Tokens have completely caught on, and the three major telecom operators have finally taken action.

On May 17th, it's World Telecommunication Day.

China Telecom took an action on this day, launching a trial commercial Token package nationwide. The minimum price for individual users is 9.9 yuan, which includes 10 million Tokens. Shanghai Mobile launched a package of 400,000 Tokens for 1 yuan, and Beijing Mobile launched a monthly package of 10 million Tokens for 24.99 yuan. China Unicom was also not idle. Shanghai Unicom provided 30 million Tokens for free to OPC customers for testing, and Hubei Unicom combined communication, broadband, and Tokens.

As of today, all three major operators have launched Token packages. Thus, following phone charges, text message fees, and data traffic fees, a new charging item will appear on the monthly communication bills of Chinese people.

In the view of those who have been in the communication industry for 26 years, the significance of this event far exceeds the innovation of pricing. The last time operators collectively adjusted the billing unit was during the popularization of 4G. It changed from call minutes to the number of text messages, and from data traffic in MB to the current Tokens. However, Tokens are different from the previous ones. Minutes are sold for connections, data traffic is sold for pipelines, and Tokens are sold for "intelligence".

As computing power is packaged into packages like water and electricity and the phone bill is deducted monthly, we are witnessing a transformation at the infrastructure level.

Why Tokens Become Hard Currency

Let's first study the packages of the three operators.

China Telecom made the most significant move. It set a unified price at the group level, covering the whole country. Both individuals, developers, and enterprises have three options. China Telecom's package integrates its own Xingchen large model and DeepSeek V3.2 and other ecological large models, which can be used for daily office assistance, writing, learning, and creation.

China Mobile's approach is more flexible, advancing province by province. Shanghai Mobile is the most aggressive, offering 400,000 Tokens for 1 yuan, supporting payment with phone bills and cross - platform use, and also jointly launching an AI native workbench with Tencent. Beijing Mobile launched a package of 10 million Tokens for 24.99 yuan per month. Existing users can get additional discounts, and the lowest price for a single - purchase package is 5.99 yuan. Jiangsu Mobile even launched an experience package of 2.5 million Tokens for 5 yuan, setting the trial threshold extremely low.

China Unicom's approach is different. Hubei Unicom took action at the end of April, launching three levels of Token Plan and two levels of Coding Plan, integrating communication, broadband, and cloud desktops. Shanghai Unicom targeted single - person companies directly, providing 30 million Tokens during the testing period. Sichuan Unicom launched a "Family Token Package", with a fixed monthly fee combined with China Unicom's cloud desktop and WorkBuddy intelligent agent, aiming to solve the pain point of "not knowing how to use" with "ready - to - use" services.

Let's do a simple conversion: The lowest - level package of China Telecom for individuals is equivalent to 0.99 yuan per million Tokens, which should be the cheapest among the unified prices of the three groups; Shanghai Mobile's 1 - yuan package for 400,000 Tokens is equivalent to 2.5 yuan per million Tokens; the renewal price for OPC customers of Shanghai Unicom is 1 yuan per million Tokens.

Just looking at the unit price, China Telecom offers the most competitive price, but China Mobile wins in terms of flexibility, as users don't need to subscribe monthly and can buy as much as they use. China Unicom's differentiation lies in scenario integration, packaging Tokens with cloud desktops and intelligent agents, and selling "ready - to - use" services.

However, if you think that operators are engaging in a price war, you're underestimating it. At this stage, this pricing is more like the logic of "cultivating user habits" in the early days of data traffic management.

After discussing the packages, we need to return to a fundamental question: What is a Token?

Strictly speaking, a Token is the smallest computing unit for large models to process text, images, and voice. When you say "What's the weather like in Beijing today?", the AI breaks it into five or six Tokens for processing; every word it replies to you is also a Token.

In the past two years, the circulation of Tokens on the Chinese Internet has experienced an exponential growth in a physical sense. According to the statistics of the National Data Bureau, at the beginning of 2024, the daily average Token calls in China were 100 billion; by the end of 2025, it soared to 100 trillion; in March 2026, it exceeded 140 trillion. It has increased by more than a thousand times in two years.

In March this year, the National Science and Technology Terminology Approval Committee officially set the standard Chinese translation of "Token" as "Word Element". When a technical term requires a national - level press conference to determine its name, it means that it has changed from a pricing unit within large - model manufacturers to a consumable at the whole - society level. It's not data traffic, nor bandwidth, but a brand - new 'basic resource'.

Jensen Huang pointed out this clearly at the GTC conference. He said: "Tokens are the new commodities." The term "commodities" is not used casually. It has specific meanings: standardization, gradability, and the ability for large - scale trading.

At GTC, NVIDIA launched a five - level Token pricing system, ranging from a free level to an ultra - high - speed level of 150 US dollars per million Tokens, graded by speed and context length. This is a bit like what the telecommunications industry has always wanted to do but failed to achieve - differentiated pricing of data traffic according to quality. Jensen Huang implemented this idea first for them.

Also in this month, the domestic large - model industry experienced a round of sharp price re - evaluation. Tencent Cloud took the lead in raising the price of its Hunyuan large model. The input price of the core model increased from 0.0008 yuan per thousand Tokens to 0.004505 yuan, a rise of 463%. Subsequently, Alibaba Cloud and Baidu Smart Cloud followed suit.

By May, the pressure was transmitted to the C - end. ByteDance's Doubao quietly posted three levels of paid subscriptions on the App Store page. At that time, the daily average Token usage of the Doubao large model had exceeded 120 trillion, a thousand - fold increase compared to its release. According to the calculation of Zheshang Securities, ByteDance's capital expenditure in 2025 is about 160 billion yuan, most of which is spent on AI computing power procurement.

DeepSeek, which is open for commercial use, took a different path. The input pricing of V4 - Flash was reduced to 1 yuan per million Tokens, and after cache hits, it was only 0.2 yuan, almost at the cost line. However, even so, the pricing curve of the entire industry has clearly turned upward.

That is to say, at this critical point of transitioning from "free all - you - can - eat" to "pay - as - you - go", operators entered the market with phone bills.

The Operators' Calculations

The fact that the three major operators simultaneously launched Token packages is not a coincidence.

The traditional communication business has reached its ceiling. The unit price of data traffic has been decreasing, and the voice and text message business has been shrinking for many years. Operators' anxiety about being reduced to mere pipelines has been around for a long time - you build the network, transmit the data, but the Internet companies make the money.

Tokens give them a chance to turn things around.

First, they have an existing computing power base. China Mobile's total intelligent computing scale reaches 92.5 EFLOPS, China Telecom's self - owned and connected intelligent computing reaches 91 EFLOPS, and China Unicom reaches 45 EFLOPS. The three operators together account for nearly 50% of the IDC market share. In 2025, China Mobile's computing power service revenue was 89.8 billion yuan, a year - on - year increase of 11.1%, accounting for 20.2% of its main business revenue. China Mobile has officially listed "computing power service" and "communication service" as two of its three major main businesses.

Second, they have channels that others don't have. A mobile phone number, bound to a phone bill, is a payment system covering 1.7 billion users. For any large - model manufacturer, this is a much - desired distribution channel. Users don't need to register a cloud platform account, bind a credit card, or understand what an API is. They can simply open the operator's App, select a package, pay, and use AI. This "low threshold" is something that even large Internet companies can't achieve with huge investments.

Third, they have the network. AI inference has much higher requirements for latency than ordinary cloud computing. The closer a company's computer room is to users, the better the experience. The tens of thousands of edge computer rooms of operators across the country are a natural advantage. According to a research report from Galaxy Securities, relying on the national backbone network, distributed computing power nodes, and cloud - network integration capabilities, operators are transforming from "data traffic pipeline providers" to "computing power service providers".

However, on the other hand, compared with large Internet companies, operators also have inherent shortcomings: they are far from users' application scenarios. Some professionals point out that when targeting the individual market, operators have difficulty competing flexibly with Internet companies; but when targeting government and enterprise customers, especially sensitive departments and enterprises, operators' advantages in security, controllability, algorithm response, and tool integration will be evident.

So, if you look at the current product layouts of the three operators, they all show a tendency to "focus on the B - end".

Pan Helin, an expert from the Ministry of Industry and Information Technology, put it more specifically: Government and enterprise customers, especially state - owned enterprises, have comprehensive requirements for security, controllability, algorithm response, and tool integration, and operators have advantages in these aspects.

The new computing power operation model of "Agent + Token + AI Cloud" proposed by China Unicom, the construction of China Telecom's intelligent cloud system, and China Mobile's strategic path of upgrading to a technology service enterprise all point to the same thing: not competing for the C - end AI application market, but serving as the infrastructure for the entire AI industry.

The Entire Industry is Repricing Tokens

Operators are not the only ones eyeing the Token business. Expanding our view, the entire AI industry chain recalibrated its relationship with Tokens in the first half of 2026.

Alibaba established the Alibaba Token Hub business group this year, with CEO Wu Yongming taking the lead. It integrated Tongyi Laboratory, MaaS business line, Qianwen Division, etc. The core goal can be summarized in nine words: "Create Tokens, Deliver Tokens, Apply Tokens". And Jack Ma led the core management teams of Alibaba and Ant to Yungu School for a closed - door seminar on the challenges and opportunities brought by AI, and then began to appear frequently in various parks. It's hard not to connect these signals.

Zhipu raised the API price in the first quarter of this year, but the call volume increased instead. This phenomenon is "extremely rare". When the industry is generally engaged in a price war, someone can raise the price and increase the volume, indicating that in high - value scenarios, pricing power has begun to concentrate.

ByteDance's Volcengine announced the commercial pricing of the Seedance 2.0 video generation model in March this year. The cost of generating a 15 - second video is about 15 yuan. The official has said that enterprise customers are now more concerned about the "overall cost of completing things end - to - end" rather than the cost per Token.

Looking at the Token business of operators together, Alibaba wants to control the entire Token industry chain, ByteDance wants to monetize C - end traffic, and operators want to turn computing power into new revenue. They have different starting points and paths, but their destinations all revolve around one word, Token.

At this stage, the cooperation between operators, large - model manufacturers, and cloud service providers far outweighs the competition.

Operators integrating AI manufacturers' models into their package systems is equivalent to helping AI manufacturers with large - scale distribution. For users, the most direct feeling is that "using AI has become easier". They don't need to research which platform is cheaper, which model is stronger, or how Tokens are billed. You buy a package, and the operator takes care of everything behind the scenes.

However, whether this model can work in the long run depends on how to solve several problems. An expert in the industry pointed out a reality: The Token charging mechanism is still like a "black box" at present. Users lack a transparent perception of the consumption of each call. Clarification and cost - reduction are the keys to the growth of the service.

In simple terms, there is no unified standard for the measurement, pricing, and circulation of Tokens. The pricing systems of the three operators are different, and the definitions of Tokens by large - model manufacturers and cloud providers are not entirely the same. When users buy a package and use it on different models, the consumption speed may vary greatly. The standard - setting work being promoted by the China Academy of Information and Communications Technology may be the first step in solving this problem, but it's still a long way from "unification".

In addition, the downstream applications in the industry are far from mature. At present, most users' demand for Tokens is not comparable to their demand for text messages and data traffic in the past. Text messages and data traffic are necessities. Is Token a necessity? It's not clear yet.

In the longer term, whether the operators' Token packages can succeed depends on several variables:

One is whether AI applications can penetrate into the daily lives of ordinary users, just like mobile payment and short - video did in the past. If Tokens are only consumables for developers and heavy users, the market ceiling will be very low.

The second is whether the Token billing system can be transparent and stable, so that users can truly develop the consumption habit of "paying as much as they use". The third is whether operators can continuously invest in computing power infrastructure, rather than regarding the Token package as a one - time pricing innovation.

Regardless of the result, the significance of the operators' move may lie in declaring the positioning of "computing power" as a basic service. This is a more profound story than the commercialization of 5G. And when the three major operators simultaneously put Tokens on the shelves, it shows that this matter has changed from the strategy of a single large company to the consensus of the entire industry.

The communication industry is moving from transporting information to producing intelligence. For us users, when new things come, we should first observe, then try, and finally decide whether to pay. Don't be in a hurry. Let the bullet fly for a while.

This article is from the WeChat official account "New Vision" (ID: xinmouls), written by Lu Yao and republished by 36Kr with permission.