In China, daily use of artificial intelligence surpasses model training

A report by the China Center for Information Industry Development indicates that, as of June 2026, the volume of inference data for Chinese artificial intelligence exceeded, for the first time, the volume devoted to model training. The figure, reported in two separate articles by Sina News citing the same research institute and, in one of the two, also the state broadcaster CCTV, marks, according to the report, China’s shift from the model research and development phase to that of industrial-scale application.

Average daily token consumption — the unit used to measure text processing by language models — exceeded 175 trillion, an increase of more than a thousandfold compared with early 2024. Over roughly two and a half years, this increase measures the acceleration of daily use of artificial intelligence systems in the country, beyond the purely experimental development phase.

In the week examined by the report, the top five models by usage volume worldwide were all of Chinese origin, with combined usage of 93.54 trillion tokens, compared with 22.47 trillion recorded by US models over the same period. The comparison between the two totals offers a snapshot of the spread of usage of Chinese models relative to American ones in that specific monitored time window.

The institute’s head, Liu Wenqiang, commented on the shift toward the application phase with a statement translated from Chinese:

“Artificial intelligence is reshaping the industrial system at an unprecedented pace” (transl. from Chinese) — Liu Wenqiang, Communist Party committee secretary and deputy director of the China Center for Information Industry Development

The shift described in the report — from the model training phase to large-scale daily use — indicates, according to the Chinese institute, that the domestic market is now in a condition in which the operational consumption of artificial intelligence systems exceeds, in terms of data volume processed, the computational effort devoted to the development of the models themselves.