A New Commodity.

May 17th, World Telecommunication Day.

China Telecom made a move on this day, launching a trial nationwide commercial Token package. Individual users can get 10 million Tokens for as low as 9.9 yuan. Shanghai Mobile offers 400,000 Tokens for 1 yuan, while Beijing Mobile provides a monthly package of 10 million Tokens for 24.99 yuan. China Unicom is also active, with Shanghai offering 30 million free Tokens to OPC customers for testing, and Hubei Unicom bundling Tokens with telecom and broadband services.

As of today, all three operators have launched Token packages. This marks the introduction of a new billing category on Chinese communication bills, following phone charges, text message fees, and data fees.

For someone who has worked in the telecommunications industry for 26 years, this signifies far more than just an innovation in billing. The last time operators collectively adjusted their billing units was during the 4G rollout, shifting from call minutes and text message counts to data MBs, and now to Tokens. However, Tokens are different. Minutes sold connectivity, data sold bandwidth, but Tokens sell 'intelligence.'

As computing power is bundled into packages and billed monthly like utilities, we are witnessing a transformation at the infrastructure level.

01

Why Token Has Become a Hard Currency

Let's examine the packages offered by the three operators.

China Telecom has made the boldest move, with unified pricing at the group level covering the entire country. Both individuals and developers, as well as enterprises, have three tiers to choose from. China Telecom's packages integrate its own Stellar Large Model and ecosystem models like DeepSeek V3.2, suitable for daily office assistance, copywriting, learning, and creative tasks.

China Mobile adopts a more flexible approach, advancing province by province. Shanghai Mobile is the most aggressive, offering 400,000 Tokens for 1 yuan, supporting payment via phone bills and cross-platform use, and partnering with Tencent to launch an AI-native workstation. Beijing Mobile offers a monthly package of 10 million Tokens for 24.99 yuan, with additional discounts for existing customers, and a pay-as-you-go option starting at 5.99 yuan. Jiangsu Mobile even offers an introductory package of 2.5 million Tokens for 5 yuan, setting the barrier for trial extremely low.

China Unicom takes a different path. Hubei Unicom launched three tiers of Token Plans and two tiers of Coding Plans in late April, integrating telecom, broadband, and cloud desktop services. Shanghai Unicom directly targets one-person companies, offering 30 million Tokens for free during the testing period. Sichuan Unicom introduced a 'Family Token Package,' combining a fixed monthly fee with Unicom's cloud desktop and WorkBuddy AI agent, aiming to address the pain point of 'not knowing how to use' with 'out-of-the-box' solutions.

A simple comparison: China Telecom's lowest individual tier costs 0.99 yuan per million Tokens, likely the cheapest among the group's unified pricing; Shanghai Mobile offers 400,000 Tokens for 1 yuan, equating to 2.5 yuan per million Tokens; Shanghai Unicom's OPC customer renewal price is 1 yuan per million Tokens.

In terms of unit price, China Telecom is the most aggressive, but China Mobile excels in flexibility, allowing pay-as-you-go without monthly commitments. China Unicom differentiates itself through scenario integration, bundling Tokens with cloud desktops and AI agents, selling 'out-of-the-box' solutions.

However, viewing this as merely a price war would be shortsighted. At this stage, pricing resembles the early days of data traffic management, aiming to 'cultivate user habits.'

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

Strictly speaking, a Token is the smallest computational unit for large models to process text, images, and voice. When you ask, 'How's the weather in Beijing today?' AI breaks it down into five or six Tokens for processing; every word in its response is also a Token.

Over the past two years, Token circulation on the Chinese internet has experienced exponential growth in a literal sense. According to the National Data Bureau, China's daily average Token usage was 100 billion in early 2024; by the end of 2025, it surged to 100 trillion; and in March 2026, it exceeded 140 trillion. A thousandfold increase in two years.

In March this year, the National Committee for Terms in Sciences and Technologies officially designated 'Token' as the standard Chinese translation. When a technical term requires a national-level announcement for naming, it indicates that it has evolved from an internal billing unit among large model firms to a society-wide consumable. Not data traffic, not bandwidth, but a brand new (brand-new) 'basic resource.'

Jensen Huang clarified this at the GTC conference, stating, 'Token is the new commodity.' The term 'commodity' is not used lightly. It has specific connotations: standardized, gradable, and tradable at scale.

At GTC, NVIDIA introduced a five-tier Token pricing system, ranging from a free tier to an ultra-high-speed tier at $150 per million Tokens, graded by speed and context length. This resembles what the telecom industry has long attempted but failed to achieve—differentiated pricing for data traffic based on quality. Huang has brought this concept to fruition.

Also in March, China's large model industry underwent a violent (drastic) price revaluation. Tencent Cloud led by raising prices for its Hunyuan Large Model, with the core model's input price increasing from 0.0008 yuan per thousand Tokens to 0.004505 yuan, a 463% hike. Alibaba Cloud and Baidu Intelligent Cloud followed suit.

By May, pressure mounted on consumer-end services. ByteDance's Doubao quietly listed three paid subscription tiers on the App Store. By then, Doubao's daily Token usage had surpassed 120 trillion, a thousandfold increase since its launch. Zheshang Securities estimates that ByteDance's 2025 capital expenditures would reach approximately 160 billion yuan, with most allocated to AI computing power procurement.

Open-commercialized DeepSeek took a different approach, pricing V4-Flash inputs at 1 yuan per million Tokens, and as low as 0.2 yuan after cache hits, nearly at cost. Despite this, the industry's pricing curve has clearly trended upward.

In other words, just as the shift from 'all-you-can-eat' to 'pay-as-you-go' reaches a tipping point, operators have entered the fray with phone bills.

02

Operators' Calculations

The simultaneous launch of Token packages by the three major operators is no coincidence.

Traditional telecom services have hit a ceiling. Data unit prices continue to decline, while voice and text message services have atrophy ed (shrunk) for years. Operators have long faced the anxiety of being reduced to mere 'pipes'—building networks and transmitting data, while internet firms reap the profits.

Tokens offer them a chance to turn over (turn the tables).

First, they possess ready-made computing power infrastructure. China Mobile's total intelligent computing capacity reaches 92.5 EFLOPS, China Telecom's proprietary and accessed intelligent computing stands at 91 EFLOPS, and China Unicom's at 45 EFLOPS. Together, they command nearly 50% of the IDC market share. By 2025, China Mobile's computing power service revenue will reach 89.8 billion yuan, up 11.1% year-on-year, accounting for 20.2% of its main business revenue. Mobile has officially listed 'computing power services' alongside 'communication services' as two of its three primary businesses.

Second, they have unparalleled distribution channels. A mobile number linked to a phone bill represents a payment system covering 1.7 billion users. For any large model firm, this is a highly desirable distribution channel. Users need not register cloud platform accounts, bind credit cards, or understand APIs. They simply open the operator's app, select a package, pay, and start using AI. This 'low barrier to entry' is something internet giants cannot replicate, no matter how much they invest.

Third, they control the network. AI inference demands far lower latency than ordinary cloud computing, favoring providers with data centers closer to users. Operators' tens of thousands of edge data centers nationwide represent a natural advantage. Galaxy Securities notes that operators are transitioning from 'traffic pipe providers' to 'computing power service providers,' leveraging their national backbone networks, distributed computing nodes, and cloud-network integration capabilities.

However, operators also face inherent weaknesses compared to internet giants: they are distant from user application scenarios. Experts point out that while operators struggle to compete flexibly with internet firms in the consumer market, they hold advantages in security, controllability, algorithm responsiveness, and tool integration for government and enterprise clients, especially sensitive departments.

Thus, the current product layouts of the three operators reflect a 'B-end focus.'

Pan Helin, an expert from the Ministry of Industry and Information Technology, elaborates: Government and enterprise clients, particularly state-owned enterprises, have comprehensive demands for security, controllability, algorithm responsiveness, and tool integration—areas where operators excel.

China Unicom's proposed new computing power management model of 'Agent + Token + AI Cloud,' China Telecom's intelligent cloud system construction, and China Mobile's strategic path toward upgrading into a technology service enterprise all point to the same goal: not to compete for C-end AI application territory but to serve as infrastructure for the entire AI industry.

03

The Entire Industry is Repricing Tokens

Operators are not the only ones eyeing the Token business. Broadening the perspective (perspective), the entire AI supply 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 personally leading it, integrating the Tongyi Laboratory, MaaS business line, and Qianwen Business Division. The core objective is nine words: 'create Tokens, deliver Tokens, apply Tokens.' Meanwhile, Jack Ma led Alibaba and Ant Group's core management to Yungu School for closed-door discussions on AI's challenges and opportunities, after which he began frequently visiting various campuses. These signals are hard to ignore when viewed together.

Zhipu AI raised its API prices in Q1 this year, yet usage volume expanded further—a 'highly unusual' phenomenon in an industry embroiled in price wars. The ability to raise prices and still grow indicates that pricing power is concentrating in high-value scenarios.

ByteDance's Volcano Engine announced commercial pricing for its Seedance 2.0 video generation model in March, with each 15-second video costing approximately 15 yuan. Officials stated that enterprise clients now care less about per-Token costs and more about the 'overall cost of getting things done end-to-end.'

Viewed alongside operators' Token ventures, Alibaba aims to dominate the entire Token supply chain, ByteDance seeks to monetize C-end traffic, and operators aim to transform computing power into new revenue. Their starting points and paths differ, but their endpoints converge on one word: Token.

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

Operators integrating AI firms' models into their package systems effectively distribute these models on a large scale. For users, the most immediate benefit is that 'using AI becomes simpler'—no need to research which platform is cheaper, which model is stronger, or how Tokens are billed. Users buy a package, and operators handle everything behind the scenes.

However, the long-term viability of this model hinges on resolving several issues. Experts point out a practical reality: Token billing mechanisms remain a 'black box,' with users lacking transparent awareness of consumption per call. Clarity and cost reduction are key to service growth.

Simply put, no unified standards exist for Token measurement, pricing, or circulation. The three operators' pricing systems differ, and large model firms and cloud providers define Tokens inconsistently. A user purchasing a package may experience vastly different consumption rates across models. The China Academy of Information and Communications Technology is advancing standardization efforts, but 'unification' remains distant.

Additionally, downstream applications are far from mature. Most users' current demand for Tokens bears little resemblance to their past demand for text messages and data. While text messages and data were essential, whether Tokens are essential remains unclear.

From a longer-term perspective, the success of operators' Token packages depends on several variables:

First, whether AI applications can penetrate ordinary users' daily lives—like mobile payments and short videos did. If Tokens remain consumables for developers and power users, their market ceiling will be low.

Second, whether Token billing systems can become transparent and stable, enabling users to truly adopt 'pay-for-what-you-use' habits.

Third, whether operators can sustain investments in computing power infrastructure, rather than treating Token packages as one-time billing innovations.

Regardless of the outcome, operators' move may signify the positioning of 'computing power' as a fundamental service—a story more profound than 5G commercialization. With all three operators now offering Tokens, this indicates that what was once a strategy of individual firms has become an industry consensus.

The telecommunications industry is transitioning from transporting information to producing intelligence. For users, new offerings have arrived—observe first, try next, then decide whether to pay. No need to rush; let the dust settle.

— END —