Northern China’s next AI infrastructure test is not simply whether it can add more servers. It is whether low-carbon computing capacity can be turned into an exportable service. At the Green Computing Power conference in Hohhot on August 22, Xinhua reported that 12 projects worth 186.46 billion yuan, or about $27.49 billion, were signed alongside the start of a platform intended to connect foreign developers with Chinese models and computing resources.
That makes the event more consequential than a conventional data-center announcement. The signed projects involve Volcano Engine, ByteDance’s cloud platform, Cambricon Technologies, and China Telecom and cover green intelligent-computing centers, token factories, and equipment for computing power and electricity. The list puts cloud delivery, domestic accelerator supply, telecom infrastructure, and energy equipment in the same regional buildout. That combination is central to China’s attempt to make AI capacity usable at industrial scale rather than merely available on a rack.
Hohhot’s New AI Export Platform Targets Developers, Not Hardware Buyers
The platform is planned for Hohhot’s area of the Inner Mongolia pilot free-trade zone. Xinhua describes its role as giving overseas developers access to Chinese models and low-carbon compute while creating a route for Chinese AI firms to serve foreign markets. The planned service stack spans compliance review, capacity dispatch, token metering, settlement across borders, and help for developers.
Each element addresses a practical barrier that conventional server exports leave unresolved. A developer that wants access to a model needs a way to schedule inference capacity; a company that sells an API needs to measure usage; an overseas customer needs payment and compliance processes that work across jurisdictions. The Hohhot plan therefore frames AI capacity as a bundle of model access, metering, support, and settlement rather than as a shipment of machines.
That approach extends the shift already visible in China’s data-center strategy. EastFrontier previously examined DeepSeek’s planned one-gigawatt AI hub in Inner Mongolia, which focused on the scale of compute that a leading model developer could secure. The Hohhot platform turns the emphasis outward: the goal is to package regional capacity for developers and AI companies beyond China’s borders.
The choice of a free-trade-zone location also matters. Cross-border AI services create issues that a domestic computing cluster does not, including contractual terms, data handling, payments, and the accounting of tokens consumed by an application. The announced platform does not settle those questions by itself. It does, however, establish a local mechanism meant to handle them as part of AI delivery.
Wind, Solar, and Token Demand Reshape the Data-Center Equation
Inner Mongolia is building on a substantial physical base. Xinhua puts Horinger New Area’s footprint at 62 computing centers and 150,000 PFlops in aggregate, of which 143,000 PFlops are classified as intelligent computing. It reports a further 172,000 PFlops already operating in nearby Ulanqab, where intelligent-computing capacity represents over 95% of the figure. The region’s wind and solar resources are central to the green-computing narrative attached to those figures.
Those numbers should not be read as a direct measure of model quality or usable cloud availability. They describe regional compute capacity. Their importance lies in the scale of infrastructure that can be connected to model providers, cloud companies, and the new service platform. For a country trying to reduce AI bottlenecks, capacity is only useful when customers can reach it, workloads can be routed to it, and electricity costs remain manageable.
China’s own demand creates the immediate rationale. Xinhua cited official figures that place daily token calls at 140 trillion in March 2026, compared with roughly 100 billion at the start of 2024. The jump illustrates why token measurement has moved from a billing detail to an infrastructure concern. A platform handling API consumption at this scale needs a common way to track usage across developers, models, cloud systems, and payment channels.
The Hohhot projects also follow earlier moves by companies seeking AI capacity in the same region. RedNote’s proposed Inner Mongolia data-center investment showed how internet platforms have been drawn to the area’s energy and land advantages. Hohhot’s new initiative broadens the cast beyond a single social platform or model lab to include telecom operators, chip companies, and cloud providers.
China’s AI Capacity Push Moves From Construction to Service Delivery
The strategic question is whether these projects can make Chinese AI services easier to consume internationally. Building a compute center and operating a developer-facing service are separate tasks. The former depends on hardware, power, cooling, and network connections. The latter requires dependable model access, transparent metering, customer support, and compliance arrangements that overseas users can understand.
The new platform’s inclusion of cross-border settlement is particularly revealing. It acknowledges that AI services are not exported only through server hardware or finished software licenses. They are increasingly delivered through recurring inference calls. In that model, the ability to measure and settle token usage becomes an operational part of China’s AI trade infrastructure.
For domestic firms, Hohhot offers a possible route to make regional green capacity commercially accessible outside China. For foreign developers, the proposition will depend on whether model choice, reliability, pricing, data practices, and support match the promise of the platform. The conference created the structure and signed the projects. The harder work will be converting a large regional compute base into a service developers actively choose.