Recent discussions surrounding AI in China have focused on how the country's new models compare in performance to the latest models from the United States. However, performance comparisons alone do not fully capture the changes in China's AI policies. To understand Chinese AI, it is essential to examine not only the performance of the models but also the actual changes occurring in factories. A representative example is the 'Artificial Intelligence + Manufacturing' action plan released by Hebei Province on August 10. This plan details how AI will be applied to key local industries such as steel, chemicals, biopharmaceuticals, and automotive production processes.
This shift is also evident in the policies of the central government. Earlier this year, eight ministries, including the Ministry of Industry and Information Technology, released guidelines for the 'Artificial Intelligence + Manufacturing' initiative, which aims to expand the application of AI from research and development to encompass the entire manufacturing process, including production, marketing, and operations management. The goal is to deploy 1,000 industrial AI agents and 500 representative use cases by 2027. The accompanying 'AI Application Guidelines for Manufacturing Enterprises' require companies to assess their level of digitalization and identify bottlenecks in production before prioritizing areas for AI implementation. This indicates that the central government is focusing on how companies can apply AI to specific processes and the effects of such applications, beyond just developing large models.
Hebei Province has tailored this policy to its local industries, announcing plans to cultivate 4 to 5 large industrial models and 30 industrial AI agents by 2027, as well as establish 300 advanced smart factories. What stands out is the focus on specific applications: in steel, the emphasis is on steelmaking processes and energy management; in chemicals, on safety management and predictive maintenance; in biopharmaceuticals, on new drug development; and in automotive, on intelligent production scheduling, smart cockpit technology, and driving assistance. Even within a national strategy, the areas and processes for AI application can vary based on regional industries.
These differences extend beyond application areas. The policy tools used by local governments to implement AI in industrial settings also vary. Shenzhen is developing industrial AI agents as 'digital employees' for use in research and development, production, and supply chain management, while also promoting the establishment of innovation centers and industry knowledge alliances. This approach emphasizes broadening the base for developing and utilizing industrial agents. In contrast, Beijing's 'AI + New Materials' initiative goes beyond merely developing new materials and related equipment and components; it also requires demand companies to validate applications and encompasses a technology package that includes intelligent design, simulation, experimental data, and manufacturing processes. This approach embeds the need for research outcomes to transition into large-scale validation and actual industrial applications.
The relationship between manufacturing AI and regional industrial structures is inseparable. Each factory has different equipment specifications, process sequences, and quality standards, and the experience of skilled workers varies. There are also differences in the quantity, format, and management level of data held by companies. Understanding where delays occur in processes, the causes of recurring defects, and whether investments in automation can yield returns requires knowledge of the specific operational context. The role of local governments and regional support agencies becomes crucial in identifying company-specific needs, connecting necessary technologies, and disseminating validated methods to other companies.
However, the importance of local roles does not guarantee success. If multiple regions compete to create similar models, data centers, and industrial complexes, it can lead to redundant investments and resource waste. The Chinese central government has also urged regions to develop policies suited to their conditions while encouraging differentiated development among companies, warning against 'involutionary competition' that leads to wasteful rivalry. No matter how many models and agents are developed, if there is a lack of usable data or if they do not connect properly with existing equipment, production environments will not change. A critical consideration is whether companies can continue to use these systems at their own expense after pilot applications conclude.
South Korea has also begun efforts to connect manufacturing AI with regional industries. The Ministry of Trade, Industry and Energy is promoting the Manufacturing AI Transformation (M.AX) initiative, launching the 'MINI Alliance' involving local companies and universities in existing demonstration industrial complexes in Changwon and Gwangju, with plans to expand this framework to three additional complexes selected this year. The aim is for companies in these complexes to jointly utilize AI-related infrastructure, technology, and personnel. The challenge remains whether these systems will lead to tangible changes in the field. The ministry has supported the AI transformation of over 170 business sites through its AI Factory Leadership Project. Among these, 42 sites that started their projects relatively early have begun to show concrete results, with productivity increasing by an average of 30.1% and defect rates decreasing by an average of 15.5%. While these results are noteworthy, they cannot be generalized across all supported sites. It remains to be seen whether similar improvements will occur in other sectors and companies, and whether the benefits will persist after government support ends.
To assess the effectiveness of policies, it is essential to examine how production times, defect rates, equipment utilization, and energy consumption have changed, as well as to evaluate the costs incurred in adopting technologies against the benefits gained. Instances where expected results were not achieved can also provide valuable data. Documenting where technical limitations emerged and why economic viability was not secured can help reduce trial and error in future projects. The key is not only the success of individual demonstrations but also how those results inform subsequent policies. At this juncture, the questions raised by China's local government manufacturing AI policies for South Korea become clear. Therefore, when pursuing manufacturing AI policies, South Korean local governments should focus on designing the post-demonstration phase rather than just the demonstrations themselves. Ensuring that confirmed results can be replicated across different companies and sectors, and that businesses can sustain AI utilization independently after government support ends, is crucial. Ultimately, the success of manufacturing AI policies will be reflected not in the number of projects but in how deeply these changes take root in the field.
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.