Token going global is emerging as a brand new business.

On the evening of October 18, a $200 million (equivalent to 1.338 billion yuan) AI order has placed Runjian Co., Ltd. in the spotlight of Token offshore business.

This A-share company, whose main business covers communication network management, operation and maintenance and has been continuously laying out computing power services in recent years, has just obtained a cross-border AI service contract worth 200 million US dollars. The contract has a term of one year, overseas customers call large model inference services through API interfaces, and all relevant computing power resources are deployed in intelligent computing centers within China's territory.

More notably, the contract includes a minimum purchase guarantee clause: if the service fee corresponding to the Token actually consumed by the customer during the contract year is less than 200 million US dollars, the difference shall be made up as agreed.

The amount of this single order is equivalent to about 35.6% of Runjian Co., Ltd.'s operating revenue in the first half of 2026. What it sells is neither servers nor chips, and what customers purchase is the Token consumed during the operation of the large model.

A new profit-making model is emerging in China's AI offshore business: converting domestic computing power resources into services, allowing overseas customers to call them via the network and pay according to usage.

This transaction also brings a concept that originally mostly appeared in AI industry discussions into the commercial contracts of listed companies.

In the past, when the market paid attention to China's AI going global, it usually focused on whether large models could catch up with overseas competitors and whether domestic chips could enter the international market. Nowadays, model invocation services, inference computing power and Token billing have also become new directions for enterprises to explore overseas revenue.

For China's AI industry chain, this means new market space. For Runjian Co., Ltd., a large contract also brings a more realistic test question: when the intelligent computing centers built in China start to serve overseas customers, how much profit can this business really make?

After all, 200 million US dollars is only the contract amount. What really determines the value of this transaction also includes computing power cost, equipment depreciation, customer payment collection, and whether service providers can retain sufficient profits in the fierce price competition.

The first large order of Token offshore business has been signed. Next, the market needs to see how this business turns from contract amount to real revenue and profits.

01

China's AI Going Global Has Opened Up a New Path

To understand Token offshore business, we must first figure out what exactly customers are purchasing.

When users ask questions to large models, generate texts or write codes, the model will process the input content and generate output results. Inputs and outputs are usually measured by Token, and service providers charge fees according to model types, input and output quantities and specific billing rules.

Developers can call the model through API and integrate it into their own applications. Therefore, overseas enterprises can directly use AI capabilities without building a complete model inference system on their own.

This model reduces the initial investment for customers to use AI, and also gives model service providers the opportunity to charge by usage.

The increasing activity of Chinese large models in overseas developer communities has laid a market foundation for this business.

Chinese AI enterprises including DeepSeek, Moonshot AI, Zhipu AI, MiniMax and other Chinese AI enterprises are expanding overseas markets through different model products and technical routes. Model performance, invocation price, open weight strategy and developer tools will all affect the choices of overseas customers.

For developers, when choosing models, they often need to consider the answer quality, response speed, invocation cost and stability at the same time. As long as the model can meet specific business needs and is competitively priced, it will have the opportunity to enter application development and enterprise workflows.

However, there is still a gap between getting attention for the model and achieving commercial success. Developer trial, actual payment, enterprise procurement and long-term renewal represent market progress at different stages. The increase in invocation volume also needs to be combined with price changes and cost structure to judge the quality of revenue growth.

This order of Runjian Co., Ltd. provides another perspective for observation.

The company has long been engaged in communication network management, operation and maintenance, and has gradually expanded its digital network, energy network and computing power network businesses. With the accumulation of intelligent computing center construction and operation capabilities, the company has begun to explore new computing power services such as Token.

In this contract, Wuxiang Cloud Valley provides large model inference invocation services to overseas customers through API, and the computing power facilities remain in China. Customers can use relevant resources remotely and settle accounts in accordance with the contract agreement.

This kind of business arrangement provides a feasible path for China's computing power infrastructure to participate in the global AI market. Even if enterprises do not independently develop basic models, they can still participate in this market by virtue of their computing power resources, technical operation and service delivery capabilities.

02

Competition Lies in Price, Computing Power and Payment Collection

In the Token industry chain, there are at least two types of enterprises looking for commercial opportunities.

The first type is model vendors. They provide API services of self-developed models to customers, and their revenue is related to customer invocation volume, product pricing and contract arrangements. The model's capabilities, inference efficiency and developer ecosystem determine whether enterprises can continuously attract customers.

The second type is computing power service providers. They provide services around computing resources, inference deployment, API access and operation delivery. Some enterprises have their own intelligent computing centers, while others provide services by integrating computing resources from different sources.

The contract of Runjian Co., Ltd. this time is closer to the latter model.

The capability requirements for the two types of enterprises are different. Model vendors need to continuously improve model performance and control the cost of each inference; computing power service providers need to do a good job in resource scheduling, equipment maintenance and customer service, so as to obtain higher utilization rate of existing computing resources.

The two sides can also form a cooperative relationship. Models, computing power, interface services and overseas customer channels can be provided by different enterprises respectively, and then connected through commercial contracts.

This division of labor expands the space for China's AI industry chain to participate in the overseas market, and also brings new competition.

When more and more enterprises can provide similar inference services, customers will compare invocation prices, response speed, service stability and technical support capabilities. If service providers lack obvious cost advantages or customer stickiness, revenue growth will easily be affected by price competition.

Therefore, Token invocation volume is only one indicator to measure business scale. Customer retention rate, single inference cost, service gross profit margin and payment collection situation also determine the value of this business.

03

How Much Profit Can It Bring to Runjian Co., Ltd.?

The scale of this contract of Runjian Co., Ltd. is quite considerable.

The company achieved an operating revenue of 3.756 billion yuan in the first half of 2026, and the contract amount of 200 million US dollars is equivalent to about 35.6% of its operating revenue in the first half of the year.

However, there are obvious differences between contract amount, current period revenue and profit.

According to the announcement, the contract term is from October 19, 2026 to October 18, 2027, the service adopts the post-payment mode by usage, and the settlement cycle is a natural month. The finally confirmed revenue amount needs to be determined according to actual performance and accounting treatment.

The contract also sets an annual minimum purchase guarantee clause. If the service fee corresponding to the Token actually consumed by the customer during the contract year is less than 200 million US dollars, the difference shall be made up as agreed. This provides certain purchase guarantee for the contract, but customer payment, contract performance and revenue confirmation still need subsequent verification.

The profit performance of Runjian Co., Ltd.'s existing computing power business can also help understand the cost pressure faced by this order.

According to the company's 2026 semi-annual report, the computing power network business achieved an operating revenue of about 519 million yuan in the first half of the year, a year-on-year increase of 50.31%; the operating cost was about 483 million yuan, a year-on-year increase of 55.08%; the gross profit margin was 6.97%, down 2.86 percentage points year-on-year.

Revenue is growing rapidly, while the cost is growing even faster. The company stated that the low gross profit margin of the Token factory business is mainly affected by factors such as high depreciation during the period of fixed assets put into operation.

This is related to the operating characteristics of computing power services. Server procurement requires a large amount of capital. After the equipment is put into use, depreciation will be generated continuously, and power, operation and maintenance and resource scheduling will also affect the cost. Even if customer demand grows, enterprises still need to improve equipment utilization to improve profitability.

We can make a simple scenario calculation with the gross profit margin of existing business: assuming that the total contract amount of 200 million US dollars is finally converted into revenue, and the gross profit margin reaches 6.97% of the computing power network business in the first half of the year, the corresponding gross profit is about 93.2 million yuan.

Of course, the above is only a simple scenario calculation. The commercial value of this order ultimately depends on whether the company can deliver services at a reasonable cost and retain sufficient gross profit from the revenue.

04

Computing Power Alone Is Far From Enough

Token service seems to be able to be delivered quickly through API, but a complete set of operation capabilities is still required behind it.

The first is computing power supply. Large model inference needs to continuously respond to customer requests, and equipment failure, insufficient resources or service interruption may affect customer experience. Service providers must ensure the stable operation of computing resources, networks and systems.

The second is customer expansion. A single order can generate phased revenue, and long-term renewal and continuous procurement can create opportunities to build a stable business scale. Service providers also need to prove that they have sufficient price competitiveness and service capabilities to make customers willing to continue to use their services.

Cross-border business will also bring additional requirements. Service providers need to pay attention to customer identity, service content, data processing methods and regulatory provisions in relevant markets. The deployment of computing power in China does not mean that cross-border services can ignore compliance and operational risks.

Runjian Co., Ltd. also prompted in the announcement that the Token business puts forward high requirements for capital investment, technical capabilities and customer expansion capabilities, the market competition is fierce, and there is still uncertainty in future order acquisition and large-scale commercialization.

The company's financial situation further illustrates the capital investment required for transformation.

In the first half of 2026, the net cash flow generated by Runjian Co., Ltd.'s operating activities was about -2.876 billion yuan. The company explained that the expansion of net cash outflow is mainly related to the increased investment in new businesses such as computing power and energy, as well as the concentration of customer payment collection in the second half of the year.

This set of data needs to be understood in combination with the characteristics of the company's business, but it also reminds the market: computing power service providers usually need to invest funds first to build or prepare resources, and then gradually recover costs through subsequent services.

If customer demand continues to grow and existing equipment is fully utilized, the depreciation of fixed assets can be shared by more businesses. If customer procurement falls short of expectations and the equipment utilization rate is low, depreciation and other fixed costs may drag down profits.

Therefore, service providers need to do two things well at the same time: strive for stable customer demand, and control the delivery cost per unit of Token.

05

Can It Become A New Export of China's AI Industry?

Token offshore business provides a new internationalization method for China's AI industry.

Model vendors can sell model capabilities to overseas customers through API, while computing power service providers can provide inference services relying on domestic intelligent computing centers. With the increase of application development and enterprise deployment, a more detailed division of labor may be formed among models, computing power and service delivery.

This is particularly important for enterprises that have invested a lot of funds in the construction of intelligent computing infrastructure. Overseas customers can increase potential demand sources, help enterprises expand service scope, and find more commercial uses for existing computing resources.

However, whether this business can continue to grow still depends on several practical indicators.

First, whether overseas customers can continue to purchase, and gradually form renewal and repurchase; second, whether the gross profit margin of the computing power network business can be improved, proving that revenue growth has begun to bring better profitability; third, whether contract revenue can be converted into stable cash flow to cover computing power investment and daily operating costs.

These three indicators can better reflect the maturity of the business model than the amount of a single order.

Runjian Co., Ltd.'s 200 million US dollar order shows that cross-border Token services have entered the stage of specific commercial contracts.

Next, the company needs to prove that this order can produce sustainable business results through actual performance, customer payment collection and profit performance.

For the whole industry, opportunities are also accompanied by competition. Model technology continues to iterate, inference prices may change, and overseas customers will constantly compare the cost and quality of different service providers. Only by building advantages in service capabilities, cost control and customer relationships can enterprises gradually accumulate long-term value from short-term orders.

Token offshore business is becoming a new business. Whether it can develop into a stable overseas revenue source for China's AI industry ultimately depends on whether enterprises can convert computing resources into continuously paid services and maintain reasonable profits and cash flow during the expansion process.

This article is from the WeChat official account "Alpha Workshop Research Institute", written by Fan Qing, and published with authorization from 36Kr.