Since March 2026, the three major telecom operators have successively incorporated Token (the smallest unit of information processed by large models, featuring measurability, priceability, and tradability) operations into their strategic layouts. Provincial branches of China Mobile, such as Guangdong Mobile and Jiangsu Mobile, as well as provincial branches of China Unicom, such as Hubei Unicom and Shanghai Unicom, have taken the lead in launching Token packages for individual users.
Around the World Telecommunication and Information Society Day (May 17), telecom operators have made varying degrees of progress in the national layout of Token packages, and the industry as a whole has gradually shifted from decentralized provincial exploration to large-scale promotion. Since 2025, Token invocation volume has shown explosive growth. According to the National Data Administration, by the end of March 2026, China's average daily Token invocation volume had exceeded 140 trillion, an increase of over a thousand times compared to the beginning of 2024. The industry is transitioning from the experimental period to the large-scale application period.
However, while providing new types of intelligent services, this new business has also brought about urgent issues to be resolved in terms of defining business attributes, clarifying compliance boundaries, and safeguarding user rights. Therefore, this article systematically analyzes the business nature of Token packages and proposes suggestions for the development direction of the industry.
01. Current Status of Business Development
Token packages have become an important exploration direction for the strategic transformation of telecom operators. On May 17, China Telecom officially launched a nationwide trial commercial Token package, marking that the industry has entered the stage of unified national layout from the stage of decentralized exploration by provincial branches.
Background of Strategic Transformation: Slower Growth of Traditional Business, Computing and AI Integration Becoming a New Growth Engine
In recent years, the revenue growth of traditional businesses for the three major telecom operators has been sluggish, while the revenue from computing power and AI (Artificial Intelligence) services has achieved rapid growth. According to the data disclosed in the 2025 annual reports of each operator: China Mobile's computing power service revenue reached 89.8 billion yuan, a year-on-year increase of 11.1%, of which the intelligent computing service revenue growth rate reached 279%, becoming the "primary engine" for computing power service growth; China Unicom's AI business revenue increased by more than 140% year-on-year; China Telecom's intelligent service revenue reached 12.3 billion yuan, with strategic emerging businesses maintaining rapid growth. Driven by the dual factors of traditional traffic operations approaching the "ceiling" and the rapid development of the AI industry, the three major telecom operators have taken Token operations as the direction of strategic transformation. At the 2026 Mobile Cloud Conference, China Mobile explicitly proposed "communication services, computing power services, and intelligent services" as its three main businesses, making Token operations an important foothold for computing power and intelligent services.
Current Product Forms: Significant Differences in Billing Methods and Underlying Models, Low Degree of Product Standardization
Judging from the products currently launched in various regions, telecom operators' Token packages mainly present the following characteristics.
First, the billing form is mainly "monthly packages", but the measurement units show diversity. Currently, the AI large model packages of the three major telecom operators are mainly monthly packages, with differences in billing units. The personal version of China Telecom has a monthly fee of 9.9 to 49.9 yuan, including 10 million to 80 million Tokens; the enterprise version has a monthly fee of 39.9 to 299.9 yuan, including 15 million to 150 million Tokens. The personal version of China Unicom has a monthly fee of 15 to 45 yuan, including 6 million to 18 million Tokens; the team version has a monthly fee of 198 to 1,398 yuan, including 25,000 to 250,000 Credits (virtual billing units, with no unified standard for the conversion ratio with Tokens). China Mobile prices its packages independently by province, with varying monthly fees and Token volumes, and some provinces have additionally set up computing power usage packages. The billing units of the three major telecom operators include "Tokens" and "Credits", making it difficult for users to horizontally compare the actual available volume.
Second, the underlying models present an integrated pattern of "third-party foundation + self-developed vertical models". Currently, at the general foundation level, telecom operators' Token packages mainly integrate third-party large models such as DeepSeek, Qwen, and MiniMax, leveraging their technological maturity and open-source ecosystems to quickly build service capabilities. Meanwhile, operators' self-developed models are accelerating their implementation. China Mobile's "Jiutian", China Telecom's "Xingchen", and China Unicom's "Yuanjing" have launched industry large models for vertical scenarios such as government affairs, education, and industrial manufacturing, and achieved commercialization. However, in the underlying foundation of consumer-grade Token packages, third-party models still account for a large proportion. The processing channels are mainly online. Individual users can subscribe online through the operators' APPs or cloud platforms, while services for enterprise users support offline docking and customized delivery.
02. Analysis of Business Characteristics
This article discusses Token packages provided by telecom operators to individual, family, and enterprise users. These packages adopt a package billing model and are mainly sold through self-owned channels. Users purchase the usage volume of AI model inference services rather than traditional communication services. Their service forms possess the composite characteristics of both telecom services and intelligent services. Compared with the model of cloud service providers, which mainly targets enterprises and developers and is primarily based on pay-as-you-go billing, telecom operators' Token packages focus more on reaching end consumers in the form of monthly packages and usage packages, making the telecom service attributes more prominent.
Business Evolution Logic: Token Packages as a Natural Extension of the Expansion of Telecom Operators' Business Boundaries
From the perspective of business evolution, the business boundaries of telecom operators have been continuously expanding. They have experienced or are experiencing multiple evolutions, transitioning from basic communication to value-added services, from fixed-mobile convergence to cloud-network convergence, and from traffic operations to Token operations. Token packages belong to the natural extension of this evolution context and serve as a specific carrier for operators to transform from "connectivity providers" to "computing power operators" and "intelligent operators".
Service Content and Attributes: Users Purchase the Usage Volume of AI Model Inference Services Rather Than Traditional Communication Services
In terms of service content, what users purchase is not traditional communication services (such as voice, SMS, data traffic, etc.), but the quota for invoking intelligent services based on large language models. The number of Tokens is used as the measurement and billing unit, which essentially belongs to the category of intelligent services. However, from the perspective of the sales model, operators have adopted the "package" billing method of traditional telecom services, charging users fixed fees in the form of monthly packages or usage packages. In addition, from the perspective of the provider, operators face end users, possessing self-owned channels and a massive user base. From the perspective of user rights, it involves core rights such as user tariffs, service quality, and personal information protection, which is similar to the dimension of user rights protection in telecom services.
Analysis of Rule Application: Regulatory Application Paths for Service Models and Technical Cores
Given that Token packages are provided by operators to individual, family, and enterprise users, adopt the package billing model of traditional telecom services, are sold through operators' self-owned channels, and involve core rights such as user tariffs, service quality, and personal information protection, they possess the composite characteristics of "the form of telecom services and the essence of intelligent services".
Regarding the issue of rule application, it is advisable to examine it from two dimensions: business model and technical core. The former covers package sales, user rights protection, notification obligations, tariff disclosure, and fair trade. At the institutional level, it is highly correlated with traditional telecom services regulated by the "Regulations of the People's Republic of China on Telecommunications" and relevant normative documents of the Ministry of Industry and Information Technology. The latter covers links such as AI model scheduling, data flow security, and content generation, as well as emerging technical elements such as Token measurement. It intersects with the regulatory scope of laws and regulations such as the "Interim Measures for the Management of Generative AI Services", the "Personal Information Protection Law of the People's Republic of China", and the "Data Security Law of the People's Republic of China". Meanwhile, the designs of billing and settlement, service quality, and data security for this service can also seek corresponding regulatory references within the framework of the current legal and standard systems.
03. Potential Issues Requiring Industry Attention
As an emerging business form, Token packages have structural issues worthy of attention in terms of billing transparency, data security protection, service quality notification, and the uniformity of operational norms. If not effectively addressed, these issues may constrain user cognition, industry synergy, and sustainable development.
Complex Billing Rules Pose Challenges to User Expectation Management
The billing rules for Token packages are more complex than those for traditional telecom services, mainly reflected in the following four aspects.
First, the measurement units are not unified. Each operator bills using "Tokens", "Credits", "number of requests", etc. There is a lack of conversion standards between units, making it difficult for users to compare prices. For example, the actual available volume of "10 million Tokens" and "10 million Credits" varies significantly due to model differences.
Second, the consumption speed is difficult to predict. Token consumption is affected by multiple factors such as input and output length, the model's Token counting rules, model complexity, and invocation strategies. It is difficult for users to estimate usage based on application scenarios, which is easy to cause "usage anxiety". Even if the units are the same, different models have differences in the segmentation methods for the same text and billing calibers (such as whether to distinguish between input and output), leading to different actual task completion volumes for the same number of Tokens, further exacerbating the sense of unpredictability.
Third, there are significant differences in the capabilities of different models. Various packages generally integrate mainstream models such as DeepSeek and Qwen. Different models have differences in reasoning ability, response speed, professional fields, and knowledge timeliness. It may be somewhat difficult for users to compare and form expectations at the purchase stage.
Fourth, the quality of generated content is difficult to evaluate in a standardized manner. AI-generated content has uncertainty, and the same question may generate different answers, making it easy for user expectations to deviate from actual experiences. Unlike traditional telecom services that have clear quantitative indicators, the quality of intelligent services lacks a unified measurement standard, which may lead to user dissatisfaction or commercial disputes, forming a potential impact on the industry's service reputation.
Long Data Flow Chain Requires Improvement in Personal Information Protection Mechanisms
Token packages involve deep data interaction between users and large models. Their data flow chains are relatively long, and service responses may be completed through multiple links such as the operators' unified service platforms and large model service providers. Currently, special rules regarding security protection, notification and consent mechanisms, and data usage agreements for AI large model interaction data during the flow process are still being continuously improved. In terms of notifying key information of Token packages such as data retention periods and cross-border processing risks, existing privacy policies are mostly general provisions, and special notifications for AI interaction data are still insufficient. The agreement methods and transparency regarding whether data is used for secondary utilization such as model training in privacy policies are not yet clear, which may become a weak link in industry trust building.
Diverse Promotion Models Warrant Attention to Cross-Regional Experience Consistency
Currently, there are significant differences in the promotion pace and models of Token packages among the three major telecom operators. China Telecom took the lead in launching a nationwide trial commercial Token package at the group level, achieving unified pricing, setting multiple tariff tiers for individual, family, developer, and small, medium, and micro enterprise users, and covering the national market. China Mobile adopts the strategy of "provincial company pilot first + group ecosystem promotion". At the group level, it innovatively introduced a unified computing power dimension, packaging Token consumption of different models and specifications into standard points. Provincial companies such as Beijing, Shanghai, and Guangdong took the lead in pilot programs, and on May 20, the group company announced a nationwide launch. China Unicom mainly relies on provincial branches for pilot testing. Regions such as Shanghai, Hubei, and Sichuan have launched differentiated Token packages, with varying focuses on tariffs and package design. The promotion models of the three major telecom operators are different, and the tariff standards vary greatly, leading to inconsistent package service experiences for users in different regions. Moreover, Token packages in different provinces of the same operator also differ in functions, prices, and user coverage. Coupled with the high pricing authority and diverse strategies of each province, it is easy to cause cognitive confusion among users, forming a certain impact on the overall brand image and user trust of the industry.
04. Current Status of Business Development
In response to the composite characteristics of Token packages having "the form of telecom services and the essence of intelligent services", as well as the current situation where the existing rule system is still being improved, the industry can strengthen forward-looking research and self-discipline exploration from the following dimensions to promote the healthy and sustainable development of the business.
Promoting the Formation of Industry Consensus on Safeguarding User Rights
Given the composite characteristics of Token packages, to balance innovation inclusiveness and rights protection, it is recommended that the industry form consensus in the following directions within the existing service specification framework, focusing on users' right to know and right to choose.
First, enhance the transparency of model information. Marketing promotions should clearly, understandably, and prominently display key matters such as the name, version, provider, functions, applicable scenarios, known limitations, and risk warnings of the accessed AI models on the same page. For those involving multi-model switching, the switching rules, billing differences, and capability boundaries should be explained together, avoiding concealing or downplaying the above information through folding, hiding, blurring, or other forms.
Second, strengthen informed confirmation during business processing. When handling Token package subscription, modification, renewal, and other businesses, it should be ensured that users give informed consent. In particular, user responsibilities and restrictive clauses such as model capability boundaries, data usage scope, and tariff calculation rules should be explicitly stated in the agreement and users should be proactively reminded to pay attention. Business processing should obtain explicit user consent through methods such as SMS reply confirmation and mobile phone verification code confirmation.
Third, explore a consumption threshold reminder mechanism. Referring to the existing traffic service consumption reminder mechanism, operators should provide Token package users with a Token consumption estimation reference guide, explaining the typical consumption of common AI application scenarios and the reasons for estimation deviations. At the same time, they should provide Token consumption threshold reminder and alarm functions, promptly informing users via SMS, APP, and other methods when usage approaches the limit. If services need to be continued after the expiration of free trials or promotional activities, user consent must be obtained again.
Fourth, pay attention to the research on quantitative indicators of service quality. For matters such as quantitative indicators of service quality and relief and compensation rules for disputes over the quality of generated content, it is recommended that the industry jointly carry out research, and explore and promote them as appropriate after the relevant technical standards and institutional rules are relatively mature.
Strengthening Industry Self-Discipline in Data Security and Personal Information Protection
In response to the characteristics of Token packages involving user and large model interaction data, it is recommended that the industry strengthen self-discipline exploration from the following aspects: First, before business processing, in accordance with the "Personal Information Protection Law of the People's Republic of China" and other relevant regulations, fully inform users of data collection, usage, and storage policies, and obtain explicit user consent. If it involves providing user personal information to third-party large models, separate consent from users should also be obtained in accordance with Article 23 of the law. Second, when invoking third-party large models, it must be ensured that users are informed in advance, the process is controllable, and responsibilities are traceable regarding the acquisition of user data by third parties, retention periods, and usage boundaries. Third, enterprises should bear the main responsibility for user data security and establish a security assessment and exit mechanism for third-party model service providers.
Promoting Standardization Research on Business Rules and Making Forward-Looking Layouts for Risk Prevention
Given that Token packages belong to an emerging business format, it is recommended that the industry carry out forward-looking research around three directions: billing rules, service quality standards, and risk prediction. In terms of billing rules, the focus should be on exploring industry standards for Token consumption, researching standardized billing conversion systems for different models, as well as applicable scenarios and pricing strategies for pay-as-you-go and package-based billing. In terms of service quality standards, the focus should be on researching the AI service quality evaluation indicator system (such as response speed, accuracy, accessibility, etc.), exploring relief mechanisms for user dissatisfaction with generated content, and researching the applicability boundaries of AI services in different scenarios. In terms of risk prediction, typical dispute scenarios should be systematically sorted out and response plans formulated, and security assessment methods for third-party model service providers should be researched.
This article was published in the 7th issue of "Telecom World" in 2026; original title: "Analysis of the Business Nature and Industry Evolution Logic of Token Packages under the Background of 'Computing and AI Integration'".
Authors: Wang Haiyan, Wu Jun (Corresponding Author), Sun Lixue, Zhang Yang, China Academy of Information and Communications Technology.