In the era of traffic, operators were pipeline workers. In the Token era, whoever controls the pricing unit holds the 'electricity bill' of the AI era. Behind the 140 trillion daily average calls lies a new future of layered coexistence—not three companies, but three models.

Original New Entropy AI New Tech Team

In mid-2026, China Telecom's Ningxia branch signed a contract worth 16.451 billion yuan for a 'Token factory' project. This marked the first time the three major operators included 'Token' in the name of a hundred-billion-level procurement project.

Soon after, Shanghai Telecom launched a 9.9 yuan Token package for consumers, with China Mobile and China Unicom quickly following suit.

These veterans, who have been selling broadband and traffic for two decades, are shifting their Token layout (Token layout can be translated as 'Token strategy' or 'Token deployment' depending on context; here, 'strategy' is used for smoother readability) from tentative moves to in-depth deployment.

The reason is simple: operators refuse to accept a slow decline. They aim to stage a life-and-death breakthrough in the AI era—transforming from 'data pipeline workers' to 'actuaries of intelligence'.

Token Infrastructure and Operators' Collective Shift

What is a Token? It is the smallest semantic unit for large models to process information.

Every time you ask an AI a question, have it write copy, or revise a weekly report, Tokens are consumed. In the industrial era, productivity was measured in kilowatt-hours; in the information era, connection efficiency in GBs; in the AI era, intelligence density in Tokens.

What is truly alarming is the speed of transition.

In early 2024, the domestic daily average Token calls were about 100 billion; by March 2026, they surged to 140 trillion. Two years, three orders of magnitude. This is not linear growth but exponential explosion.

Tokens are racing along the same path as cellular data traffic did two decades ago: from scarce technical parameters in labs to standardized, packageable, and mass-distributable basic resource commodities. China Telecom's 9.9 yuan Token package is not a promotion but a clear signal—Tokens now possess the attributes of utilities like water, electricity, and gas, and can be sold to every ordinary person like traffic.

But Tokens mean far more to operators than just selling an additional service. They are redefining the value of 'connection' itself.

Traditional traffic billing measures data transmission volume, while Token billing measures intelligent processing volume. When intelligent interactions occur at the large model level, Tokens are not a substitute for traffic but a dimensionality reduction attack on it. If operators cannot transform from 'data pipeline workers' to 'actuaries of intelligence', they will be kicked aside by new era giants.

Over the past two decades, the three major operators have been trapped in the pipeline dilemma of increasing traffic but declining ARPU. The more traffic they sold, the thinner their revenues became. In the AI era, if user interactions with agents bypass traditional apps and traffic consumption, the 'connection' value of operators will be completely undermined.

Thus, all three have embarked on difficult breakthroughs.

China Telecom is the most aggressive: with hundred-billion-level procurement, 9.9 yuan consumer packages, and 'Tianyi Token Coins', it aims to establish industry standards for billing and circulation systems. The logic is clear—set standards first, then rules, and finally prices. Whoever controls the Token pricing unit holds the 'electricity bill' of the AI era.

China Mobile chooses a platform-based path: relying on computing power infrastructure and ecological alliances, it partners with Tencent, Alibaba, Huawei, and others to build distribution capabilities. The approach is pragmatic—not vying for the coolest but the most stable. It uses the nation's largest base station network and billing access for billions of users as leverage.

China Unicom takes a differentiated route: bundling Tokens with AI cloud desktops and agents, it offers solutions for lightweight B-end scenarios like 'one-person companies'. It does not compete on scale but on penetration.

The three paths differ, but the underlying logic is surprisingly consistent: never be a pipeline worker again. National-level network infrastructure and billing access for billions of users are their shared core assets—a moat that no cloud vendor or third-party platform can easily replicate.

Three-Tier Division of New Forces and the Distribution Layer Dilemma

While operators are still seeking paths at the entry layer, the industrial chain has quietly divided into three layers. The fiercest battles are fought where you cannot see them.

The production layer is the domain of Huawei Ascend and others.

They build computing power clusters, responsible for the physical generation of Tokens, and are the 'upstream miners' of the entire chain. No computing power means no Tokens; no chips mean no computing power. This layer has extremely high technical barriers, with few players, and is a true 'hardcore arena'. Whoever masters advanced-process AI chips and ultra-large-scale computing power clusters holds the 'faucet' of Token production. At this level, the story of domestic substitution is unfolding fiercely, but a visible gap remains compared to international top levels.

The distribution layer is the battleground for third-party platforms.

Companies like Silicon-Based Flow package Tokens produced by computing power clusters into APIs and sell them to developers and enterprises. However, this layer's dilemma is particularly prominent—Silicon-Based Flow's 2025 annual report showed a gross margin as low as -119%, essentially exchanging capital for scale and subsidies for users. Pure Token distribution lacks network effects and differentiation barriers. With computing power costs rigidly declining, the profit models of independent distributors remain under sustained pressure.

Why is the distribution layer so brutal?

Because Tokens are logical commodities, not physical monopolies. They have no scarcity barriers; anyone can buy or sell them, with prices as transparent as glass. If you sell for one cent today, someone will sell for half a cent tomorrow. Distribution layer players can only engage in an endless war of attrition—either being crushed by giants or dragged down by costs.

The application layer is the domain of internet newcomers like Alibaba, ByteDance, and Tencent.

They do not care about the price of Tokens per ten million but whether users are willing to pay for AI features. Doubao hides Tokens in chat boxes, letting you generate copy while chatting; Tongyi Qianwen embeds them in document assistants, automating weekly reports and meeting minutes; Hunyuan integrates them into code completion, letting programmers write half as much repetitive code. Users consume Tokens unknowingly, while giants quietly collect revenues.

Do you see the difference?

One type of player burns money to gain market share in the distribution layer, trying to outpace rivals; the other, represented by Alibaba and ByteDance, swallows Tokens into products, using application layer premiums to cover costs and creating their own closed loops. True high-dimensional players never sell naked Tokens—they turn Tokens into the air users cannot live without.

The distribution layer's dilemma will not disappear. As OpenAI and domestic large models continue to wage price wars, Tokens are becoming increasingly cheap. When Tokens drop to a few cents per ten million, players' profit margins will be completely flattened.

The deeper issue is that the distribution layer lacks irreplaceable moats. The computing power layer has chip barriers; the application layer has user stickiness. What does the distribution layer have? An API interface? A billing system? These are not core competencies that cannot be replicated.

This means the distribution layer's ultimate paths are only two: either extend upstream, build its own computing power clusters, and become a vertical integrator of 'production + distribution'; or penetrate downstream, offer industry solutions, and become a scenario Deep cultivator (scenario Deep cultivator can be translated as 'scenario deep-diver' or 'scenario specialist'; here, 'specialist' is used for clarity) of 'distribution + application'. Pure distributors stuck in the middle are destined to be the first to exit the large model industrial chain.

Final Judgment: Three Models, Not Three Companies

After all this, who will be the 'new three major operators' of the future?

The answer is already emerging—not three independent companies but three irreplaceable business models.

The first is operator-led basic pipeline layers, making Tokens affordable through scale effects and entry advantages. They hold the national network and billing access for billions of users, packagizing Tokens into universal services, becoming the 'water, electricity, and gas' of the AI era. No matter how upper-layer applications change, Tokens must flow through their pipelines. This is their irreplaceable trump card.

The second is cloud giants-led enterprise platform layers, enabling enterprises to 'use effectively' through vertical integration. Alibaba, Tencent, Huawei, and others do not sell naked Tokens but one-stop solutions of 'Token + model + toolchain'. For enterprise clients, buying Tokens is just the beginning; how to turn Tokens into productivity is where cloud giants truly add value.

The third is third-party and developer-led application aggregation layers, enabling developers to 'use smoothly' through scenario innovation. In the future, there will be a batch of Token aggregation platforms deep cultivation ( deep cultivation can be translated as 'deeply cultivating' or 'specializing in'; here, 'specializing in' is used) specific industries—finance, healthcare, law, education versions. They do not produce Tokens or build general-purpose large models but process Tokens into 'plug-and-play' modules for vertical scenarios.

As logical commodities rather than physical resources, Tokens inherently possess layered attributes, with no single entity achieving absolute monopoly. Old forces control network entries and billing pipelines; new forces control computing power production and application transformation. The true game lies in whether each party can build irreplaceable moats within its own layer.

For developers, the core proposition is never 'which Token is two mao cheaper' but whether applications can run continuously in a stable, high-concurrency, low-latency environment. Tokens will become increasingly cheap, even approaching zero, but 'running stably' will always be a hundred times more important than 'buying cheaply'.

For the industry, the final outcome is never a zero-sum, three-legged standoff but a coexistence pattern ( pattern can be translated as 'landscape' or 'structure'; here, 'landscape' is used) of multi-layered architectures—each layer guards its position and builds its barriers. Each layer requires different capabilities, and each layer may spawn new giants.

Behind the 140 trillion daily average calls lies not just a number's surge.

It is a footnote to the entire AI era—when Tokens become as ubiquitous as traffic, a new era where intelligence measures everything has quietly begun.

The era of pipeline workers is over. The era of actuaries has begun.

References:

Silicon-Based Flow 2025 Annual Report: Gross Margin -119%

MIIT/AI Association Joint Release in March 2026: Domestic Daily Average Token Calls at 140 Trillion

Public Data from Vendors: China Telecom's 'Tianyi Token Coins', China Mobile's 'Computing Power Ecosystem Alliance', China Unicom's 'AI Cloud Desktop + Agent' Solutions

China Telecom Ningxia Branch's July 2026 Bid Announcement: 'Token Factory' Smart Computing Center Procurement Project, Amounting to 16.451 Billion Yuan

Shanghai Telecom's June 2026 Launch of Consumer Token Package (9.9 yuan/month); China Mobile and China Unicom's Follow-Up Reports in July-August 2026

- END -