China's Token Export Trade: Runjian's $200M AI Contract Tests a New Frontier
A $200 million AI services contract announced on the evening of October 18 thrust Runjian into the spotlight of a nascent but consequential trend: the cross-border export of Chinese computing power, billed by the token.
The one-year deal, signed through Runjian's subsidiary Five Elephant Cloud Valley, grants an overseas client API access to large language model inference services hosted entirely within domestic intelligent computing centers in China. The contract includes a minimum purchase guarantee: should the client's actual token consumption fall short of the $200 million threshold during the contract year, it must pay the difference.
The figure is striking in context. Runjian reported revenue of RMB 3.756 billion yuan (approximately $518 million) for the first half of 2026. The contract value alone equals roughly 35.6% of that figure—a single deal, for a service measured not in hardware shipments or software licenses, but in the computational units consumed each time a language model processes a request.
A New Commercial Model for Chinese AI
The transaction represents a structurally distinct approach to China's AI internationalization. Rather than selling chips, servers, or proprietary software, Runjian is monetizing inference capacity: overseas developers and enterprises access model capabilities via API, pay according to usage, and never need to build or operate their own inference infrastructure.
This model lowers the barrier for international clients adopting Chinese AI capabilities. It also allows domestic operators of intelligent computing centers to generate foreign-currency revenue without physically exporting any equipment—the compute stays in China; the service crosses the border.
Chinese AI developers including DeepSeek, Moonshot AI, Zhipu AI, and MiniMax have been expanding their presence in international developer communities, competing on model performance, open-weight strategies, and pricing. The growing international interest in Chinese foundation models provides the demand-side foundation for this type of compute export service to take hold commercially.
Runjian, whose core business spans telecommunications network management, operations and maintenance, and increasingly digital and computing network infrastructure, has been building intelligent computing center capacity in recent years. This contract reflects a deliberate effort to convert that infrastructure into an exportable service.
Margin Pressure Behind the Headline Number
The commercial logic is straightforward; the economics are more demanding.
Runjian's existing computing network business offers a clear-eyed precedent. In the first half of 2026, the segment generated revenue of approximately RMB 519 million, up 50.31% year-on-year. But operating costs rose faster, climbing 55.08% to roughly RMB 483 million, pushing the gross margin down 2.86 percentage points to 6.97%. The company attributed the thin margin primarily to elevated depreciation charges as newly commissioned fixed assets entered service.
Applying that margin mechanically to the full contract value—a scenario that assumes complete revenue recognition and no improvement in unit economics—would imply gross profit of approximately RMB 93.2 million on the $200 million contract. That figure, while directionally useful, is only an illustrative estimate. Actual profitability will depend on utilization rates, depreciation schedules, power and operational costs, and the pace at which client payments are received.
The cash flow picture adds further texture. Runjian's net cash flow from operating activities was approximately negative RMB 2.876 billion in the first half of 2026, reflecting the capital-intensive nature of compute infrastructure build-out and the company's pattern of collecting receivables more heavily in the second half of the year. Infrastructure-heavy businesses must front capital expenditures before service revenues materialize, meaning the path from contract signing to realized profit requires disciplined cost management and reliable client collections.
What the Guarantee Clause Does—and Doesn't—Resolve
The minimum purchase guarantee embedded in the contract provides a degree of revenue protection. If the client does not consume sufficient tokens to reach the $200 million threshold, contractual obligations require it to make up the shortfall.
This mechanism offers a floor, not a ceiling—and it does not eliminate execution risk. Revenue recognition will depend on actual service delivery and applicable accounting standards applied on a monthly settlement basis. Client creditworthiness, cross-border payment flows, and regulatory compliance across jurisdictions all remain variables that will define the ultimate financial outcome.
Runjian acknowledged in its disclosure that the token services business carries substantial requirements for capital commitment, technical capability, and client acquisition, and that future order growth and large-scale commercialization remain uncertain.
The Broader Stakes for China's AI Supply Chain
Runjian's contract illustrates how the Chinese AI value chain is developing an international revenue model that extends beyond model development itself. Two distinct commercial roles are emerging: model providers that sell API access to their own foundation models, and compute service operators that supply inference infrastructure and delivery capabilities. The two can collaborate, with model capability, compute resources, API interfaces, and overseas client relationships supplied by different entities linked through commercial agreements.
This division of labor expands the pool of Chinese companies that can participate in global AI demand—including those without proprietary models. It also intensifies competition. As more operators offer comparable inference services, clients will increasingly arbitrate on price, latency, service reliability, and technical support. Operators lacking durable cost advantages or strong client retention will find revenue growth vulnerable to margin compression.
Three metrics will ultimately determine whether the token export model becomes a sustainable revenue stream rather than a series of headline-generating contracts: whether overseas clients renew and deepen their purchasing commitments; whether gross margins in the computing network segment improve as utilization rises and depreciation is spread across greater volumes; and whether contract revenues convert into stable operating cash flows sufficient to cover ongoing capital requirements.
Runjian's $200 million deal marks the arrival of token-based cross-border AI services in the formal commercial contract stage. Whether it marks the beginning of a scalable business—for Runjian and for the broader industry—will be answered in subsequent quarters of fulfillment data, margin trends, and client retention figures.
Related Coverage:
Token Exports: How China Is Selling AI Compute Power to the World
Beyond Token Prediction: The 2026 Pivot to "World Models" and the Industrial Shakeout of Embodied AI