China pushes token-based AI services and planned standards as computing capacity grows

By
CTOL Staff Reporter
1 min read

China is pushing token-based AI services and planning standards for measurement, billing and quality as computing capacity grows. We forecast 80% odds that no token-measurement or token-service-quality national standard draft will be published for public comment on the national standards platform by 30 June 2027; we would be wrong if such a draft is published there for public comment by that date.

Alibaba Cloud prices a qwen3.7-plus cache hit at 10% of its ordinary input-token rate. Baidu quotes ERNIE-5.1 by the thousand tokens, while Volcano Engine quotes doubao-seed-2.1-pro by the million. All three sell the unit China wants to make central to its AI economy, yet their invoices speak different dialects. The meter is already running; the common ruler is still being assembled.

At the World Computing Conference in Changsha on 20 November 2026, National Data Administration chief Liu Liehong described tokens as a unit for AI service delivery and measurement and spoke of “token factories” packaging computing, models and data into services. The administration plans to improve standards for measurement, billing and service quality. Buyers cannot yet assume that the same token label means the same consumption count or bill across providers. Baidu's per-thousand-token quotes and Alibaba's and Volcano's per-million-token quotes require conversion to matching units, with currency differences accounted for, before buyers can compare costs. A standards announcement alone gives investors no evidence that token usage will produce earnings at the quoted tariffs.

We call the mismatch Meter Lag: providers can sell services under a common unit name while their accounting conventions remain their own. The National Data Administration's 15 October 2026 notice solicited three proposed national standards, covering multimodal conversion, token measurement and production-service quality. Declarations closed on 31 October 2026. In August 2026, the AI Industry Alliance had already issued AIIA/T 0308-2026, an industry specification addressing usage statistics, billing and account management. These existing billing conventions already shape how suppliers get paid.

The money begins with the number of times that meter turns. The administration's 25 March 2026 account put average daily token calls above 140 trillion, compared with 100 billion at the start of 2024 and 100 trillion at the end of 2025. People's Daily via Xinhua on 21 November 2026 said daily calls had exceeded 500 trillion, without specifying the measurement month. The thresholds of 500 trillion over 140 trillion give roughly 3.6, a comparison of thresholds rather than a precise growth rate. At that scale, counting rules affect bills paid by customers and the prices suppliers can defend.

Capacity makes the issue more urgent. China News Service quoted Liu placing national computing capacity at 3.1 million PFLOPS, second globally. In a separate 20 November 2026 account from Xinhua, the China Center for Information Industry Development put intelligent compute at 2,592 EFLOPS on an FP16 basis at the end of September, up 146% year on year. These figures describe separate categories of computing capacity. The institute envisages RMB 120 billion to RMB 130 billion in compute trading by 2030. More capacity for sale increases the commercial value of a unit that purchasers can compare across suppliers.

Look at the bills rather than the national total. Alibaba's qwen3.7-plus lists $0.40 for a million input tokens and $1.60 for a million output tokens at its initial input-length tier, with different rates for cache hits. Baidu's ERNIE-5.1 lists RMB 0.004 input and RMB 0.018 output per thousand; multiplying each by 1,000 gives RMB 4 and RMB 18 per million. Volcano's doubao-seed-2.1-pro lists RMB 6 and RMB 30 per million. In a scenario with one million input and one million output tokens, Baidu’s RMB 4 plus RMB 18 gives RMB 22; Volcano’s RMB 6 plus RMB 30 gives RMB 36, a RMB 14 difference. The models differ in performance and terms; that nominal spread helps frame a purchasing comparison.

China Unicom Digital Intelligence, a co-author of the alliance specification, lists more than 230 models on its model-service platform. ByteDance's Volcano Engine faces the opposite conditional exposure: if common measurement makes its RMB 6 input and RMB 30 output tariffs directly comparable with Baidu's lower quoted rates, it must justify the difference through service quality or terms.

There is an excellent argument for a faster national timetable. In People's Daily's account carried by Xinhua, Cyberspace Administration deputy director Wang Jingtao called tokens “measurable, priceable and tradable.” It also cited 119.1 trillion weekly tokens for Chinese models on OpenRouter during 9 to 15 November 2026, and domestic shopping platforms offering token top-ups. China Electronics Standardization Institute gathered more than 20 organizations on 3 November 2026 to discuss the token standards system. The existing alliance specification gives those organizations an operational starting point. A fast national text would reward that existing work.

The speed argument has precedents. Data Basic Terminology reached public comment 175 days after its reported national project plan. The AI-generated content labeling standard went from public comment on 14 September 2024 to national release on 28 February 2025, a 167-day interval. Generative AI security requirements took 337 days from consultation on 23 May 2024 to publication on 25 April 2025. The TC609 first-batch needs call became an April list in about 88 days. Six important-data catalogue texts went to public comment on 17 November 2026, while that batch contained no token standard. Of the three examples with dated plan or consultation-to-publication transitions, all three advanced, but each started at a different milestone. That selection offers little basis for estimating the chance of a June token consultation.

The nearer comparison is the 242 days between the token declaration deadline of 31 October 2026 and our 30 June 2027 evaluation date. The 175-day Data Basic Terminology example began with a plan already on the books. The token measurement and service-quality proposals still require approval and consultation text. The extra approval step deserves more weight than the pace of preparatory meetings. Formal approval of either project on the national standards platform while at least 175 days remain before 30 June 2027 would move our assessment to three in ten.

National publication is only one route to common accounting. An alliance specification or TC609 technical guidance can shape contracts sooner; an administration-led statistical definition can standardize national token counts without settling a buyer's model bill. Our publication test counts a measurement or service-quality national standard on the national platform; alliance rules, statistical definitions and a multimodal-conversion-only text take other routes. Different billing units, workloads and caching terms can make superficially similar token prices produce different bills.

Buyers who overlook output and caching charges risk mistaking a low input price for the lowest total bill. We forecast 80% odds that no token-measurement or token-service-quality national standard draft will be published for public comment on the national standards platform by 30 June 2027 (20% odds that one will); we would be wrong if such a draft is published there for public comment by that date.

You May Also Like

This article is submitted by our user under the News Submission Rules and Guidelines. The cover photo is computer generated art for illustrative purposes only; not indicative of factual content. If you believe this article infringes upon copyright rights, please do not hesitate to report it by sending an email to us. Your vigilance and cooperation are invaluable in helping us maintain a respectful and legally compliant community.

Subscribe to our Newsletter

Get the latest in enterprise business and tech with exclusive peeks at our new offerings

We use cookies on our website to enable certain functions, to provide more relevant information to you and to optimize your experience on our website. Further information can be found in our Privacy Policy and our Terms of Service . Mandatory information can be found in the legal notice