China’s Ministry of Industry and Information Technology has set an explicit target for the businesses that turn AI tools into working industry deployments. A notice reproduced by Beijing News on August 31 calls for a national resource pool with more than 2,000 AI application service providers by the end of 2026 and at least 3,000 by the end of 2027.
The policy addresses a part of the AI economy that is less visible than foundation models or chips but decisive for commercial adoption. The ministry defines these service providers as companies or institutions offering AI solution consultation and planning, implementation and delivery, operations management, and security governance around a user organization’s intelligent-transformation needs. In other words, the target is not simply for more AI companies. It is for a deeper layer of firms able to move technology into complex business settings.
A National Pool for Companies That Deliver AI
The notice says the program will establish provider archives at the regional level before MIIT consolidates them into a national resource pool. It seeks organizations that understand industry pain points, technical mechanisms, security risks, and delivery operations. The wording reflects an important reality of AI adoption: a model or agent is not useful to most companies until it can be connected to data, software, workflows, compliance controls, and staff responsibilities.
MIIT’s two-stage target is specific. By the end of 2026, the national pool should contain more than 2,000 providers and form a multi-level structure with stronger ability to deliver complex settings. By the end of 2027, the number should be at least 3,000, supporting what the notice calls a full-factor, whole-chain, all-scenario AI application-service ecosystem. Provinces that host national AI industrial innovation and application pilot zones are expected to have at least 100 enterprises in their own provider pools by the end of 2027.
The focus on delivery follows recent government work on AI inputs. EastFrontier’s August 31 report on a 32-city data-annotation pilot described an effort to strengthen the data side of AI development. MIIT’s service-provider action addresses a later stage. It is about taking models, data, compute, and industry knowledge into customer environments where results must be delivered, operated, and governed.
The notice also envisages “AI application service groups” that would bring together upstream and downstream organizations around areas such as model-data coordination and compute-power coordination. It calls on regions to submit at least 10 such groups in principle. The aim is to build integrated solutions rather than leave a customer to assemble separate models, infrastructure, applications, and safety controls alone.
Procurement, Standard Products, and Safety Are Central to the Plan
MIIT asks providers to package high-frequency, urgent, reusable business needs into modular and standardized solution products. That emphasis matters because it suggests policy support for services that can be replicated across sectors rather than customized from the ground up for each buyer. The notice describes these products as small, fast, light, and precise, a formulation often used for solutions that can be deployed with lower barriers for smaller organizations.
The ministry also encourages regions to open real application settings to providers and to support practical deployment through incubators, testing facilities, and AI application pilot bases. It mentions procurement mechanisms such as first purchases, first use, and risk compensation to increase the purchase of large models, agents, and token services. Those are encouragements rather than a quantified funding pledge. The notice does not specify a national spending total, a list of eligible models, or guaranteed contracts.
Security and compliance are woven into the program rather than left to a final review. The notice calls for providers to incorporate cybersecurity, data security, ethics governance, and business compliance throughout research, deployment, and application. It also says providers should receive information on security risks involving AI software and hardware, including operating systems, databases, and training and inference chips.
That breadth makes the providers a potential interface between China’s AI policy objectives and enterprise users. They are being asked to solve technical integration challenges while also managing risks that may arise from the systems they deploy. The ministry’s definition includes operations management and security governance, indicating that its vision extends beyond selling software licenses.
China’s Next AI Competition Is About Execution Capacity
The notice links application services to the broader infrastructure that China is building. It encourages providers to connect with national compute-network hubs, interoperability nodes, and the China Computing Platform, and it points to compute vouchers as a policy tool for reducing costs. EastFrontier’s August 23 coverage of Inner Mongolia’s AI service exports showed the policy interest in pairing compute resources with services that can travel beyond local markets. MIIT’s new action adds a national talent and delivery framework to that objective.
The program also calls for universities and vocational schools to build training bases with leading providers. The intended result is a workforce that understands business needs, models, safety, and implementation. It further encourages companies to build frontline deployment-engineer teams that work at user sites, an acknowledgement that many AI projects require sustained on-the-ground adaptation.
Whether the target succeeds will depend on quality as well as quantity. A pool of 3,000 listed organizations would not automatically prove that China has 3,000 capable AI integrators. The meaningful test will be whether providers can deliver reliable systems in difficult environments, make them affordable for customers, and keep risk management integral to the work. MIIT has supplied the numerical goal. The next challenge is building the execution capability behind it.