Artificial intelligence is moving from screens and cloud services into machines, factories and other physical environments. Mitsubishi Electric plans to put that transition at the center of its CEATEC 2026 showcase, demonstrating how AI can combine with industrial data and decades of field expertise to help machines perceive changing conditions, make decisions and act more autonomously.
The next phase of industrial AI may be less about generating text and more about teaching machines how to operate in the physical world.
That is the direction Mitsubishi Electric will emphasize at CEATEC 2026, where the Japanese industrial technology company plans to showcase technologies and partner initiatives under the theme “Creating the Future of Industry with Physical AI.”
The event takes place October 13–16 at Makuhari Messe in Chiba, Japan.
Mitsubishi Electric’s focus reflects a broader shift in artificial intelligence. Large language models have demonstrated how AI can process and generate information, but industrial environments present a different problem. Machines must respond to changing physical conditions, incomplete information and unpredictable events while meeting strict requirements for safety and reliability.
That is the challenge behind physical AI.
Unlike software agents operating entirely within digital environments, physical AI systems interact with factories, machinery, robots and other real-world infrastructure. They need to perceive their surroundings, interpret sensor information, understand operational conditions and execute actions within defined safety boundaries.
Mitsubishi Electric plans to demonstrate that approach through hands-on exhibits, demonstrations and presentations at CEATEC, using its Serendie digital platform as part of the company’s broader technology architecture.
The company describes Serendie as a digital platform supporting its effort to connect field knowledge, data and AI. The strategy is closely tied to what Mitsubishi Electric calls “Circular Digital-Engineering”—an approach that combines accumulated engineering and operational knowledge with digital technologies to create new products and services.
The idea addresses one of the biggest challenges facing industrial AI adoption: enterprises often have enormous quantities of operational knowledge, but that knowledge is distributed across equipment, engineering teams, maintenance records, production systems and individual employees.
Turning that experience into machine-readable information could make AI systems more useful in physical environments.
For example, an industrial AI system might combine sensor data from manufacturing equipment with historical maintenance information and operational rules. Rather than simply identifying an anomaly, the system could potentially determine what is happening, evaluate possible responses and recommend or execute an appropriate action.
That progression—from detection to understanding to action—is central to physical AI.
Mitsubishi Electric says its CEATEC demonstrations will show how AI can perceive, understand and act safely in real-world environments while adapting to uncertainty and variability.
The emphasis on uncertainty is important.
Factory environments are not controlled software environments. Components wear out. Materials vary. Temperatures change. Production lines encounter unexpected conditions. Robots interact with objects that may not always appear exactly as expected.
A physical AI system therefore needs to be robust to variation rather than simply optimized for a fixed set of inputs.
This differentiates industrial AI from many early enterprise generative-AI deployments. A chatbot can produce an imperfect answer that a user corrects. An industrial control system making the wrong physical decision can damage equipment, disrupt production or create a safety incident.
Consequently, industrial AI requires a much tighter connection between AI capabilities, deterministic control systems, safety engineering and human oversight.
Mitsubishi Electric’s focus on autonomous factories illustrates where the company believes this technology is heading.
Fully autonomous manufacturing remains a difficult goal, but factories are already becoming increasingly automated through robotics, industrial controllers, machine vision, digital twins and predictive analytics. Physical AI could potentially connect these previously separate systems, allowing machines and production processes to respond dynamically to changing circumstances.
The concept also fits into the wider industrial strategies of companies such as NVIDIA, Siemens, Microsoft and Amazon, which are developing combinations of AI infrastructure, simulation, robotics and industrial software.
NVIDIA has pushed the concept of AI-powered physical systems through its robotics and simulation platforms, while industrial companies such as Siemens are integrating AI into manufacturing and engineering workflows. Cloud providers are also building AI services that can connect models to enterprise data and applications.
The emerging competitive question is therefore not simply who has the most capable AI model.
It is who can connect AI models to physical systems, industrial data and reliable control mechanisms.
For Mitsubishi Electric, that advantage may come from its installed base and accumulated field knowledge. The company operates across factory automation, power systems, building systems, transportation and other industrial markets. Those environments generate the operational data and domain expertise that physical AI systems need.
That creates a potential feedback loop: equipment generates data, engineers contribute domain knowledge, AI analyzes operational patterns and the resulting insights can be fed back into engineering and automation.
The challenge is making that loop secure and useful without creating unacceptable operational risks.
Mitsubishi Electric’s “co-creation” emphasis is therefore significant. Rather than presenting physical AI as a standalone product, the company plans to demonstrate initiatives involving internal and external partners. That suggests industrial AI development is increasingly becoming an ecosystem exercise involving equipment manufacturers, software companies, system integrators and customers.
The strategy also highlights an important reality about autonomous factories: autonomy will likely arrive incrementally.
Factories are unlikely to move overnight from human-operated production lines to fully independent systems. More realistic adoption paths involve individual processes becoming increasingly autonomous—predictive maintenance, inspection, scheduling, material handling and process optimization—before broader factory-level coordination becomes possible.
Physical AI could provide the connective layer between those capabilities.
Mitsubishi Electric’s CEATEC 2026 showcase will therefore be less about AI as a standalone technology and more about how artificial intelligence can become embedded in the physical infrastructure of industry.
If the approach succeeds, the factory of the future may not simply have more robots or more sensors. It may have systems capable of understanding what is happening around them, adapting to conditions and coordinating actions with increasingly limited human intervention.
That would represent a meaningful shift in industrial automation—from programmed machines toward AI-enabled physical systems.
Market Landscape
The physical AI market is emerging at the intersection of industrial automation, robotics, edge AI, digital twins and generative AI.
Traditional industrial automation relies heavily on deterministic rules and programmed logic. AI introduces probabilistic reasoning and perception, allowing machines to operate in environments where every possible condition cannot be explicitly programmed.
That creates opportunities across manufacturing, logistics, energy, transportation and infrastructure.
Major technology companies are moving into the space from different directions. NVIDIA is supplying AI compute, simulation and robotics platforms. Microsoft and Amazon are integrating AI with enterprise and cloud infrastructure. Industrial technology companies such as Siemens and Mitsubishi Electric bring decades of expertise in automation, controls and physical systems.
The resulting market is unlikely to be dominated by one technology layer.
Instead, physical AI will require an integrated stack spanning sensors, edge computing, industrial controls, AI models, simulation, connectivity, robotics and safety systems.
For manufacturers, the business case will depend on measurable outcomes: fewer unplanned shutdowns, higher production throughput, better quality control, lower energy consumption and greater flexibility when production conditions change.
That makes the conversion of existing industrial data and field knowledge into AI-ready information particularly important.
Top Insights
- Mitsubishi Electric is showcasing physical AI at CEATEC 2026, focusing on machines that can perceive, understand and safely act in changing industrial environments.
- Serendie provides a digital foundation for connecting Mitsubishi Electric’s field knowledge, operational data and AI technologies across industrial applications.
- Autonomous factories are a major target, with physical AI potentially connecting robotics, sensors, industrial controls and production intelligence into coordinated systems.
- Industrial AI differs from generative AI because physical actions require greater emphasis on safety, reliability, real-time response and adaptation to unpredictable conditions.
- The emerging physical AI ecosystem includes industrial companies and technology providers such as Mitsubishi Electric, NVIDIA, Siemens, Microsoft and Amazon.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
