China Mobile Token Plans Turn AI Compute Into a National Telecom Product

China Mobile plans to introduce nationally standardized AI token packages after testing several service levels across multiple provinces during the first half of 2026. The China Mobile token plans would package model usage with connectivity, devices, customer benefits, and other technology services.

That combination creates a direct challenge to the way cloud companies sell generative AI. Instead of requiring customers to compare separate model APIs, China Mobile wants to make AI consumption resemble a familiar telecom subscription.

The company disclosed the plan during its interim results briefing on August 14, according to a report carried by 36Kr. China Mobile has not yet published the package allowances, launch date, supported models, or service-level terms.

Those missing details matter because a token is not a standardized unit of useful work across different models. One model can consume more tokens than another while completing the same task.

China Mobile is betting that customers will value a unified purchasing channel more than they value direct control over every model and infrastructure choice. Its MoMA platform sits at the center of that strategy.

The contest is therefore not simply China Mobile against another telecom carrier. It is the carrier bundle against the conventional cloud model, where buyers assemble models, computing resources, security, and network services separately.

China Mobile Is Taking Token Packages Nationwide

The announcement moves China Mobile’s token business from scattered trials toward a coordinated national product.

China Mobile said it completed the design of several token package levels during the first half of 2026. Those packages have already entered trials in multiple provinces, although the company did not identify every participating market.

The carrier now intends to establish unified national pricing for both consumer and enterprise customers. It will focus on frequently used AI scenarios and customers that generate higher service value.

This is more than an expansion of geographic availability. National pricing would replace a collection of local experiments with a product that China Mobile can market, provision, and bill across its broader network.

That distinction is important in China’s telecom sector. Provincial subsidiaries often test new services before the parent company standardizes successful offerings for wider distribution.

Earlier regional packages showed that China Mobile was willing to sell model consumption through existing telecom channels. A national structure would give those experiments common commercial rules.

China Mobile says customers will be able to combine tokens, models, mobile data, devices, and additional benefits. They would order the components they need instead of buying a single fixed AI service.

The company has not explained how flexible those combinations will be. It also has not said whether unused allowances will expire, transfer between models, or remain tied to individual accounts.

For consumers, the package could provide access to chatbots, content generation, cloud storage features, or device-based assistants. The carrier could place those services inside applications that customers already use.

Enterprise packages can address different needs. A business might consume tokens through customer-service agents, document analysis, coding assistants, or automated workflows connected to internal systems.

China Mobile has already described retail inspections, ecommerce, cloud storage, and enterprise services as target scenarios. Its May announcement also referenced partnerships with government and corporate customers.

The carrier is building this effort around Token operations rather than a single proprietary model. Here, a token means a unit of text, code, or other input processed or generated by an AI model.

That approach lets China Mobile sell access to multiple models under one commercial framework. It also gives the company room to change the underlying model without redesigning every customer package.

The reported nationwide plan remains an announcement, not a complete product release. Customers still need precise documentation covering quotas, model access, latency, security, and acceptable-use restrictions.

China Mobile must also explain what unified pricing means in practice. A consistent national list does not guarantee identical performance, support, or computing capacity in every province.

Still, the direction is clear. China Mobile wants AI inference to become a recurring carrier service rather than an occasional cloud purchase.

Why the China Mobile Token Plans Are Arriving Now

China Mobile is moving now because it has already assembled the platform, partnerships, and internal organization needed to sell AI consumption at scale.

The company introduced its broader Token operating system at the 2026 Mobile Cloud Conference in May. It described tokens as a common measurement connecting computing power, models, applications, and users.

China Mobile also announced a unified operating layer for authentication, billing, and settlement. These functions are ordinary in telecom services but remain fragmented across many AI platforms.

The carrier’s Token operating system brings those functions into a structure designed for large-scale distribution. It links infrastructure supply with customer-facing applications and commercial settlement.

China Mobile also formed an alliance with major technology partners. The participants named in the announcement included Tencent, Alibaba, Huawei, ZTE, and iFlytek.

A separate application alliance included cloud and model providers such as Alibaba Cloud, Volcano Engine, and Huawei Cloud. China Mobile presented the alliance as a way to standardize measurement, service quality, and compliance.

This partnership strategy reflects an important limitation. Telecom carriers possess networks, billing relationships, data centers, and customer channels, but they do not control every leading AI model.

China Mobile can address that weakness by aggregating outside models. It can then compete through distribution, orchestration, security, and account management rather than model leadership alone.

The company has also created a dedicated Token Office to coordinate commercialization. According to organizational reporting, the office oversees product upgrades, pricing, and compliance.

That internal change suggests the token initiative has moved beyond a technical demonstration. A dedicated group can align provincial trials, infrastructure investment, partner contracts, and national commercial policies.

Traditional telecom economics provide another reason for the timing. Mobile connectivity remains essential, but mature markets offer limited room for rapid growth through ordinary data packages.

China Mobile reported computing-service revenue of RMB 89.8 billion for 2025, an increase of 11.1 percent. Its total operating revenue grew much more slowly during the same period.

The company’s annual results also highlighted stronger AIDC supply and a model aggregation engine. AIDC refers to data centers designed for AI training and inference workloads.

That revenue pattern gives management a direct incentive to connect AI demand with infrastructure investment. Token packages create a retail and enterprise channel for selling the output of that infrastructure.

Demand is also becoming easier to package. AI agents consume tokens repeatedly as they plan tasks, call tools, analyze results, and revise their output.

A single chatbot request may involve one visible exchange. An agentic workflow can trigger many model calls before the user sees a completed result.

That difference supports recurring allowances. Customers may prefer predictable access instead of receiving an uncertain bill for every API request, tool call, and model response.

However, predictable packaging shifts some risk to the service provider. China Mobile must estimate usage patterns, manage capacity, and prevent a small group of heavy users from degrading service.

The company’s timing therefore reflects both opportunity and operational readiness. It has model partnerships, a common platform, billing experience, regional pilots, and a growing computing business.

The remaining step is turning those assets into a national service customers can understand and trust.

MoMA Makes the Carrier Bundle Possible

MoMA is the mechanism that lets China Mobile sell many AI models as one managed service instead of competing through one model.

MoMA is China Mobile’s model management and orchestration platform. It provides a common gateway between customers, applications, outside models, and the carrier’s computing resources.

China Mobile said in May that the platform had gathered more than 300 open and closed models. Named examples included Jiutian, Doubao, DeepSeek, Qwen, and MiniMax.

The MoMA model gateway is designed to reduce the work required to connect every application with each model separately. It can also support common authentication and billing.

In a conventional setup, an enterprise selects a model provider, negotiates access, integrates an API, and builds monitoring around that service. Adding another model can repeat much of that work.

MoMA offers a different route. A customer connects to one platform, while the orchestration layer selects or exposes models behind that interface.

The platform can evaluate a request according to quality, latency, and cost requirements. It can then route the task to a suitable model, according to China Mobile’s description.

This mechanism makes flexible packages technically plausible. A customer allowance does not need to represent access to only one model or one data center.

China Mobile could define a package by application scenario, performance class, model group, or service priority. The customer would buy an outcome-oriented bundle rather than one isolated API.

For example, an enterprise package could send simple classification requests to a smaller model. It could reserve a more capable model for difficult analysis or long-form generation.

A consumer assistant might use one model for conversation and another for image generation. The customer would see a unified allowance even when the platform uses several underlying services.

That abstraction resembles telecom network management. Customers buy connectivity, while the operator handles routing, capacity, interconnection, and many operational details behind the service.

China Mobile wants to extend that role into AI inference. Its advantage is not necessarily a better model; it is the ability to combine model access with networks and computing capacity.

The mobile AI index developed by GTI and Omdia describes MoMA as a gateway between model development and industry applications. The report says it supports task-based model selection.

The same report says China Mobile claims its optimized inference engine can lower token costs by about 30 percent. That figure remains a company claim and requires independent performance testing.

Cost comparisons are difficult because model quality, hardware, prompt caching, response length, and latency can all change the result. A lower unit cost does not automatically mean a better customer outcome.

MoMA also creates strategic leverage over model providers. China Mobile can bring partners access to its customer base, while retaining control over billing and the primary service interface.

That arrangement can benefit smaller model developers that lack nationwide distribution. It can also reduce their visibility if customers experience everything through the carrier’s brand.

Large model providers face a different calculation. They gain another sales channel, but they risk becoming interchangeable suppliers within an orchestration layer.

China Mobile faces dependency risks as well. Its service quality will partly depend on model partners, licensing terms, and the continued availability of outside systems.

A model provider can change usage policies, withdraw a model, or prioritize its own platform. China Mobile needs fallback options that do not disrupt customer workflows.

The company’s own Jiutian models can provide one layer of control. Yet the presence of hundreds of outside models shows that breadth remains central to MoMA’s value.

The national token plan therefore depends on successful orchestration. Without MoMA, China Mobile would be selling a collection of disconnected AI services.

With it, the carrier can present tokens, connectivity, models, and applications as one managed product.

The Real Contest Is Carrier Bundles Versus Direct Cloud Access

China Mobile is testing whether distribution and integration can matter more than direct relationships with cloud and model providers.

Cloud platforms usually sell AI inference through metered APIs. Developers select models, submit requests, and pay according to the provider’s measurement rules.

That model offers transparency and control. Technical teams can compare model behavior, monitor exact usage, and optimize their applications around specific APIs.

It also creates complexity. Enterprises may manage several vendors, separate accounts, inconsistent token definitions, different security controls, and multiple billing systems.

China Mobile’s bundle addresses that friction. It can combine connectivity, inference, devices, support, and identity under an existing customer relationship.

The carrier also has a distribution network that most model startups cannot match. It can place AI packages in mobile applications, retail stores, enterprise sales channels, and account portals.

This advantage is particularly relevant outside dedicated software teams. A small business may want an AI customer-service function without building a cloud architecture or managing several model contracts.

Large enterprises may value unified procurement for different reasons. They often need account controls, regional support, auditability, and predictable service administration.

China Mobile can use its network operations experience to address those requirements. It can also bundle private connectivity or security functions with model access.

The company has promoted confidential inference for organizations handling sensitive data. Confidential computing protects information while a workload is being processed, although implementation quality still requires verification.

Direct cloud access retains important strengths. Developers can adopt new models quickly, configure infrastructure closely, and avoid an intermediary that may limit available features.

A carrier bundle might expose only common capabilities across its model catalog. Advanced provider-specific tools can arrive later or remain unavailable.

Cloud platforms also operate mature developer ecosystems. Their documentation, observability tools, deployment options, and global regions can matter more than consolidated billing.

China Mobile must therefore avoid treating tokens like undifferentiated mobile data. AI model outputs vary greatly in accuracy, reasoning ability, latency, safety behavior, and supported context.

Ten million tokens from one model do not necessarily provide the same value as ten million from another. Model choice can matter more than the nominal allowance.

China’s other major carriers are pursuing similar strategies. China Telecom introduced nationwide token packages before China Mobile’s latest announcement, while China Unicom tested consumer and business offerings.

The early carrier package comparison shows that the entire sector is exploring AI consumption as a telecom product. China Mobile is not entering an empty market.

Competition among the carriers can accelerate adoption. It can also recreate familiar telecom problems, including complicated bundles, promotional allowances, and difficult comparisons.

Cloud providers will not remain passive. They can simplify their own purchasing systems, expand partner channels, and offer enterprise commitments tied to broader cloud usage.

Model companies can also sell subscriptions directly to consumers. Those products provide a clearer connection between the model brand, its capabilities, and the user experience.

China Mobile needs its bundle to provide more than convenience. It must deliver dependable performance and enough choice to justify placing the carrier between customers and model providers.

This contest will develop differently across customer groups. Consumers may respond to simple activation and phone-bill payment, while developers will examine APIs and performance controls.

Enterprise buyers will ask harder questions about data location, audit logs, support obligations, and model changes. Their procurement cycles will also take longer than consumer adoption.

The carrier bundle can win where customers want managed access. Direct cloud services will remain attractive where technical control is the primary requirement.

China Mobile’s national rollout is a test of how large the first group can become.

Unified Pricing Does Not Make AI Tokens Uniform

The largest unresolved issue is whether one token allowance can remain meaningful across hundreds of models and many application types.

Tokenization methods vary. The same sentence can produce different token counts depending on the model, language, tokenizer, and formatting.

Multimodal systems add another layer of complexity. Images, audio, and video may be converted into internal units that do not correspond neatly with text tokens.

Agentic applications amplify the problem. One visible request can cause repeated planning, retrieval, tool execution, verification, and response-generation steps.

Customers can therefore consume allowances at very different rates while performing tasks that appear similar. Without clear measurement rules, a national package may be predictable only on paper.

China Mobile needs to disclose how it normalizes consumption across models. It could apply conversion ratios, separate model pools, or different weights for various workloads.

Each approach has tradeoffs. Conversion ratios simplify billing but can obscure actual provider costs. Separate pools provide clarity but weaken the promise of flexible usage.

The company must also explain model substitutions. If MoMA routes a task automatically, customers should know whether that selection changes token consumption, data handling, or output quality.

Enterprise customers will want controls that prevent an application from switching to an unapproved model. They may also require logs showing which system processed each request.

Service quality creates another uncertainty. Regional trials do not establish that a national product can deliver consistent latency during periods of heavy demand.

China Mobile says it will accelerate AIDC deployment to increase token production. In this context, token production means operating computing infrastructure that generates model outputs at scale.

More infrastructure can increase capacity, but it does not guarantee efficient utilization. AI data centers require accelerators, energy, cooling, networking, software optimization, and skilled operations.

Demand forecasting is difficult during an early market. China Mobile could build capacity ahead of adoption, or it could face congestion if attractive packages stimulate unexpected usage.

The economics also depend on customer behavior. Light users can subsidize heavy users in a fixed allowance structure, but extreme usage can undermine that balance.

Usage controls may protect the service, yet aggressive limits can make a generous package less useful. Buyers need clear policies covering throttling, concurrency, and peak-period access.

Data governance requires equal attention. Consumer prompts may contain personal information, while enterprise requests can include contracts, source code, or internal records.

China Mobile has discussed confidential computing and security partnerships. Customers still need specific commitments on retention, training use, encryption, access controls, and deletion.

The model catalog creates compliance questions. Different models may have separate licenses, permitted uses, moderation systems, and geographic restrictions.

A unified interface cannot erase those differences. MoMA must preserve them through policy enforcement and customer documentation.

There is also a risk of package complexity. Combining tokens, models, data, devices, and benefits sounds flexible, but too many variables can make products difficult to compare.

Telecom customers already understand mobile data reasonably well. AI consumption is less intuitive because users cannot easily predict how many tokens a task will require.

China Mobile can reduce that uncertainty with usage dashboards and scenario estimates. It can show how common tasks affect allowances without promising identical results.

Enterprise buyers need more detailed tools. They should be able to set budgets, assign quotas, monitor individual applications, and detect abnormal consumption.

These capabilities will determine whether the service supports production workloads or remains a promotional add-on.

The national announcement creates a credible commercial direction. It does not yet establish that China Mobile has solved measurement, quality, and governance across its model catalog.

Three Signals Will Show Whether the Strategy Works

The next evidence must come from product terms, real usage, and infrastructure performance rather than another alliance announcement.

The first signal is the final national product specification. China Mobile needs to publish allowances, eligible models, conversion rules, launch coverage, and service limitations.

Those terms will show whether the package delivers genuine model flexibility. They will also reveal whether unified pricing means one national product or several bundles under a common label.

Model-specific weighting deserves close attention. A complicated conversion system would weaken the argument that tokens can function like a simple telecom allowance.

The second signal is adoption beyond promotional trials. China Mobile should eventually disclose active users, recurring customers, enterprise workloads, and consumption patterns.

Registration totals alone will not demonstrate demand. Free allowances and bundled activation can create accounts that generate little sustained usage.

Recurring token consumption would provide stronger evidence. It would show that customers are incorporating MoMA into routine work instead of testing it once.

Enterprise renewals will matter even more. Businesses will continue paying only if the platform reduces operational friction without sacrificing security, reliability, or model quality.

Watch for specific deployment examples with measurable workloads. Customer-service agents, document systems, coding tools, and retail inspection applications can demonstrate repeatable demand.

The third signal is AIDC utilization and service performance. China Mobile plans to accelerate data center commissioning, but new capacity must translate into dependable inference.

Investors and customers should look for changes in computing-service revenue, infrastructure utilization, latency, availability, and unit economics.

If computing revenue rises alongside sustained token consumption, the carrier will have evidence that its network and customer base can support a new AI business.

If infrastructure spending rises without durable usage, the package strategy will look more like capacity promotion than a stable commercial market.

Competitor responses will provide supporting evidence. Simpler cloud purchasing or improved packages from China Telecom and China Unicom would indicate that China Mobile is applying real pressure.

The broader significance extends beyond China. Telecom operators in other markets are also considering AI applications, devices, and computing services as additions to conventional connectivity.

China Mobile offers a particularly large test because it can combine national distribution, cloud infrastructure, and established billing relationships.

Success would not make cloud platforms obsolete. It would show that many customers prefer AI delivered through a managed service layer instead of a direct model contract.

Failure would expose the limits of treating model consumption like mobile data. Tokens may prove too inconsistent, technical, or application-dependent for conventional package design.

For developers and enterprise buyers, the immediate task is not choosing a package that lacks final terms. It is identifying which workloads benefit from aggregation and which require direct model control.

Teams should document their model requirements, data restrictions, latency targets, and expected consumption before evaluating the national offering. A searchable AI knowledge base can help organize those decisions across technical and business stakeholders.

China Mobile token plans now have a national direction, an orchestration platform, infrastructure backing, and an enormous distribution channel. The unresolved question is whether the company can make those pieces feel like one dependable product.

When the final terms arrive, look past the size of the token allowance. Ask which models are included, how usage is counted, where data travels, and what happens when demand peaks.