The 2026 Mobile Network Conference was held in Guangzhou from November 5 to 7. In his keynote, “Strengthening Network Foundations, Co-Creating Intelligence,” on the morning of November 6, China Mobile Chairman Chen Zhongyue laid out three figures: more than 50 billion yuan in total procurement in 2027, over 10 billion yuan in computing power investment, and more than 10 billion yuan in device procurement and sales.

China Mobile discloses its procurement budget every year, but how it divides that spending reveals where it sees the bottleneck. This time, the allocation is clear: more money is going into computing power, and most of that 10 billion yuan will not be spent on constructing buildings.

Computing Power Is the Budget Still Growing

Viewed together, computing power and device procurement and sales each account for 10 billion yuan, putting them on a similar scale. But Chen framed them differently.

Device procurement and sales are tied to an established business: the 2027 target is to “sell more than 100 million units,” driven by distribution channels and replacement demand. Computing power, by contrast, was described more emphatically: “China Mobile’s largest increase in investment this year is in computing power.”

Several initiatives show where that increase is going. Having already built multiple AI computing clusters equipped with tens of thousands of accelerators, China Mobile is now planning several ultra-large AI computing clusters. It aims to double its AI computing capacity successively over the next two years and build multiple gigawatt-scale AI computing campuses.

Both “successive doubling” and “doubling” appeared in the speech. The first suggests capacity will more than double once within two years, while the second refers to ensuring that AI computing capacity doubles by 2027. The two formulations do not align perfectly, but they point in the same direction: the computing power supply curve will rise steeply over the next two years.

The Bottleneck Is Not Data Center Capacity

The verbs used in the section on the computing power network reveal the company’s real assessment.

Connecting computing resources into a network, fully linking China’s national computing hubs, expanding 400G OTN capacity, increasing the use of hollow-core fiber, accelerating CDN expansion and upgrades, upgrading the computing-network control platform, and enabling computing resources to connect to the grid and be scheduled as one: none of these initiatives involves building a larger data center. They all point to the same objective: linking existing computing power so it can be scheduled and delivered where needed.

Chen described these initiatives with three goals: smoother access to computing power, more efficient interconnection between computing resources, and more agile edge distribution. “Smoother” refers to traffic coming in, “efficient” to movement within the network, and “agile” to delivery out of it. Together, the three terms cover the three bottlenecks along the chain.

If China Mobile believed there was simply not enough computing power, it would keep adding racks. Its decision to build the roads instead suggests it sees transmission and scheduling, rather than total supply, as the current constraint.

Tokens Become the New Unit of Measurement

One phrase in the speech sounded unusual for a telecom operator: “token factory.”

The original wording was: “Jointly build dedicated computing power networks to support the deployment of customized token factories.” The same section also said the computing power network would provide “stronger support for token invocation.”

Changing the unit of measurement is more than a rhetorical shift. When selling racks, the billable resources are space and electricity. When selling bandwidth, the billable resource is traffic. When selling tokens, billing is tied to the model’s actual output. Each model involves entirely different customers, contract cycles, and margin structures. The first is a one-time delivery; the last is settled by consumption and places a much heavier operational burden on the provider.

The supporting initiatives point in the same direction: launching a full-stack AIDC solution and working with partners to build customized gigawatt-scale AI computing campuses; applying integrated generation-grid-load-storage systems, direct connections to green power, and shared infrastructure for air and liquid cooling to coordinate computing with electricity and water use; and upgrading China Mobile Cloud, expanding benchmark government and industrial cloud projects, and supporting model training and inference, customized computing power, and AI agent development.

Once a company starts selling outcomes, it must take responsibility for those outcomes. That is something telecom operators have not done in decades.

The Intelligent Network Wants to Be the Switch and Router

Of the three layers, the third bears the least resemblance to a traditional telecom business. China Mobile is positioning itself as the “switch” and “router” of intelligent services.

The analogy deserves a closer look. A switch sends packets to the correct port, while a router selects the path. China Mobile does not want to produce models itself. It wants to enable models, data, and AI agents to be supplied securely and reliably and interconnected everywhere within its network, supporting end-to-end services “from scenarios to platforms, devices to the cloud, and applications to models.”

Put differently, the communications network sells connectivity, the computing power network sells computing power, and the intelligent network sells control over scheduling. The first two are resource businesses; the third is a position business. Whoever occupies the middle decides what can be invoked and at what priority.

The speech did not specify how this layer would be priced. It was described more as infrastructure than as a product that can be billed.

A Needle Built to Its Own Standard

The three networks need a common foundation to stitch them together. China Mobile has chosen a technology whose definition it led.

China Mobile describes the IPv6 Enhanced full-stack universal routing protocol as a “foundational technology for the global internet.” It has been deployed at scale across the networks of more than 300 operators in China and overseas, improving internet transmission efficiency by more than 25%.

Chen said the aim of using this technology across the board to build a new routing architecture is to “better align the communications network with AI computing, strengthen the computing power network’s supply of tokens, and enable the intelligent network to better meet service demand.”

That statement clarifies the roles of the three networks: the communications network directs the flow of computing power, the computing power network supplies tokens, and the intelligent network connects with demand. The routing protocol is the shared language of all three. Developing that language in-house and getting more than 300 operators to adopt it creates something far harder to replace than a handful of new campuses.

The Rest Goes to Devices and Distribution

Returning to the 50 billion yuan total, more than 10 billion yuan will go to device procurement and sales. Together with investment in distribution resources and the renovation of 10,000 stores, that represents the other major portion of spending.

This portion corresponds to China Mobile’s most tangible assets: more than 1 billion mobile subscribers and over 300 million household customers. The accompanying initiatives in the speech were a refresh of the GoTone service and the 12580 intelligent assistant, intended to give every customer an AI assistant they can carry with them.

China Mobile also wants to turn these customers into an entry point for its ecosystem. On the capital side, it plans to use its “Rainbow Bridge” mechanism to coordinate strategic investment with industrial development and increase capital partnerships in five areas: mobile networks, chips, models, quantum technology, and embodied AI. On the platform side, it will launch the China Mobile Ecosystem Aggregation Platform as a unified gateway for global partners, matching business opportunities with products, technologies with scenarios, demand with delivery, and industry with investment.

The inclusion of embodied AI in the capital partnership list is an easily overlooked detail of the speech. It does not appear in the technical discussion of the three networks, but it does appear among the company’s investment priorities.

Viewed layer by layer, the communications network manages connectivity, the computing power network manages supply, and the intelligent network manages scheduling. The 50 billion yuan will build infrastructure across all three. The genuinely new element is the unit of measurement in the middle: once bandwidth gives way to tokens, what telecom operators sell becomes directly tied to model output for the first time.