China touts AI milestone: models generate more tokens than used in training

The surge in "token calls" observed in China points to a significant increase in the operational use of artificial intelligence. "Inference" refers to the process where trained AI models respond to user requests by processing data units known as tokens. The reported figure of over 175 trillion average daily token calls, a substantial jump from early 2024, suggests that AI models are moving beyond initial development into widespread application.

For the first time, the volume of inference data generated by these systems has surpassed the amount of data originally used for their training. This milestone, highlighted by the government-backed CCID report, suggests a shift in focus from the creation and refinement of AI models to their active deployment and engagement with users at scale. It implies that the models are not just being built, but are being heavily utilized in real-world scenarios.

The measurement of token calls through public cloud service providers offers an indication of the practical demand for AI services. While the specific applications driving this demand are not detailed, the sheer volume suggests a robust and expanding user base interacting with AI. This data, originating from a government-backed report, frames the situation as a marker of China's accelerating adoption of AI technology.