The World Economic Forum’s 17th Annual Meeting of the New Champions, also known as the 2026 Summer Davos Forum, was held in Dalian from June 23 to 25. On June 24, Yu Feng, president of Honeywell China, spoke with National Business Daily (NBD) in an exclusive interview.

As one of the earliest foreign companies to enter China, Honeywell entered the Chinese market in 1935 and has been deeply rooted in the country for more than 90 years. Yu said China’s vast and complex industrial system is a real-world testing ground for industrial AI. AI solutions that can be successfully deployed and operate reliably in China’s complex industrial environment are likely to have the potential to be replicated across sectors and scenarios, and may become benchmark solutions for broader adoption.

When discussing AI in the industrial sector, Yu said that while general-purpose large models pursue broad applicability, industrial AI must adhere to the fundamental principles of accuracy, reliability and explainability. The industrial world does not blindly chase what is merely “advanced.” It only accepts technologies that have been validated and can be trusted.

Yu Feng Photo/provided to NBD

Industrial AI Must Adhere to the Fundamental Principles of Accuracy, Reliability and Explainability

NBD: In your view, what is the biggest difference between industrial AI and general-purpose large models? What special requirements do industrial scenarios place on AI?

Yu Feng: If general-purpose large models pursue broad applicability, industrial AI must adhere to the fundamental principles of accuracy, reliability and explainability. When AI is embedded into real production equipment and operational processes, machines evolve from execution tools into “digital workers” that can listen, see, think and collaborate. They need to combine historical and real-time data and turn it into actionable operational insights, rather than merely generating text or images.

Industrial scenarios place far higher demands on AI than typical consumer internet applications do First, there is an extreme requirement for safety and determinism. Industrial systems do not blindly trust a technology simply because it is advanced. They only accept solutions that have undergone long-term validation and whose accountability can be traced. Therefore, AI decision-making must be transparent and explainable.

Second, industrial AI depends on high-quality data and deep industry understanding. Industrial data has long been isolated and locked away, while industry expertise is highly fragmented. Collecting, cleaning and integrating such data is the foundation for AI to effectively learn industrial patterns. At the same time, AI must also understand process logic, equipment behavior and production constraints. The same set of data may carry completely different physical meanings under different processes.

Finally, industrial AI requires clear explainability and accountability. Traditional control systems such as PLCs, or programmable logic controllers, operate based on preset logic, with results that can be verified and traced. AI algorithms, however, are probabilistic by nature. Once an error occurs, it is often difficult to explain the cause or define responsibility. Such uncertainty is difficult to accept in industrial scenarios.

Industrial AI Cannot Pursue Digitalization for Its Own Sake. It Must Be Closely Integrated with Real Business Needs

NBD: What do you see as the biggest obstacle to the large-scale application of industrial AI today?

Yu Feng: Based on Honeywell’s global industrial experience, especially our practice in the Chinese market, I believe there are currently three interrelated bottlenecks in scaling industrial AI deployment.

First, data silos and the lack of data governance are the primary obstacles to the large-scale application of AI. Data lies at the core of AI, but in many companies, production and quality data is often scattered across different systems and departments, forming data silos. Many companies have huge volumes of data, yet how to identify and process key data remains a very challenging issue. AI’s value can only be truly unlocked when data can be brought to the production site and put into real-time use.

Second, the shortage of skilled professionals and cross-disciplinary talent constrains AI implementation and scaling. Many companies are facing both talent attrition and structural skills shortages. As experienced workers gradually retire and frontline job mobility increases, output can vary significantly across shifts. More importantly, interdisciplinary talent that understands both AI technology and industrial site operations is extremely scarce. As a result, even when advanced AI tools are introduced, companies often struggle to apply them deeply and iterate on them continuously. How to enable frontline workers to master and make good use of AI tools is also a key issue for scaling industrial AI.

Third, application scenarios are often not clearly identified, and value is difficult to quantify. Industrial AI cannot pursue digitalization for its own sake. It must be closely integrated with real business needs. If business departments and technical departments have different understandings of AI’s capabilities, it becomes difficult to demonstrate the value of investment to management, making it hard to secure sustained resource support. Corporate transformation must be built on tangible benefits. These benefits may be economic, environmental or social. Only when the ROI of AI is clearly visible and measurable will companies have the incentive to expand AI from pilots to the entire production network.

China’s Vast and Complex Industrial System Is a Real-World Testing Ground for Industrial AI

NBD: China’s manufacturing system is large in scale, complete across the industrial chain, and highly complex in terms of industry scenarios. What does this complexity mean for the deployment of industrial AI?

Yu Feng: China’s manufacturing sector has numerous specialized segments, complex operating conditions, and both new and legacy production lines. Generic solutions are difficult to apply directly. This is also a core reason why many technologies struggle to adapt to Chinese manufacturing and scale up after entering the market.

From the perspective of long-term industrial development, however, this highly complex scenario advantage can promote the iteration of industrial AI technologies and help form comprehensive solutions with broad applicability. China’s vast and complex industrial system is a real-world testing ground for industrial AI. Diverse production scenarios, complex operating variables and different operational logic across industries can continuously refine the adaptability, stability and fault tolerance of technologies. AI solutions that can be successfully deployed and operate reliably in China’s complex industrial environment are likely to have the potential to be replicated across sectors and scenarios, and may become benchmark solutions for broader adoption.

NBD: Honeywell has long upheld its “East for East” strategy and emphasized localized innovation. In the era of AI and industrial autonomy, has this strategy taken on new meaning?

Yu Feng: In the past, when we talked about localization, we focused more on local manufacturing and supply chains. Now, however, the “localization” we emphasize has been comprehensively upgraded. From deep insight into market demand to product definition, technology research and development, and manufacturing delivery, the entire value chain must be deeply rooted in China.

The transformation of the R&D model is very important. In the past, we mainly made localized adjustments to mature overseas products. Today, however, China’s digital and green transformation is advancing rapidly, and many scenarios are so complex and unique that overseas experience cannot simply be copied. We must move from “local adaptation” to “from zero to one” innovation. Starting from customer needs, we must continue to iterate, allowing innovation to take root, grow and evolve in China, and create solutions that truly fit the Chinese market.

As AI and industrial technologies become deeply integrated, the capabilities of any single company are far from enough. We need to turn innovation from an “internal corporate loop” into an “industrial ecosystem loop.”

In the future, industrial AI deployment pathways that we validate in China, as well as proven low-carbon and intelligent manufacturing solutions, can be replicated in more markets, turning China’s local innovation capabilities into shared experience for global industrial upgrading.