Future industrial artificial intelligence (AI) will be a comprehensive integration of data, algorithms, hardware and industry knowledge. See five major trends of industrial AI.
Industrial AI insights
- Artificial intelligence for industrial applications, industrial AI, has five major trends advancing development and use.
- Data governance, rise of industrial AI agents for full process operation models and the accelerated iteration of industrial large language models are helping industrial AI.
- Also helping industrial AI deep integration of IT and OT and a return to rational application concepts with autonomous closed-loop operation as the core goal.
As global manufacturing accelerates its shift from the Industrial Internet to the Industrial Intelligence Internet, AI has transcended its role as a mere auxiliary tool and evolved into the “brain” and “nervous system” of industrial production systems. However, challenges such as data silos, implementation difficulties, unexplainable algorithms, and persistently high application costs remain widespread pain points constraining the large-scale adoption of industrial AI.
On June 3, the inaugural 2026 Industrial AI Summit was held in Shanghai (Figure). Dozens of enterprises and institutions participated, including Emerson, Phoenix Contact, PTC, Siemens and Supcon. They shared cutting-edge practices covering data governance, industrial agents, platform architecture, hardware infrastructure and scenario-based deployment, outlining the current development landscape and future technological evolution of industrial AI. Five distinct core technological trends have emerged.
Systematic, comprehensive data governance: Solidify the industrial AI foundation
Data constitutes the core production factor of industrial AI, yet fragmented, isolated heterogeneous data remains the primary barrier to technological deployment. The summit reached a broad industry consensus that engineered, standardized full-domain data governance is imperative.
The China Academy of Information and Communications Technology (CAICT) proposed a three-layer ontology framework and XBS multi-dimensional decomposition structure. It processes unstructured industrial data layer by layer through semantic, kinetic and dynamic ontologies, converting raw data into machine-readable, iterable knowledge assets. Meanwhile, it establishes a traceable, highly rigorous framework for knowledge modeling of complex industrial systems, fundamentally eliminating data silos.
In parallel, building AI-native data systems has emerged as a new development direction. PTC proposed revamping the full product lifecycle data chain via data graphing and ontology modeling to underpin high-end AI applications, such as generative design and intelligent change analysis. This transforms industrial data from simple storage assets into core drivers of innovation across R&D, production and operations and maintenance.
Comprehensive rise of industrial AI agents: Reconstructing full-process operation models
Industrial agents emerged as the most discussed technological track at the summit, marking a pivotal shift of industrial AI from isolated functional tools to system-level intelligence enabling autonomous cross-process collaboration. Multiple enterprises launched targeted agent products covering full scenarios including automation engineering, production control, R&D design and supply chain management.
Baosight Software developed the “Skywalker” AI agent to address low efficiency in industrial automation engineering. It automatically generates technical schemes, compiles PLC codes and conducts automated testing. Leveraging a “cloud training, edge inference, on-site control” architecture, it delivers real-time intelligent equipment regulation, integrating accumulated industrial domain expertise with large model reasoning capabilities.
Siemens’ AI Engineering Agent seamlessly connects mainstream industrial software platforms, generating PLC codes and human-machine interfaces (HMIs) via natural language queries and drastically lowering barriers to engineering development.
LinkCrux AI targets discrete manufacturing with executive AI digital employees that automate repetitive manual workflows such as order entry, design change synchronization and supplier quotation comparison. These agents bridge business gaps between disjointed systems with tangible, visible implementation outcomes.
Industry experts widely predict that a cross-domain interconnected ecosystem of industrial agents will take shape in the coming years, enabling continuous self-learning of AI through closed-loop feedback mechanisms.
Accelerated iteration of industrial large language models
General large language models struggle to meet the rigorous working condition requirements of industry. Vertical domain-specific large models have become the mainstream of R&D, with companies in different tracks creating differentiated model products based on business characteristics.
Supcon, a provider of process manufacturing, launched the Time Series Pretrained (TPT) model built on a Mixture-of-Experts (MoE) architecture. Optimized for analyzing industrial time-series data such as equipment operation curves and quality fluctuations, it integrates process mechanism knowledge to deliver process anomaly early warning, equipment health assessment, parameter optimization and soft-sensing capabilities. It paves the way for autonomous process factories, establishing fully self-decision-making production systems analogous to autonomous driving.
Siemens is investing in industrial foundation models paired with its Flamingo data analytics platform to realize adaptive control of process parameters, unifying analytical and generative AI capabilities. Unlike generic large models prioritizing universal generalization, specialized industrial large models center on real-world scenario deployment, tightly coupled with domain mechanisms, production rules and safety standards. Boasting improved stability and professional precision, they are increasingly deployed in core production links including formula optimization, fault prediction and quality control.
Deep integration of IT and OT: Open control platforms as the mainstream architecture
Traditional industrial control architectures struggle to be compatible with AI technology, and the barrier between information technology (IT) and operational technology (OT) severely restricts AI from penetrating the production floor. At this conference, open control platforms featuring software-hardware integration and multi-language compatibility became the key to breaking the deadlock.
Phoenix Contact’s PLCnext virtual control platform, relying on Linux container technology, opens high-level language interfaces such as C++ and Python while remaining compatible with traditional industrial programming standards. This allows AI inference models and real-time control tasks to run collaboratively on the same hardware. The built-in AI assistant can also generate engineering code via natural language, realizing full-chain predictive maintenance from data collection and model training to edge inference.
HollySys built an integrated management and control architecture relying on the XMagital platform, promoting the deep integration of IT and OT. It not only achieves “lights-out” production on certain lines but also uses AI to optimize material costs and simplify equipment programming, shortening programming tasks that previously took one or two days down to just 10 minutes. These open platforms break the closed nature of traditional industrial control systems, establishing an underlying architecture compatible with cutting-edge technologies and adaptable to long-term iteration, serving as the core carrier for implementing industrial AI at the on-site control layer.
Return to rational application concepts: Autonomous closed-loop operation as the core goal
As the hype around Industrial AI climbs, the industry has begun to discard gimmicks like “unmanned factories” and “fully black screens.” Rational implementation and value prioritization have become the consensus. Emerson pointed out that Industrial AI must possess determinism and inheritability to avoid the “hallucination” issues of general AI. It clarified that the core of Industrial AI is to build an autonomous closed-loop of “perception – decision-making – control,” distinguishing the different development stages of “autonomous operation” within a factory and “autonomous management” across enterprises, reminding the industry to view conceptual hype rationally.
Delta, drawing on its own manufacturing experience, proposed that smart factory construction must follow a progressive path of “lean, automated, digitalized and intelligent.” Guided by actual value, it leverages smart equipment, digital twins, and AI algorithms to optimize processes and monitor personnel operations, enabling equipment to self-monitor, self-adjust, and self-learn, thus completing the leap from “automation” to “intelligence.” Practices from multiple companies show that industrial AI implementation no longer pursues comprehensive intelligence blindly. Instead, it focuses on specific issues like process difficulties, efficiency bottlenecks and quality hazards. Entering through small scenarios and gradually replicating and promoting has become the mainstream implementation model.
Five trends broaden industrial AI applications, with edge AI hardware to help
Currently, industrial AI has completed its conceptual exploration stage and fully entered a new cycle of technological integration, scenario deepening and value realization. The five major trends—comprehensive data governance, industrial agents, dedicated industrial large language models, IT/OT integration platforms, edge AI hardware and rational implementation operations—are intertwined, jointly driving the transformation of manufacturing, with edge AI hardware to help.
In the future, industrial AI will no longer be a competition of one technology, but a comprehensive contest of data, algorithms, hardware and industry knowledge. As technologies continue to iterate and the ecosystem constantly improves, more implementable, replicable and lower-cost industrial AI solutions will emerge. These will continue to resolve the pain points of manufacturing transformation, helping the industrial sector truly achieve digital and intelligent upgrades.
Stone Shi is executive editor-in-chief, Control Engineering China; Edited by Mark T. Hoske, editor-in-chief, Control Engineering, WTWH Media, [email protected].
Keywords
Industrial AI, digital transformation
Consider this
Are you applying industrial AI as effectively as your competitors? When will you know?
You also might like
See a prior article by Stone Shi, Control Engineering China: How industrial AI deepens sustainability in 3 ways
