--- title: "China's AI Policy Now Comes as a Catalogue. The Test Is Whether Anyone Copies It" date: 2026-11-07 category: China AI Decoded site: NeuroAI canonical: https://neuroai.site/a/na-decode-scenario-catalogue-replication language: en --- # China's AI Policy Now Comes as a Catalogue. The Test Is Whether Anyone Copies It > MIIT published 1,000+ AI use cases in October 2026, and a CAICT vice-president said nearly half of AI+ application scenarios are still stuck at the pilot stage. A plant manager in a parts business hears the same pitch from six vendors in a month. Each brings a slide from a different factory. None of them brings the factory. From this month he has somewhere else to look: the industry ministry has put the country's working cases in one place, grouped by sector, with the maturity of each one marked. The theory is that a solved problem in one plant becomes a purchasable specification in the next. Cataloguing is a different job from inventing. ## Key takeaways - MIIT published more than 1,000 high-value AI scenarios across 20 industrial sub-sectors on 15 October 2026 (MIIT, via CCTV). - A CAICT vice-president said statistical data show nearly half of "AI+" (人工智能+) scenarios remain at the pilot stage without scaled rollout (CCTV, 15 October 2026). - CAICT described a three-tier maturity grading of those scenarios: proven and replicable, validated but unscaled, and still under technical development (CCTV, 15 October 2026). - The eight-ministry AI+Manufacturing plan of 25 December 2025 targets 500 scenarios and 1,000 industrial agents by 2027 (工信部联科〔2025〕279号). - A RMB 60 billion (≈US$8.5bn / HK$66bn) national AI industry investment fund is already investing, MIIT said in October 2026. ## The instrument changed Chinese AI policy has largely been about supply: chips, compute (算力), models, datasets. The October document is about demand, and it is a different kind of paper. MIIT released it on 15 October, describing the 1,000-plus entries as a navigation map for industry, drawn from production lines rather than from laboratories, with 18 national AI innovation application pilot zones supported so far (MIIT, via CCTV, 15 October 2026). At a State Council briefing on 20 October the ministry put numbers on the surrounding machinery: 285 typical cases published, roughly 30 matchmaking events that connected more than 5,000 enterprises, and more than 100 national and industry standards, with over 70 international standards led from China (MIIT, 20 October 2026). None of those are achievements. They are instruments, and the instrument that matters is the catalogue, because it fixes the unit of policy at the single use case. ## Three tiers, three different claims The useful part of the document is that it does not treat its entries as equivalent. Each scenario is graded by maturity, in three bands whose names are worth learning, because Chinese partners will use them. - **推广级 (replicable):** a deployment has stabilised and can be copied by another firm. - **验证级 (validated):** a pilot has been completed, and expansion is the open question. - **攻关级 (under development):** a technical problem is still unsolved, and no buyer should treat it as available. That is rarer than it sounds. Most national AI programmes publish targets; a maturity label per use case is a claim a buyer can test, and the mix of labels is itself a measurement. ## The number the adviser volunteered The admission came from the technical side rather than the political one. Wei Liang (魏亮), a vice-president of the China Academy of Information and Communications Technology (中国信通院), said on the same release that statistical data show nearly half of "AI+" application scenarios remain at the pilot stage and have not achieved scaled rollout (CCTV, 15 October 2026). The same report noted the problem the catalogue is meant to answer: some enterprises follow the crowd, build the same system twice, and cannot move a proven solution from one plant to the next (CCTV, 15 October 2026). Note carefully what that number is not. It is a statement about AI+ scenarios in general, from an unspecified sample, not a count of the 1,000 catalogue entries. Multiplying the two — half of a thousand, so five hundred — produces a figure nobody published, and it is the most likely error a reader of the Chinese coverage will make. ## The money sits underneath A catalogue does not pay for anything. Three layers do. - **The national fund.** A RMB 60 billion (≈US$8.5bn / HK$66bn) national AI industry investment fund has been established and has invested in projects, MIIT said in October 2026. It is an investment fund, not a grant programme: it takes a position and expects a return. - **Local money.** City and district governments run their own vehicles; Guangzhou alone assembled a RMB 20 billion (≈US$2.8bn / HK$22bn) AI fund with a mother-fund structure (Guangzhou municipal government, 14 April 2026). - **Vouchers for the buyer.** The eight-ministry plan, issued as 工信部联科〔2025〕279号 on 25 December 2025, encourages local governments to issue compute vouchers (算力券) and model vouchers (模型券) to lower the cost of trying. The division of labour is the point. The state pays for the first copy; the customer has to pay for the second, and the catalogue exists because too few did. ## Honest limitations The 1,000-plus entries are not published as an itemised list with deployment status, so the count cannot be audited from outside, and tier assignments are made by the compilers. The 285 cases are selected, not sampled, which makes them evidence of possibility rather than of prevalence. The "nearly half" figure has no published sample size or method behind it. The national fund's portfolio is not itemised, and no independent body has measured how many catalogued scenarios were adopted by a second site. ## What readers can do now - When a Chinese partner cites a "high-value scenario," ask which of the three tiers it sits in. A 推广级 claim should come with a reference site you can contact; a 验证级 claim means one pilot, not a product. - Use 285, the number of published cases, as the denominator for demonstrated deployments, and treat the 1,000-plus list as a menu. Anyone quoting the larger number as market size has changed the subject. - Read the 2027 targets in 工信部联科〔2025〕279号 — 500 scenarios, 1,000 industrial agents, 100 datasets — as procurement language, then check whether your counterpart's product maps to a named sector in the plan rather than to the general category. --- Published by NeuroAI (https://neuroai.site/) — https://neuroai.site/a/na-decode-scenario-catalogue-replication Free to quote with attribution and a link to the original.