China Reports AI Systems Generate More Inference Data Than Training Data

China’s artificial intelligence systems have generated more inference data than the data used to train them, marking the first time this has happened in the country. The development illustrates China’s growing adoption of AI at scale, according to a new government-backed report.

Token calls point to rising AI demand

The report was released on Saturday by the China Centre for Information Industry Development, which operates under China’s Ministry of Industry and Information Technology. It cited recent official data showing that China’s average daily token calls exceeded 175 trillion.

The figure represents an increase of more than 1,000 times from early 2024. Tokens are units of data processed by AI systems, while inference refers to running trained models to respond to user requests.

The difference between training data and inference data reflects the expanding use of trained AI models after they are developed. Inference data is generated when users or systems interact with those models and receive responses.

The report said the volume of token calls refers to AI usage through public cloud service providers. That makes token calls an indication of the growing demand for AI services as adoption accelerates.

The milestone therefore places the focus on the scale of AI use in China, rather than only on the data involved in building the models. The report’s figures point to a sharp rise in activity through public cloud services and in the use of trained models to process requests.

Source: South China Morning Post