of 2026, with penetration above 50%.
