500 AI application service providers: that is the resource-pool target Zhejiang has set for the end of 2028. Under this action plan, the pool is not limited to model developers. Firms or institutions that do consulting and planning, project delivery, operations, or security governance also count.
By the end of 2027, the pool should reach at least 300, including 50 professional-level and 10 leading-level firms; by the end of 2028, it should expand to 500, including 100 professional-level and 20 leading-level firms. These are Zhejiang targets, not results already achieved.
Who is covered, and where?
The issuing body is Zhejiang's Office of the Leading Group on High-Quality Manufacturing (and Digital Economy) Development. The public source is the "documents and notices" column on the Zhejiang EIT department website. The scope is Zhejiang. The plan also calls for a link to a national online platform, but it does not automatically enroll firms outside the province. Digital service firms, manufacturers, and AI-native companies are listed as cultivation paths. The notice asks recipients to implement the plan; it is not a draft for comment. Intake rules and grading standards still have to be set.
What are the 2027 and 2028 numbers?
The figures above are two-year targets. The ladder is "basic, professional, and leading." Basic-level firms are encouraged to offer lightweight, fast-to-deliver solutions and to file for the pool. Professional-level firms should scale in a specific domain and work on vertical models or specialized agents. Leading-level firms are expected to have a model or agent platform and a toolchain. The plan says Zhejiang will "research and formulate" classification standards; this notice has no scoring sheet. A firm calling itself "leading-level" is not, on this document alone, officially designated.
Can firms apply to the pool now?
Qualified enterprises and institutions are to register, build archives and a capability list, and go through surveying, intake, filing, selection, evaluation, publication, and removal. The list is to be published regularly. This page gives no application URL, checklist, or deadline, and has no attachments. Companies that want in will need follow-up notices. Zhejiang buyers of AI services may later use the published capability list and government "AI+" scenarios.
How much are the vouchers worth?
Corpus vouchers: the plan "encourages qualified localities" to use them to cut data costs. That is a local-pilot option, not a province-wide consumer voucher, and no amount is stated. Compute and Token vouchers: Zhejiang will increase such policy support to lower compute costs, with no amount, eligibility list, or start date. Public data may be used by qualified service providers if data security is protected; that is not a promise to open all public data. "Pay for results" and "try first, pay later" are encouraged business-model experiments, not a government promise to pay users' bills. FDE and agent orchestrator are job categories to be "explored and studied," not existing credentials.
How we see it
Our view: the real opening is in the ability to implement and operate these services, not in waiting for subsidies. Consulting, delivery, operations, and security governance are written into the definition, so Zhejiang is trying to build teams that can put models into real work, not just add more model shops. The vouchers appear only as support tools; there is still no usable price list or application window. Reading the plan as if applications were already open would go beyond the text. For developers and AI delivery firms, watch the intake rules, classification standards, and the capability list still to be published.
Source: Zhejiang EIT Department notice, issued October 29, 2026, posted November 12, 2026 at 16:47 Beijing time, no document number on the page. Original notice