AI inference data surpasses training data in China

Nov 2, 2026
AI inference data in China has surpassed model training data for the first time, as average daily token calls exceed 175 trillion.

AI systems in China now process more data during inference than is used to train them, reports the South China Morning Post. A report from the China Center for Information Industry Development, which operates under the Ministry of Industry and Information Technology, said average daily token calls exceeded 175 trillion, more than 1,000 times the level in early 2024.

The report did not disclose how much data is used to train AI systems in China. A separate estimate from the China Academy of Information and Communications Technology put average daily token consumption at about 500 trillion as of June when private cloud use was included.

China's National Data Administration estimated in April that computing demand for AI inference could eventually reach three times the demand for model training. Capital spending by 13 major Chinese companies on AI infrastructure is projected to more than double to 932 billion yuan in 2026 and exceed 1.2 trillion yuan in 2027, according to Rhodium Group.

We hope you enjoyed this article

Consider subscribing to one of our newsletters like Silicon Brief or Daily AI Brief.

Also, consider following us on social media:

AI Chips & Datacenters AI Brief AI Brief (X)
Free newsletter

Silicon Brief

Weekly coverage of AI hardware developments including chips, GPUs, cloud platforms, and data center technology.

Guide
AI interconnect and networking startups

AI interconnect and networking startups

29 private startups developing the chips, optical switches and memory fabrics that move data through AI infrastructure.

Whitepaper

Tensordyne Napier: What If One Rack Could Do the Work of Nine?

Tensordyne

This Tensordyne whitepaper presents Napier, an inference-focused AI processor and rack-scale system based on the company’s TDN Math logarithmic number system. It examines infrastructure requirements for large mixture-of-experts and agentic models, compares major inference architecture approaches, and details the TDN AIP processor, TDN72 pod, TDN Link fabric, and Napier Ultra configuration. The paper reports simulation-based performance, cost, and accuracy-validation results, including Tensordyne’s projected comparison of one Napier rack with a nine-rack Nvidia Rubin plus Groq deployment; the chip is reported as taped out and in fabrication.

Read more

You may also like

Free, six days a week

Daily AI Brief: the AI news that matters, in your inbox.