Jinan drafts 2026-2028 action plan for ‘AI + Manufacturing’

In September 2026, Jinan’s Bureau of Industry and Information Technology and 12 other departments issued the ‘Jinan Implementation Action Plan for “AI + Manufacturing” (2026-2028).’ It sets targets for developing more than 40 specialized industrial large models and smart agents, more than 30 typical application scenarios, at least three artificial intelligence capability centers and at least five Qilu ‘Model Power Spaces’ by 2028.

Jinan’s plan is a government framework for connecting AI with manufacturing’s ‘on-site data’ and ‘process management,’ rather than simply supporting research and development. Its public targets and cross-process approach mean adopters will be judged not only on model performance but also on their ability to implement AI, including data preparation and safety management.

Jinan’s plan aims to expand AI adoption from ‘individual demonstrations’ to ‘the entire manufacturing process,’ extending a policy theme already seen in other regions. But its presentation as a 2026-2028 action plan jointly issued by 13 departments, with numerical targets of more than 40 industrial large models, more than 30 typical scenarios, at least three capability centers and at least five ‘Model Power Spaces,’ signals an effort to move beyond slogans and toward operational planning. Public information already shows that manufacturing sites often keep sensor data, equipment inventories, quality records and maintenance histories separately, making it difficult to prepare them for AI training and inference. The plan’s emphasis on data catalogs, corpus products and integrated mechanisms appears to reflect that practical constraint. This newspaper has no related articles, so a direct comparison with its previous coverage is not possible. But even from public information, the central question is not the development of large models itself, but how to standardize on-site data and integrate AI into specific processes. A case involving Inspur Yunzhou Industrial Internet’s coal-chemical AI agent, which converted the experience of more than 20 veteran technicians into a knowledge graph with 36,000 entries, illustrates an approach based on structuring workers’ tacit knowledge and linking it to quality. Meanwhile, even with 10,000 P-scale computing power, delays in preparing manufacturing data can leave model training far removed from on-site deployment. Conversely, if data collaboration and application development are integrated, AI is likely to take hold first in areas where results are easier to measure, such as process management, quality assurance and maintenance, rather than design support. On the market side, it is important that Jinan’s plan is based on a ‘13+34’ industrial-chain structure. This is a policy framework for bringing together not only AI suppliers but also manufacturing demand within the region; results will depend more on the mix of industries and the complexity of their processes than on general AI investment. Competition will test how roles are divided among model developers, industrial software companies, equipment vendors and industrial internet companies, but this article alone does not show which approach will become dominant. The plan also calls for promoting lightweight tools among small and medium-sized enterprises, but does not specify actual deployment costs or pricing models. Three points warrant follow-up. First, which processes will be served by the more than 40 industrial large models, and which will go into operation first? Second, will the at least three artificial intelligence capability centers handle data preparation and safety management as well as computing resources? Third, how will individual cases such as Inspur Yunzhou Industrial Internet’s be extended to mass deployment across Jinan?

For Japan’s manufacturers, a key issue is how to organize equipment data and quality records on the factory floor. Jinan’s approach of prioritizing data collaboration and integrated mechanisms over model development offers particular lessons for industries where factory data remains fragmented.