Physix FrontierPhysix Frontier · News Briefing Card (IT Home · Nov 1, 2026)
CCID: China's AI inference data volume tops training data for first time
KEY FACTS
- The China Center for Information Industry Development (CCID) released a report stating that China's AI inference data volume has surpassed training data volume for the first time.
- As of June 2026, China's average daily token calls exceeded 175 trillion, more than 1,000 times the level at the start of 2024.
- The iteration cycle for top-tier large models has shrunk from months to weeks or even days.
- China has built more than 500 excellence-grade smart factories, covering 31 provincial-level regions.
- Domestic compute and large models have been deployed across more than 1,000 application scenarios in 20 industries.
KEY DATA
Over 175 trillionAverage daily token calls
More than 1,000xToken call growth vs. early 2024
More than 500Number of excellence-grade smart factories
Over 700 millionGenerative AI user base
PHYSIX OBSERVATION
Inference data volume overtaking training data volume is the watershed marking AI's move from the lab into industry. Thousand-fold token call growth shows demand has scaled up, but the real test is whether compute costs can sustain this consumption. For vendors, the era of selling models is giving way to selling services; whoever can drive inference costs down holds the ticket into the next stage.
Source: IT Home report