China plans national standards for counting and billing AI tokens

China's National Data Administration says it will set rules for measuring and billing AI tokens, and a draft national standard already spells out how reasoning and cached tokens should be counted.

Illustration of a factory conveyor belt carrying glowing coin-like tokens past a large meter and a clipboard with a checklist, in warm orange light. Image: AgenticFeed illustration

China's National Data Administration says it will write standards for how AI tokens are measured and billed, along with quality tests, as Beijing tries to turn the units that AI models process into a regulated product. Liu Liehong, the agency's head, set out the plan in a speech on November 20 at the opening of the 2026 World Computing Conference in Changsha, Hunan, according to Yicai, which cited Securities Times.

The remarks were posted on the agency's website on Saturday, Bloomberg reported. The agency did not say when the standards would be published, The Next Web noted.

Token factories

Liu described tokens as the link that carries value from computing power through to models and data, calling them a "value anchor" of what Beijing terms the digital and intelligent economy, according to Yicai's report of the speech. He said an industry built around producing and selling tokens is "accelerating" into shape.

He singled out what officials call "token factories," businesses that package rented computing power and model supply, plus data capabilities, into a token production service. Liu said these were emerging quickly as a new form of industrial organization, Lianhe Zaobao reported.

He also said tokens were creating a new way to deliver services. Industry datasets are turned into measurable intelligent services through model training and inference, then sold by the token and priced per call, according to the speech.

Next, Liu said, the agency would expand the ways token services are used and improve standards for token billing and quality evaluation. It would also "resolutely avoid low-quality duplicate construction," Yicai reported.

Usage figures

Beijing has pushed the idea of a token economy all year, The Next Web reported. In March, Liu said China's daily token use had passed 140 trillion, more than 1,000 times the level of early 2024, according to Yicai figures cited by The Next Web. By June the daily figure had passed 500 trillion, according to agency data cited by People's Daily.

Consumer demand is rising too. Tmall data showed sales of AI token packages from October 15 to November 5 rose by more than 700% compared with the preceding period, Lianhe Zaobao reported. The paper also said livestream host Li Jiaqi sold 100,000 monthly subscriptions to Alibaba's QwenWork within half an hour during the Singles' Day shopping festival.

What a counting standard covers

Some of the technical work is already public. A recommended national standard titled "Intelligent computing, Token service, Part 2: Counting rules" is listed on China's national standards information platform. It is managed by the national working group on intelligent computing standards under the Standardization Administration of China, has a 15-month project cycle and was open for public comment from September 30 to October 30, the listing shows.

According to the listing, the standard sets a common framework for counting tokens that consume computing resources. It separates the counting stages of input encoding, prefill, KV cache reuse and decoding.

The draft also sets requirements for input, output, cached, reasoning and multimodal tokens. It also defines a shared data format for usage fields so that counts from different platforms can be compared.

The listing says the rules apply to providers counting and disclosing tokens. It also says regulators and third-party auditors can use them to verify counts, and that customers can use them to check costs.

Drafters named on the listing include Zhejiang Lab, the National Institute of Metrology, the China Academy of Information and Communications Technology, Alibaba Cloud, Baidu, Ant Group, DeepSeek, Zhipu and Tsinghua University. China Mobile and China Telecom are also listed.

The separate categories matter for AI agents, which run many model calls per task and can generate large volumes of reasoning and cached tokens that never appear in the final answer. The draft is listed as a recommended standard, not a mandatory one, and the National Data Administration has not said how its billing and quality rules will be enforced.

Source: Yicai, Yicai, “国家数据局:积极拓展词元服务应用场景 健全词元计量计费、质量测评等标准规范”

Additional sources: Lianhe Zaobao, Lianhe Zaobao, “China to strengthen AI token standards as consumer demand for AI computing services surges”; Moneycontrol (Bloomberg), Bloomberg, “China to tighten AI token measurement, billing standards as 'token factories' emerge”; The Next Web, The Next Web, “China's data regulator plans standards for measuring and billing AI tokens”; National Public Service Platform for Standards Information, SAMR, “Intelligent computing, Token service, Part 2: Counting rules”

Get the AgenticFeed newsletter

A free, regular roundup of the agent stories that matter. No spam, and you can unsubscribe at any time.

Want to reach readers following AI agents? Advertise with AgenticFeed

Next event · 30 Nov, 2026

AWS re:Invent 2026 in Las Vegas

AWS’s annual conference runs Nov. 30 to Dec. 4. Matt Garman’s CEO keynote is on Dec. 1, and the catalog lists more than 2,200 sessions.

The AgenticFeed newsletter

A free, regular roundup of the agent stories that matter. No spam.