One-sentence summary
MIIT's CCID says that by June 2026 China's AI inference data volume overtook training, with daily token calls over 175 trillion, up over 1,000x from early 2024, marking a shift from building to using models.
The Data
On November 1, 2026, per CCTV, MIIT's CCID research institute released data: China's AI inference data volume has overtaken training data volume for the first time; as of June 2026 daily token calls exceeded 175 trillion.
The Growth
Daily token calls are up over 1,000x from early 2024. The bulk of AI data processing has shifted from the training phase to inference and real-world use.
What the Inflection Means
Inference overtaking training shows the industry center of gravity moving from training bigger models toward running models in high-frequency real business. Compute, inference cost, and deployment scenarios rise in importance.
Echoing Open Source
Together with the Wuzhen data on Chinese open models leading weekly usage, inference overtaking training and low-cost open-model scaling reinforce each other, reflecting fast expansion on China's AI application side.
Sources
- Ifeng Tech (via CCTV): China's AI inference data volume overtakes training