When inference overtakes training
Inference is the process of running trained AI models to respond to user requests, which involves processing units of data known as tokens. Token calls through public cloud providers are a measure of the growing demand for AI services. The report did not disclose how much data is being used to train AI systems in China.
- Average daily token calls through public clouds: more than 175 trillion.
- Increase since early 2024: more than 1,000 times.
- Total daily token consumption including private clouds as of June: around 500 trillion, according to the China Academy of Information and Communications Technology.
China's National Data Administration estimated in April that demand for computing power for AI inference could eventually reach three times that needed for training, underscoring the growing need for infrastructure.
The “token economy” in national strategy
Beijing has increasingly stressed the importance of tokens in the AI era. Several local governments, including Beijing's, have announced plans to develop the token economy, with measures to boost token consumption through “token factories” and token distribution platforms.
















