Last month, China Telecom's Ningxia branch signed a 16.45 billion yuan procurement contract — and for the first time, the word "Token" was written into the name of a ten-billion-level deal: a "Token Factory" intelligent-computing project. Days later, Shanghai Telecom launched a 9.9 yuan/month consumer token plan. China Mobile and China Unicom followed within weeks.
The carriers that sold bandwidth for two decades are making a collective pivot — from "pipefitters who transmit data" to "actuaries of intelligence." This isn't a side business experiment. It's a structural signal: Token is becoming metered infrastructure, and the people who own the meter are positioning themselves to own the AI era's utility bill.
The number behind the pivot: 1,000× in two years
Token is the minimum semantic unit of LLM processing. Every AI chat, every doc rewrite, every code completion burns tokens. In early 2024, China's daily token calls stood at roughly 100 billion. By March 2026, that number reached 140 trillion — three orders of magnitude in two years. Not linear growth; exponential.
An industrial age measured productivity in kilowatt-hours. The information age measured it in gigabytes. The AI age measures it in tokens. Tokens are now retracing the exact path cellular data took twenty years ago: from a scarce lab parameter to a standardizable, packagable, mass-distributed commodity. The 9.9 yuan plan is not a promotion — it is a declaration that token now has the properties of water, electricity, and gas, and can be sold to every ordinary consumer.
Why the pipe players are running scared
For two decades, Chinese carriers were trapped in a pipeline paradox: the more data they carried, the thinner their revenue per user became. But token threatens something worse than thin margins — it redefines what "connection" is worth. Traditional billing meters data transported; token billing meters intelligence processed. When human-AI interaction happens at the model layer, token isn't a substitute for traffic — it's a dimension-reduction attack on it. If carriers stay pipefitters, the platform giants will simply route around them.
So all three are trying different escapes:
- China Telecom is the most aggressive. A ten-billion-level bulk purchase plus the 9.9 yuan consumer plan plus a "Tianyi Token Coin" — an attempt to define the industry's metrology standard. Logic: set the standard first, set the rules second, set the price third. Whoever fixes the pricing unit of token holds the AI era's electricity bill.
- China Mobile is going the platform route. Compute infrastructure plus ecosystem alliances with Tencent, Alibaba, and Huawei for distribution. Pragmatic: not the coolest, but the steadiest — betting on the country's largest base-station network and a billing relationship with over a billion users.
- China Unicom is differentiating. Bundling tokens with AI cloud desktops and agents, targeting lightweight B-segment buyers like "one-person companies." Not competing on scale; competing on penetration.
The chain has already split into three layers
While carriers search for their entry point, the token value chain has quietly stratified — and the bloodiest fight is happening in the layer nobody looks at.
- Production layer: Huawei Ascend and its peers build the compute clusters that physically generate tokens. Upstream "mine owners." High technical barriers, few players — a real hardcore arena.
- Distribution layer: Third-party platforms like SiliconFlow package cluster tokens into APIs for developers. This is the brutal one: SiliconFlow's 2025 annual report showed gross margin as low as -119% — trading capital for scale, subsidies for users. Pure token distribution has no network effect and no differentiation. Token is a logical commodity, not a physical monopoly; its price is transparent as glass. Sell at one cent today, someone undersells you tomorrow. The only exits are integrating upstream (build your own clusters) or penetrating downstream (own an industry solution). Stuck in the middle means being the first squeezed out.
- Application layer: Alibaba, ByteDance, and Tencent hold this ground. Doubao hides tokens inside a chat window, Tongyi Qianwen buries them in a document assistant, Hunyuan embeds them in code completion. Users never feel the token; the giants quietly collect the premium. High-dimensional players never sell naked tokens — they turn tokens into air the user can't live without.
Carriers, for their part, hold something no cloud vendor can copy: a nationwide physical network and the billing front door to more than a billion users. That is their irreplaceable card.
Endgame: three business models, not three companies
So who are the "new three operators"? The answer is already visible: not three independent companies, but three irreplaceable business models.
- The carrier-led infrastructure layer — making token universally accessible through scale and entry advantage. No matter how the application layer evolves, tokens must flow through their pipes. They become the water, electricity, and gas of the AI age.
- The cloud-led enterprise platform layer — selling "token + model + toolchain" as one integrated offer. For enterprise buyers, buying tokens is only the start; converting tokens into productivity is where the real premium lives.
- The third-party application aggregation layer — vertical token platforms for finance, healthcare, legal, education. They don't produce tokens and don't build general models; they machine tokens into plug-and-play modules for specific scenes.
Token, because it is a logical commodity rather than a physical resource, is inherently layered — no single player can hold absolute monopoly. The old forces own the network entry and the billing pipe; the new forces own compute production and application conversion. The real competition is whether each layer can build a moat no one else can cross.
What this means for developers and buyers
For developers, the core question is not "whose token is twenty cents cheaper." Token prices will keep falling — possibly toward zero — as price wars continue. The question is whether your application can keep running on a stable, high-concurrency, low-latency base. "Runs reliably" will always beat "buys cheaply" by a hundred times.
For industry observers, the endgame is not a zero-sum tripod but a layered symbiosis: each layer defends its own territory and builds its own wall. The pipefitter era is over. The actuary era has begun — and the players who own the meter are already taking their positions.