August 31, 2026 (InvestinChina.asia) — The Ministry of Industry and Information Technology (MIIT) on Monday unveiled a special campaign to cultivate artificial intelligence application service providers, setting explicit headcount targets for a national resource pool and throwing policy weight behind procurement of large models, agents and tokens, as well as the overseas rollout of AI applications.
The initiative arrives as China’s AI sector shifts from a technology race to commercial value realization. By the end of 2026, the national service provider resource pool is to exceed 2,000 entities, forming a multi-tiered echelon with markedly stronger delivery capabilities for complex scenarios. By the end of 2027, the pool is to grow to no fewer than 3,000 providers, underpinning an AI application service ecosystem characterized by full-factor coordination, end-to-end connectivity and full-scenario coverage.
Defining the Service Provider
According to the official definition, an AI application service provider is an enterprise or institution that delivers AI solution consulting and planning, implementation, operations management, and security governance around the intelligent transformation needs of user organizations. Such providers must field teams of at least three people, command AI-related technologies and application models, identify users’ core AI needs, and maintain the premises, equipment and tooling required for service delivery, with standardized service manuals and documentation.
The notice lays out five service categories: consulting and planning; delivery and implementation; operations and management; security and governance; and supporting services spanning training and testing & evaluation.
Four Priority Tasks
The campaign organizes four priority tasks. First, establish the service provider resource pool — provincial-level pools must contain no fewer than 100 enterprises by end-2027 in provinces that host national AI industry innovation and application pilot zones. MIIT will aggregate regional pools into a national registry and publish it in due course.
Second, elevate service supply. MIIT will organize providers to lead the formation of “AI application service corps” built around model-data resonance and compute-power-electricity coordination, while supporting breakthroughs in hardware-software adaptation and multi-model collaboration.
Third, drive scaled adoption. Providers are guided to package modular, standardized solution bundles for high-frequency,刚性 and reusable business processes, creating “small, fast, lightweight and precise” AI products and services. The notice explores mechanisms such as first-time procurement and risk compensation to step up purchases of large models, agents and tokens, improving the quality and efficiency of model and intelligence adoption across industries.
Regions with the right conditions are encouraged to build comprehensive “going global” service systems, leveraging cooperation platforms such as the Belt and Road, BRICS and China-ASEAN to land high-quality AI application projects overseas.
Fourth, strengthen support. Providers are backed to connect with national computing hub nodes, national computing interconnect nodes and the China Computing Platform, making good use of policy instruments such as “computing vouchers” to bring down compute costs.
From Capex Race to Commercialization
The policy push lands against a backdrop of deepening commercialization. In a research note, China Securities Co. argued that since 2026 the investment thesis for AI has gradually shifted from model capability and capital expenditure races toward commercial verification via orders, revenue and profitability. Overseas agent products have taken the lead in generating revenue increments, while domestic models are rapidly closing the gap with their peers in coding and agent tasks, with inference efficiency, token consumption and product revenue rising in tandem.
Sinolink Securities observed that China’s AI industry is moving from technology catch-up to commercial monetization, with 2026 a pivotal year. The market is scenario-driven rather than a zero-sum game: tech giants control computing power and general-purpose models, while vertical specialists embed themselves in high-barrier niche industries to deliver complete solutions. Token consumption has become a key metric for measuring enterprise intelligence and commercial value, with pricing power resting increasingly with vertical vendors that own irreplaceable scenario data.
Guojin Securities noted that continuously improving open-source model capabilities and falling token costs further favor AI applications. As open-weight and open-source models expand high-quality supply and lower the barrier to enterprise deployment — with models such as DeepSeek V4-Flash and GLM-5.3 enhancing agent, tool-calling and long-horizon task capabilities while keeping API prices low — application vendors can adapt and iteratively develop at lower cost and greater freedom, expanding AI usage across high-frequency scenarios including customer service, sales, office productivity and R&D.
Once enterprise agents penetrate core business processes such as sales, finance and human resources, they still need to continuously call real-time enterprise data and rely on permission controls, approval rules, system write-back and audit trails, the brokerage added.
The notice requires all regions to submit their local resource pool information to MIIT’s Department of Science and Technology by December 1, 2026, with regular updates to follow.