Wu Qi/Wen

What has truly been rewritten is not only the factory, but also the logic of value distribution

In the past 20 years, discussions about China's manufacturing industry have mostly revolved around the same main theme: how large the scale is, how comprehensive the chain is, and how fast the upgrade is.

This main line is not wrong. China is still the only country in the world that has all the industrial categories in the United Nations Industrial Classification, and its manufacturing added value has ranked first in the world for many consecutive years. The complete industrial system, ultra large scale industrial chain, and strong engineering capabilities are still the most difficult reality to replicate in Chinese manufacturing.

But if we continue to understand the competition of the next decade only along this main line, we may misjudge the real changes.

AI? Especially the combination of large models, industrial intelligent agents, industrial software, and data infrastructure is rewriting not only several links in factories, but also the value allocation logic of manufacturing competition.

The Implementation Opinions of the "Artificial Intelligence+Manufacturing" Special Action Plan issued in 2026 propose to promote the deep application of 3-5 general large-scale models in the manufacturing industry, launch 1000 high-level industrial intelligent agents, create 100 high-quality datasets in industrial fields, promote 500 typical application scenarios, and cultivate 2-3 globally influential ecological leading enterprises by 2027. The policy focus is no longer just on "making factories smarter", but on restructuring the source of manufacturing competitiveness.

The most crucial change here can be summarized as' cognitive abundance '. Cognitive abundance in this article refers to AI significantly reducing the cost of acquiring, processing, and generating cognitive resources for enterprises, leading to the popularization of some previously scarce cognitive activities.

The so-called cognitive abundance does not mean that judgment is no longer important, nor does it mean that experience has become ineffective. Rather, it means that some cognitive activities that were originally expensive, slow, and difficult to replicate, such as analyzing, designing, screening, summarizing, identifying anomalies, and generating solutions, are being significantly reduced, accelerated, and partially replicated by models, algorithms, and data systems.

In the past, the advantages of many companies and countries were based on the premise that 'high-quality cognition is expensive, slow, and difficult to replicate'. Today, this premise is being shaken. What is shaken is not the value of cognition itself, but the exclusive foundation of some cognitive activities.

The manufacturing industry has long been perceived as' hard ', but behind it lies a large number of high-value cognitive processes: product definition, process development, parameter optimization, defect analysis, quality judgment, supply chain collaboration, and operation and maintenance feedback.

What truly loosens cognitive abundance is not the physical foundation of manufacturing, but the high cost cognitive activities that have been encapsulated in experience, processes, and organizational stability for a long time. Therefore, the real question that needs to be answered today is not only whether China's manufacturing industry can continue to grow and strengthen, but also what position China's manufacturing industry should occupy in the new value distribution structure when the rules of the industrial ecosystem are rewritten.

The advantages of China's manufacturing industry have not disappeared, but they are being re layered

If viewed from a traditional perspective, the advantages of China's manufacturing industry are still very prominent and will not easily disappear for a considerable period of time.

The most fundamental aspect is a complete industrial system, which used to be mainly reflected in cost, efficiency, and supply chain stability. But in the era of AI, it has taken on a new meaning: a complete industrial system means that various scenarios from raw materials, intermediate products to terminal manufacturing can continuously generate data, train models, validate processes, and spread capabilities within the same country. In the past, it was mainly a manufacturing advantage, but in the future it will increasingly become a training advantage, trial and error advantage, and diffusion advantage.

The high-density collaboration of industrial clusters remains a real barrier. The "close range, multi-level, and fast response" network formed by the Yangtze River Delta, Pearl River Delta, Beijing Tianjin Hebei, and several key industrial belts is still scarce globally. Parts, processes, debugging, quality improvement, and delivery coordination can be completed within a shorter spatial radius. This practical organizational capability will not be automatically weakened by the emergence of AI, but may be further enhanced by intelligent scheduling, data backflow, and rapid collaboration. No matter how strong the model is, components still need to flow in real space, and complex problems still need to be closed in real scenarios. In this sense, physical density remains a fundamental capability that is extremely difficult to replace in China's manufacturing industry.

The meaning of a super large market has also changed. In the past, the Chinese market was often understood as a "scale advantage" - the market was large enough to dilute costs; Today, it should be understood more as a "training advantage" - AI, especially industrial intelligence capabilities, cannot mature solely through laboratory research and development, but must be continuously validated in real devices, real customers, and real working conditions. According to publicly available information at the beginning of 2026, the number of domestic artificial intelligence enterprises has exceeded 6000, with an intelligent computing power scale of 1590 EFLOPS. The core industry scale is expected to exceed 1.2 trillion yuan, and its applications have covered key industries such as steel, non-ferrous metals, power, and communication, gradually deepening research and development, quality inspection, and customer service. For the manufacturing industry, this means that China's market advantage is partially transforming into model iteration and scenario validation advantages.

However, at the same time, some advantages that were once considered natural are being weakened, or rather, they are shifting from "natural advantages" to "advantages that can only be preserved by completing the transformation of form," the most typical of which is the labor efficiency advantage.

China's manufacturing industry is no longer primarily dependent on the lowest labor costs, but more on the large-scale supply of high-quality industrial workers and engineering technicians. The problem is that AI and automation will not only improve the efficiency of Chinese companies themselves, but also reduce the manpower density required for other economies to catch up. This means that China's efficiency advantage will not naturally disappear, but its sustainability will increasingly depend on whether experience can be digitized, processed, and modeled, rather than just relying on manpower scale and experience thickness. The head of the Ministry of Industry and Information Technology publicly emphasized in early 2026 the need to cultivate more compound talents who understand both artificial intelligence and manufacturing, which precisely indicates that the key to the problem has shifted from "too many people" to "whether people and systems can evolve together".

There is another type of advantage that has not disappeared, but if the conversion is not completed, it will exist in name but actually be discounted. The most typical example is industrial data resources. Data resources themselves do not automatically equal the advantage of the capability layer. It will only truly transform into new productive forces after being institutionalized, standardized, and infrastructurized.

There is another issue that has not received enough attention: the direction of AI's impact on consumer goods manufacturing and industrial goods manufacturing is not the same.

For consumer goods manufacturers, AI enhances the power of "deconstruction": demand perception is faster, flexible production is stronger, and platform based enterprises and quick responders are more likely to erode existing brand barriers. For industrial product manufacturers, AI is actually more likely to strengthen customer lock-in: the accumulated working condition data, optimization models, and operation and maintenance knowledge from the continuous operation of equipment at the customer site will turn one-time delivery into long-term relationships, and turn selling products into sustained business value.

The focus of support for consumer goods manufacturing should be more on product definition, brand autonomy, and user direct connectivity capabilities; The focus of support for industrial product manufacturing should be more on data closed-loop, continuous service, and solution capabilities. Deloitte's research on the "service first" transformation of the manufacturing industry also supports this judgment: serviceization and continuous operation capability are becoming important sources of profit improvement for industrial enterprises.

Therefore, a more accurate judgment is not whether China's advantages still exist or not, but that the advantages of China's manufacturing industry have begun to re stratify: some advantages are being amplified, some advantages are being weakened, and some advantages will gradually depreciate if they do not complete the transformation of form.

The future will not only be determined by the height of the industrial chain, but also by the position at the capability level

In the past 30 years, the main line of upgrading China's manufacturing industry has been to climb vertically along the industrial chain: from low-end to high-end, from assembly to independent research and development, and from following to leading. This route is generally correct and has achieved significant results.

But the change brought by AI is that future competition may not necessarily be reflected in "who is at the higher end of the industry chain", but more likely to be reflected in "who has stronger control and organizational power in key capability layers". The importance of Marco Iansiti and Karim Lakhani's research on competition in the era of artificial intelligence lies in their reminder that data, algorithms, and networked architectures are breaking traditional constraints of scale, scope, and learning, making it increasingly possible for businesses to compete across existing industry boundaries. For the manufacturing industry, future high-level competition is likely to occur more at the "capability level" rather than just within a single industry chain.

If the manufacturing industry is placed in a new competitive framework, it can be roughly divided into four layers.

The first layer is the physical execution layer, which includes manufacturing, delivery, engineering, and large-scale response capabilities. China has the strongest advantage in this layer, which is also the main practical basis for the global status of China's manufacturing industry today.

The second layer is the industry translation layer, which refers to the ability to translate general AI, general algorithms, and general digital capabilities into usable solutions for a specific industry. This layer determines who can turn "general intelligence" into "industry productivity".

The third layer is the intelligent infrastructure layer, which includes industrial software, digital twins, data infrastructure, industry models, and engineering knowledge base. Whoever masters this layer is closer to the "brain" of the manufacturing industry.

The fourth layer is the rules and standards layer, which includes data formats, interface protocols, security authentication, model evaluation, and the ability to shape ecological rules. Whoever can form influence at this level is closer to the high position in the future industrial ecosystem.

It is worth noting that this four layer framework is not a universal value ladder, and not all industries are 'more valuable the higher up'. In fields such as special materials, precision machining, and hot end components of aircraft engines, physical execution capability itself may be the highest value. The purpose of this framework is not to require all industries to 'move up', but to help determine where the key layers of future competition lie in your industry.

Under this framework, the strongest layer in China is currently the first layer, and efforts are being made to fill the second layer. The third layer is making rapid progress but still lacks system dominance, while the fourth layer is the most worthwhile to occupy the position ahead of schedule but still relatively weak at present. The special action plan issued by the Ministry of Industry and Information Technology and eight other departments in early 2026 is actually aimed at promoting China's manufacturing from "making factories smarter" to "filling and occupying positions at the capability level".

The global center position of China's manufacturing industry today is mainly built on the physical execution layer and the ultra large scale application layer. These advantages are still strong at present, but they are no longer automatically equated with future advantages. The future is truly determined not only by the position of the industrial chain, but also by the position of the capability layer.

The optimal path for China to develop industrial AI

It is precisely on this issue that the development path of China's industrial AI should not simply copy the United States or Europe. This is not because we deliberately "follow our own Chinese style path", but because the starting points, advantages, and weaknesses of the three parties are different, and the optimal path will naturally not be the same.

The advantages of the United States lie in its foundational models, leading platforms, computing power ecosystem, developer ecosystem, and standard spillover capabilities. The America's AI Action Plan released by the White House in 2025 is based on three pillars: accelerating innovation, building AI infrastructure, and maintaining leadership in international diplomacy and security; NIST will launch AI in Manufacturing related centers by the end of 2025, which is also strengthening the path of "occupying the intelligent core first, and then penetrating into the industry".

The advantages of Europe are different. It has a deeper foundation in industrial software, industrial equipment, industrial automation, digital twins, engineering standards, and institutional governance, so it emphasizes more on AI Factors Apply AI? Technological sovereignty and regulatory framework. The EU's AI Factors, Apply AI Strategy, and AI Act reflect a path of "infrastructure and regulatory framework first, then promoting widespread adoption in industries".

China's starting point is different from both the United States and Europe. China's biggest advantage is not a single platform or software system, but a complete industrial system, ultra large scale manufacturing scenarios, industrial cluster density, rapid engineering capabilities, and application iteration soil. The policy orientation of the "Implementation Opinions on the Special Action of 'Artificial Intelligence+Manufacturing'" is not simply copying the route of the United States' "platform first, industry later", nor is it completely copying the route of Europe's "rules first, diffusion later". Instead, it is taking a path that emphasizes more on scenario driven, industry landing, and systematic promotion.

Therefore, from the perspective of the overall upgrading of Chinese manufacturing, the most advantageous model should be a combination path that fully utilizes its comparative advantages.

Firstly, with application and scenario traction as the leading factors. The biggest practical advantage of China is that it has the most complete industrial scenarios, the longest industrial chain, and the richest real working conditions. The top priority is not to abstractly pursue the 'strongest universal model', but to transform massive real-world industrial scenarios into high-quality data, industry models, and replicable solutions. This will enable China to quickly form an advantage in the "industry AI translation layer".

Secondly, place industrial data infrastructure at a higher strategic position than general digitization. China is not without data, but rather with a large amount of data that is stored but not used, difficult to circulate, difficult to connect, and difficult to form a high-quality training loop. Industrial data infrastructure is not auxiliary engineering, but a new type of infrastructure for manufacturing upgrading in the AI era. The reason for this is not only because data is important, but also because China's most unique advantage is the source of data for all categories, processes, and scenarios.

Thirdly, using industrial clusters rather than individual enterprises as the basic unit for promoting industrial AI. This is a unique condition of China and an advantage that is difficult for both the United States and Europe to replicate. When an AI solution is successfully validated in a factory, it can quickly spread to surrounding similar enterprises because they use similar equipment, face similar process problems, and even share supplier and engineer networks. Promoting data standards, model interfaces, and industry solutions through the Yangtze River Delta automotive parts cluster, Pearl River Delta electronic manufacturing cluster, and other units is more likely to generate economies of scale and diffusion effects than allowing each enterprise to explore independently.

Fourth, use industry solutions and intelligent product layers as the main battlefield for value migration. If China only uses AI to improve factory efficiency, it may still remain in the position of "strong execution, weak platform" in the long run. The more favorable path should be to promote leading enterprises to move up towards industry models, intelligent products, continuous services, and platform nodes. Not just creating products, but turning them into a system of continuous perception, continuous optimization, and continuous service. Siemens and Nvidia will expand their cooperation in early 2026 and explicitly propose to jointly build an industrial AI operating system; Sandvik has embedded AI into manufacturing software, mining equipment, and operations services to form a customer-oriented digital solution layer. This indicates that the world's leading companies are no longer just competing for the devices themselves, but for the intelligent and service layers above the devices.

Fifth, we cannot give up on the weaknesses at the bottom level, but our strategy should focus more on the "bottleneck layer" rather than spreading it out comprehensively. China cannot bypass the underlying issues of advanced chips, core industrial software, and critical design toolchains. But the optimal strategy is not to fully replicate all the underlying layers at the same time, but to identify the most critical and influential low-level links that affect industrial capabilities and make high-intensity breakthroughs, while using algorithm efficiency, scenario advantages, and industry data to amplify the boundaries of available capabilities.

Sixth, fill the gap in the foundational layer with an open source ecosystem. Open source is not a temporary solution, but should be seen as a strategic choice. It lowers the threshold for acquiring basic abilities and shifts the focus of competition from "whose model is stronger" to "whose application is deeper", which is precisely China's strength. The Implementation Opinions of the "Artificial Intelligence+Manufacturing" Special Action clearly propose to "build a globally leading open source ecosystem", which is not only a technical arrangement, but also an arrangement for industrial ecological game.

Finally, enter the standards and rules layer as early as possible, rather than waiting for the technology to mature before participating. If China only values applications and neglects rules, even if its industrial scale continues to lead in the future, it may still suffer losses in value distribution. The data format, interface protocol, security authentication, model evaluation, and trustworthy system of industrial AI are still in the early stages. If China wants to truly transform its manufacturing advantages into global competitiveness, it cannot only be the largest application market, but also strive to become an important rule participant. NIST has listed AI standard work as a formal direction, and the EU continues to emphasize governance and regulation, which precisely shows that rules are not a subsidiary issue.

In summary, the difference between the three paths is that the United States is more like a "smart core first, then industrial spillover"; Europe is more like 'industrial system and regulatory framework first, then promoting widespread adoption'; A more reasonable path for China is to first transform the complete industrial system and large-scale scenarios into industry intelligence capabilities, and then reverse engineer the upward movement of platforms, software, standards, and rules.

The problems in different industries are not the same

China's manufacturing industry is not a homogeneous whole. Different industries are in completely different positions, so we cannot use the same mindset to talk about 'AI reshaping China's manufacturing industry'.

For industries that are already leading or have strong competitiveness, such as new energy vehicles, power batteries, consumer electronics, home appliances, communication equipment, and some construction machinery, the core issue is not how to catch up, but how to prevent staying at the execution level in the new round of value distribution. Their practical advantages mainly come from scale efficiency, supply chain integration, and rapid iteration, but in the future, if the focus of product value continues to shift towards software, data, and intelligence layers, relying solely on better hardware and faster manufacturing is not enough to ensure that value is not intercepted by the upward moving platform layer. BYD has been continuously strengthening its intelligent route in recent years, while Haier Smart Home has clearly proposed in its annual report to use big models to promote operations and product applications. These actions indicate that leading industries are beginning to realize that the next stage of competition is not just about continuing to strengthen hardware, but also about developing intelligent capabilities on top of hardware.

For industries that are rapidly catching up but not yet leading, such as commercial aircraft, high-end medical equipment, semiconductor equipment, and precision instruments, the core issue is not to replicate the old path of the leaders, but to determine whether AI has changed the path of catching up. If the bottleneck is mainly cognitive accumulation problems such as process experience, design iteration, and testing verification, AI may accelerate catch-up; If the bottleneck is mainly due to physical infrastructure issues such as core equipment, critical materials, etc., the role of AI is more to assist rather than replace. AI is an accelerator, not a shortcut.

For industries where there is still a significant intergenerational gap, such as advanced process semiconductors, core EDA, high-end aircraft engine hot end components, top precision optics, and some biopharmaceutical core processes, the problem cannot be understood as "comprehensive catch-up". The more realistic question is: which nodes may be rewritten by AI, and which nodes can only rely on long-term basic research and physical experiments to advance in the short term. What is truly important here is to identify the entry points where AI is most likely to bring about non-linear breakthroughs, rather than treating AI as a shortcut.

For industries with unique global positions, such as rare earth processing, ultra-high voltage, high-speed rail system integration, and large-scale infrastructure construction, the problem is neither catching up nor defending, but how to use AI to transform unique positions into higher value, stronger lock-in, and greater international influence. In other words, what these industries should truly do is not just to maintain their market share, but to elevate their unique position to a higher level of solution capability and rule influence.

The biggest risk is not that the direction is unclear, but that the old model is still effective

The biggest tension currently facing China's manufacturing industry is not the lack of direction, but the fact that the old model is still creating real returns, so the new direction is naturally prone to being delayed.

The successful model in the past can be summarized as: based on large-scale physical execution capability, with cost, efficiency, and response speed as the main competitive means, and continuous investment and scale expansion as the main growth methods.

Today, this model has not failed. The scale of the manufacturing industry still ranks first in the world, the advantages of the industrial chain are still obvious, and the international competitiveness of many industries has not been weakened. Because the old model is still effective, resource allocation, organizational attention, and strategic discussions naturally tend to continue amplifying existing advantages.

But what really needs to be vigilant is precisely this. The subject who misses the paradigm shift is often not because they haven't seen the new direction, but because the old model is still successful, so the investment in the new model always seems "not urgent enough". If the international standards for industrial AI are dominated by others in the next three to five years, even if Chinese manufacturing leads in scale, it may still be in a passive position in terms of interface, certification, and evaluation rules, which is a regulatory lock in; If a few platform enterprises take the lead in forming strong network effects and data flywheels in key industries, the cost of catching up with later players will rapidly increase, which is market-oriented lock-in; If industrial data infrastructure does not make substantial progress in critical stages, China's most unique data endowment may remain in a "resource state" for a long time and cannot be transformed into a capability advantage, which is infrastructure lock-in. Once these three types of locks are stacked, the judgment that the window is still there will quickly become invalid.

If resources in the next three to five years are still mainly invested in amplifying old advantages without synchronously moving up to the capability layer, then the current advantages may become path dependent in the next stage.

What really needs to enter the high-level resource allocation agenda are these five issues

If we summarize the entire text into five questions that high-level officials really need to answer.

Firstly, industrial data infrastructure should be regarded as a new type of infrastructure for the manufacturing industry in the AI era, rather than ordinary digital supporting facilities. Data classification and grading, rights confirmation and use, secure circulation, interface standards, high-quality datasets, and trustworthy exchange mechanisms are no longer matters for technical departments, but rather the fundamental issues that determine whether China's manufacturing industry can form a capability advantage.

Secondly, it is necessary to accelerate the cultivation of "industry AI translators". China does not lack general AI capabilities, nor does it lack industry leaders. What is truly scarce are intermediate level entities that can transform general model capabilities into usable productivity for specific industries. Siemens and Sandvik are worth noting not because they have "stronger AI", but because they can weave general technology, industry knowledge, and customer problems into practical industrial solutions. It is precisely this layer that China needs to fill in.

Thirdly, the adjustment of talent structure should be elevated to an equally important position as technology investment. The most scarce in the future will be composite talents who understand both the boundaries of AI capabilities and industrial scenarios and process constraints, as well as the ability to participate in data governance, model deployment, and business restructuring. The special action proposes to promote 500 typical application scenarios, which should not only undertake the function of technical verification, but also the function of cultivating composite talents.

Fourthly, some leading enterprises should be encouraged to move from "manufacturing execution centers" to "industry capability nodes". Not all enterprises need to become platforms, but without a group of entities with external influence in industrial software, industry models, intelligent product platforms, and standard shaping, the overall position of China's manufacturing industry will be difficult to truly move up.

Fifth, it is necessary to proactively take the lead in international standards and regulations for industrial AI as soon as possible. As the world's largest manufacturing country and one of the most active markets for AI applications, China has the conditions and necessity to actively enter the early game in data formats, interface protocols, security certification, model evaluation, and ethical standards. Rules are not a subsidiary issue, but a high-level issue in the industrial ecosystem.

Equally unavoidable are these four types of risks

The first is the structural imbalance of "strong application but still needs to be supplemented at the bottom layer". China has made significant progress in AI application speed, scene richness, and industry landing, but there are still shortcomings in underlying capabilities such as advanced chips, core industrial software, and high-end design toolchains. If this imbalance persists for a long time, it may lead to a pattern of "prosperity at the top and restraint at the bottom".

The second is that the difficulty of data governance may be underestimated. Rich industrial data does not automatically equate to intelligent advantages. Long term construction is needed for data ownership, security responsibilities, trust mechanisms for cross enterprise circulation, quality standards, and value distribution. If these problems are not solved effectively, one of China's biggest potential advantages will still remain in a "resource state" and difficult to transform into a "capability state".

The third is the interference of AI narrative on the fundamentals of manufacturing. One of the most realistic risks is that a large amount of resources are absorbed by conceptual projects, and there is insufficient investment in long-term factors such as quality, process, delivery, organizational capabilities, talent development, and digital infrastructure. The foundation of manufacturing competition has not changed, and AI can only magnify the advantages or defects on the foundation, but cannot replace the foundation itself.

The fourth is the risk of time mismatch in AI investment returns. Manufacturing industry is not an Internet industry. The landing of AI in the manufacturing industry requires data preparation, system integration, process adaptation and organizational change, which are slow variables. For many small and medium-sized manufacturing enterprises, if they lack sufficient patience and resource buffering, interrupting the initial investment without generating returns can easily lead to a wavering judgment of the entire direction.

Conclusion

Overall, China's manufacturing industry still has a strong foundation in the era of abundant cognition, and indeed faces an important opportunity, but what is truly changing is the value structure of its advantages.

The key to the next decade is not just to continue strengthening physical execution capabilities and industrial scale advantages, but to accelerate the extension to key capability layers such as industrial data infrastructure, industry intelligent solutions, industrial software and model platforms, intelligent product layers, and rule standard layers on this basis. Moreover, this extension should not be a one size fits all approach: for already leading industries, the core task is to transform practical advantages into advantages in the intelligent and relational layers; The core task of catching up with the industry is to use AI to accelerate cognitive accumulation, without mistakenly thinking that it can replace basic research; For industries with intergenerational disparities, the core task is to identify key nodes where AI may rewrite the path; For industries with a unique global position, the core task is to transform this position into higher value lock-in and regulatory influence.

It should be acknowledged that the above analysis of competence level competition and path selection is based on several hypotheses that have not been fully validated. There is still considerable uncertainty regarding how deeply AI can penetrate the core manufacturing processes and whether the platform based competitive logic can be applied to the physical world. If the penetration rate of AI in the manufacturing industry is much slower than expected, or if the complexity of the physical world hinders the formation of platformization trends for a long time, some of the judgments in this article may need to be changed. However, even with these uncertainties, building a capability layer based on one's own endowment, while stabilizing the fundamentals and not being biased by AI narratives, is still a choice with a high margin of safety in most reasonable scenarios.

From this, it can be seen that the true core issue regarding the future of China's manufacturing industry is no longer whether it can continue to strengthen its position as the world's factory, but whether it can gradually become a key capability node and rule participant in the future industrial intelligence ecosystem while maintaining its position as the world's factory.

This is not a posture issue, but a resource allocation issue. The watershed of China's manufacturing industry in the future may not necessarily depend on how much more can be produced, but more likely on which layer can occupy in the new value distribution structure.

The window still exists, but what truly makes sense is not knowing that the window is still open, but knowing where resources should be moved from.

(The author is the former President of Roland Berger China and Vice Chairman of Accenture Greater China)