Physical AI, the critical branch of artificial intelligence responsible for controlling robots and industrial machinery in real-world environments, is currently navigating a complex hierarchical landscape. While industry giants like OpenAI and Google focus on scaling multimodal foundation models, and Nvidia dedicates its efforts to building robust platforms and tools for physical AI development, a third, arguably more pragmatic, camp has emerged. Industrial manufacturing stalwarts such as Hitachi and Siemens are championing the argument that true machine navigation of the physical world necessitates a profound foundational understanding of that world itself. This perspective is rapidly transitioning from strategic discussions in boardrooms to practical implementation on factory floors, as exemplified by Hitachi's recent insights shared with Nikkei Asia.

Kosuke Yanai, deputy director of Hitachi's Centre for Technology Innovation-Artificial Intelligence, articulates a clear distinction between theoretical and viable physical AI. He emphasizes that "Physical AI cannot be implemented in society without a systematic understanding that begins with foundational knowledge of physics and industrial equipment." Hitachi's compelling advantage lies in its extensive repository of this foundational knowledge, meticulously gathered over decades through its involvement in developing critical infrastructure like railways, power systems, and intricate industrial control systems. This deep expertise is underpinned by advanced technologies such as thermal fluid simulation, capable of modeling gas and liquid behaviors, and sophisticated signal-processing tools for real-time equipment condition monitoring. Yanai refers to this as the "engineering foundation" that supports Hitachi’s comprehensive knowledge in product design and control logic construction.

Central to Hitachi's overarching physical AI strategy is the Integrated World Infrastructure Model (IWIM), conceptualized as a mixture-of-experts system designed to seamlessly integrate diverse specialized models and data sets. Although the IWIM itself remains in the concept verification phase, its underlying principles are already yielding tangible results in real-world applications. A notable collaboration with Daikin Industries has led to the deployment of an AI system capable of diagnosing malfunctions in commercial air-conditioner manufacturing equipment. This intelligent system, meticulously trained on extensive equipment maintenance records, procedure manuals, and detailed design drawings, can now accurately identify the likely failing component upon detecting an anomaly – a level of operational intuition traditionally held only by highly experienced human engineers. Similarly, in partnership with East Japan Railway (JR East), Hitachi has developed an AI solution that pinpoints the root cause of control device malfunctions within Tokyo's vast metropolitan railway traffic management system, subsequently assisting operators in formulating swift response plans. In a network where delays can impact millions of daily commuters, the capability to accelerate fault diagnosis holds immense operational significance.

Hitachi's commitment to advancing physical AI is also evident in its prolific research and development output. In December 2025, the company presented findings from two significant projects at ASE 2025, a leading software engineering conference, directly addressing a persistent challenge in industrial AI: the considerable time and effort required for writing and adapting control software. In the automotive sector, Hitachi, alongside its subsidiary Astemo, pioneered a system leveraging retrieval-augmented generation (RAG) to automatically produce integration test scripts for vehicle Electronic Control Units (ECUs). By intelligently drawing from hardware-specific API information and frontline engineering knowledge, this technology demonstrated a remarkable 43% reduction in integration testing man-hours during a pilot involving multi-core ECU testing, compared to traditional manual execution. Furthermore, in the logistics domain, Hitachi developed innovative variability management technology. This approach modularizes robot control software into reusable components, structured around a robot operating system (ROS). By pre-mapping environmental variables and operational requirements across diverse warehouse settings, the system empowers operators to adapt robotic picking-and-placing workflows to new products or facility layouts efficiently, without the need for extensive software rewrites from scratch.

A consistent and paramount theme across all of Hitachi’s physical AI endeavors is an unwavering emphasis on safety guardrails. This isn't merely a box-ticking exercise for compliance but rather an intrinsic engineering constraint fundamentally woven into the very design of their systems. Yanai underscored this commitment to Nikkei, explaining that Hitachi integrates its robust control and reliability technology, honed through years of social infrastructure development, to rigorously prevent AI outputs from veering outside human-approved operating parameters. This comprehensive approach encompasses multi-layered safeguards: input validation to meticulously filter out unsuitable training data, output verification to guarantee that machine actions pose no threat to people or property, and continuous real-time monitoring of the AI model itself for any operational anomalies. This distinction is critical: physical AI systems operate and fail in the real world, not in simulated sandboxes. Consequently, the stakes associated with an AI controlling vital railway signaling or intricate factory robotics are fundamentally different and far more critical than those governing a conversational chatbot.

To match its ambitious physical AI objectives, Hitachi is also bolstering its infrastructure. Hitachi Vantara, the group’s dedicated data and digital infrastructure arm, is proactively adopting NVIDIA’s cutting-edge RTX PRO Servers, which are built upon the powerful RTX PRO 6000 Blackwell Server Edition GPU. This advanced hardware is specifically engineered to accelerate demanding agentic and physical AI workloads. These servers are being synergistically paired with Hitachi's iQ platform to construct sophisticated digital twins – virtual replicas of physical systems – capable of simulating a vast array of real-world phenomena, from grid fluctuations to complex robotic motions, at an unprecedented scale. Concurrently, the IWIM concept is designed to bridge Nvidia’s open-source Cosmos physical AI development platform with specialized Japanese-language LLMs and visual language models through the Model Context Protocol (MCP). This protocol essentially serves as a robust framework to seamlessly stitch together the diverse models, simulation tools, and industrial datasets that sophisticated physical AI systems inherently require.

While the broader race in physical AI remains fiercely contested and far from settled, Hitachi's unwavering position — that profound domain expertise and rich operational data are as fundamentally important as advanced model architecture — is becoming increasingly undeniable. This argument is now robustly validated as its real-world deployments with key partners like Daikin and JR East demonstrably showcase the tangible value and practical efficacy of such deep-seated expertise in action.