Introduction: Transforming domestic computing resources into services, enabling overseas clients to tap into them via the internet, and implementing a pay-per-use model.

On the evening of October 18, Runjian Co., Ltd. found itself thrust into the limelight of token export, thanks to a staggering $200 million (RMB 1.338 billion) AI order.

This A-share listed company, primarily known for its communication network management and operations, has been increasingly venturing into computing power services in recent years. It has now clinched a $200 million cross-border AI services contract. This one-year agreement empowers overseas clients to utilize large model inference services through API interfaces, with all computing resources housed in smart computing centers within China.

Remarkably, the contract features a minimum purchase guarantee clause: should the client's actual service fees for consumed tokens fall short of $200 million within the contract year, the deficit must be compensated as stipulated.

The order's value equates to roughly 35.6% of Runjian's revenue for the first half of 2026. What's being sold here are not servers or chips, but tokens consumed during large model operations.

A novel avenue for monetizing Chinese AI overseas has emerged: transforming domestic computing resources into services, enabling overseas clients to access them via the internet, and charging based on usage.

This transaction has also brought a concept, previously discussed mainly within the AI industry, into the commercial contracts of listed companies.

Historically, market attention on Chinese AI's global expansion centered on whether large models could keep pace with overseas rivals and whether domestic chips could penetrate international markets. Now, model access services, inference computing power, and token-based billing are emerging as new avenues for enterprises to tap into overseas revenue streams.

For the Chinese AI industry chain, this signifies fresh market opportunities. For Runjian, a hefty contract brings more tangible challenges: when domestically constructed smart computing centers begin serving overseas clients, how much profit can this venture yield?

After all, $200 million is merely the contract's face value. The transaction's true worth also hinges on computing power costs, equipment depreciation, client payments, and whether service providers can maintain adequate profits amidst fierce price competition.

The token export narrative now boasts its first major order. Next, the market awaits to see how this business can convert contract values into tangible revenue and profits.

01 A New Pathway for Chinese AI's Global Expansion

To grasp token export, it's crucial to understand what clients are purchasing.

When users pose questions, generate text, or write code using large models, the model processes the input and produces output. Input and output are typically quantified in tokens, with service providers charging based on model type, input/output volume, and specific billing criteria.

Developers can access models via APIs and integrate them into their applications. This enables overseas enterprises to leverage AI capabilities directly without constructing complete model inference systems themselves.

This model reduces clients' initial AI investment and offers model service providers opportunities to charge based on usage.

The burgeoning activity of Chinese large models in overseas developer communities lays a market foundation for this business.

Chinese AI firms like DeepSeek, Yuezhi, Zhipu, and MiniMax are expanding overseas through diverse model products and technical approaches. Model performance, access pricing, open-weight strategies, and developer tools all influence overseas clients' decisions.

For developers, selecting a model often entails considering response quality, speed, cost, and stability. As long as a model meets specific business needs and offers competitive pricing, it stands a chance of being integrated into application development and enterprise workflows.

However, model attention and commercial success remain worlds apart. Developer trials, actual payments, enterprise procurement, and long-term renewals represent distinct stages of market progression. Increased usage must be weighed alongside price fluctuations and cost structures to gauge revenue growth quality.

Runjian's order provides another perspective.

The company has long been involved in communication network management and operations, gradually branching out into digital, energy, and computing power networks. With accumulated expertise in smart computing center construction and operation, it began exploring novel computing power services like tokens.

In this contract, Wuxiang Yungu furnishes large model inference access services to overseas clients via APIs, with computing facilities remaining in China. Clients can remotely utilize these resources and settle accounts as agreed.

This business arrangement offers a viable pathway for Chinese computing infrastructure to participate in the global AI market. Even without self-developed base models, enterprises can participate through computing power resources, technical operations, and service delivery capabilities.

02 Competing on Price, Computing Power, and Payments

At least two types of enterprises are vying for business opportunities in the token industry chain.

The first category comprises model vendors. They provide API services for their self-developed models, with revenue tied to client usage, product pricing, and contract terms. Model capabilities, inference efficiency, and developer ecosystems determine their ability to continuously attract clients.

The second category consists of computing power service providers. They offer services centered around computing resources, inference deployment, API access, and operational delivery. Some own smart computing centers, while others integrate resources from various sources.

Runjian's contract aligns more closely with the latter model.

The capability requirements differ between the two. Model vendors must continually enhance model performance and rein in inference costs per operation, while computing power service providers must prioritize resource scheduling, equipment maintenance, and client services to maximize computing resource utilization.

The two can also forge partnerships. Models, computing power, interface services, and overseas client channels can be provided by different enterprises and linked through commercial contracts.

This division of labor expands the space for China's AI industry chain to participate in overseas markets but also introduces new competition.

As more enterprises offer similar inference services, clients will compare access prices, response speeds, service stability, and technical support capabilities. Service providers lacking clear cost advantages or client loyalty may find their revenue growth easily undermined by price competition.

Thus, token usage is merely one indicator of business scale. Client retention rates, cost per inference, service gross margins, and payment collection also determine the business's value.

03 How Much Profit Can Runjian Anticipate?

Runjian's contract is substantial in scale.

The company reported RMB 3.756 billion in revenue for the first half of 2026, making the $200 million contract equivalent to approximately 35.6% of its H1 revenue.

However, contract value diverges significantly from current revenue and profits.

According to the announcement, the contract spans from October 19, 2026, to October 18, 2027, with services billed after usage on a monthly basis. The final confirmed revenue amount hinges on actual performance and accounting treatment.

The contract also includes an annual minimum purchase guarantee. If the client's actual service fees for consumed tokens fall short of $200 million within the contract year, the deficit must be compensated as agreed. This provides some purchase assurance, but client payments, contract performance, and revenue recognition still necessitate subsequent verification.

Runjian's existing computing power business profitability also sheds light on the cost pressures faced by this order.

According to its 2026 H1 report, the computing power network business reported approximately RMB 519 million in revenue, up 50.31% YoY; operating costs were approximately RMB 483 million, up 55.08% YoY; and the gross margin was 6.97%, down 2.86 percentage points YoY.

Revenue surged rapidly, but costs escalated even faster. The company attributed the low gross margin of its token factory business mainly to high depreciation during the fixed asset production launch period.

This relates to the operational characteristics of computing power services. Server procurement demands significant capital, with ongoing depreciation after deployment. Electricity, operations, and resource scheduling also impact costs. Even with growing client demand, enterprises must enhance equipment utilization to bolster profitability.

A straightforward scenario calculation can be made using the existing business gross margin: assuming the full $200 million contract value is ultimately recognized as revenue and achieves the H1 computing power network business gross margin of 6.97%, the corresponding gross profit would be approximately RMB 93.2 million.

Of course, this is merely a simplistic scenario calculation. The order's commercial value ultimately hinges on whether the company can deliver services at a reasonable cost and retain sufficient gross profit from revenue.

04 Computing Power Alone Is Insufficient

While token services may appear readily deliverable via APIs, they still necessitate a comprehensive set of operational capabilities.

First and foremost is computing power supply. Large model inference must continuously respond to client requests, and equipment failures, resource shortages, or service interruptions could tarnish the client experience. Service providers must ensure the stable operation of computing resources, networks, and systems.

Second is client acquisition. While a single contract can generate phased revenue, long-term renewals and ongoing purchases are imperative to build a stable business scale. Service providers must also demonstrate they possess sufficient price competitiveness and service capabilities to retain clients.

Cross-border operations also impose additional requirements. Service providers must consider client identity, service content, data processing methods, and regulatory requirements in relevant markets. Having computing power deployed domestically does not imply cross-border services can overlook compliance and operational risks.

Runjian also noted in its announcement that the token business imposes stringent requirements on capital investment, technical capabilities, and client acquisition abilities, with intense market competition and uncertainty in future order acquisition and large-scale commercialization.

The company's financial situation further underscores the capital investment required for this transition.

In H1 2026, Runjian reported approximately -RMB 2.876 billion in net cash flow from operating activities. The company explained that the expanded net cash outflow was primarily due to increased investments in new businesses like computing power and energy, coupled with client payments concentrated in the second half of the year.

These figures must be interpreted in the context of the company's business characteristics but also remind the market that computing power service providers often need to invest capital upfront to build or prepare resources, then gradually recover costs through subsequent services.

If client demand continues to surge and existing equipment is fully utilized, fixed asset depreciation can be spread across more business. If client purchases fall short of expectations and equipment utilization remains low, depreciation and other fixed costs may drag down profits.

Thus, service providers must accomplish two tasks simultaneously: secure stable client demand and control delivery costs per token.

05 Can It Evolve into a New Export Avenue for Chinese AI?

Token export presents a novel internationalization approach for China's AI industry.

Model vendors can peddle model capabilities to overseas clients via APIs, while computing power service providers can offer inference services through domestic smart computing centers. As application development and enterprise deployments proliferate, a more intricate division of labor may emerge among models, computing power, and service delivery.

This is particularly crucial for enterprises that have heavily invested in constructing smart computing infrastructure. Overseas clients can augment potential demand sources, aiding enterprises in expanding their service scope and discovering more commercial applications for existing computing resources.

However, the sustainability of this business still hinges on several practical indicators.

First, whether overseas clients persist in purchasing and gradually form renewal and repurchase patterns; second, whether gross margins in the computing power network business improve, proving that revenue growth is yielding better profitability; third, whether contract revenue can be converted into stable cash flow to cover computing power investments and operational costs.

These three indicators better mirror the maturity of the business model than a solitary contract value.

Runjian's $200 million order demonstrates that token cross-border services have entered the specific commercial contract stage.

Next, the company must leverage actual performance, client payments, and profitability to prove this order can generate sustainable business outcomes.

For the entire industry, opportunities come hand in hand with competition. Model technology continues to iterate, inference prices may fluctuate, and overseas clients will perpetually compare costs and quality among service providers. Only by establishing advantages in service capabilities, cost control, and client relationships can enterprises accrue long-term value from short-term orders.

Token export is emerging as a new business opportunity. Whether it can evolve into a stable source of overseas revenue for China's AI industry ultimately hinges on whether enterprises can transform computing resources into continuously paid services while maintaining reasonable profits and cash flow during expansion.