Hitachi Digital knows which valve to check.

A factory has dozens of valves. An AI assistant, asked about a breakdown, consults its records and concludes that there is a valve problem. This sounds promising until somebody has to find the valve. In a Hitachi researcher’s account of early tests, retrieval-based systems could handle familiar failures, then stumble when the fault looked different. A useful answer needed more than a fluent explanation. It needed a map of the machinery.

A manual cannot tell you everything

Daikin, the air-conditioning manufacturer, began trial operations with Hitachi in April 2025 on an agent for factory equipment diagnostics. The approach converted equipment drawings into knowledge graphs, representing components and their relationships. It combined those maps with operating records and the diagnostic steps experienced engineers use. The purpose was to investigate unfamiliar faults, where simply retrieving a previous answer could lead the maintenance crew astray.

The distinction is easy to miss at a software demonstration. Finding a relevant paragraph and tracing a failure through interconnected equipment can look similar on a screen. On the plant floor, they lead to different actions. A system needs to know which component affects which other component, and what evidence would distinguish one possible cause from another. That requires engineering knowledge someone has taken the trouble to encode.

A company built to make the handoffs work

Hitachi Digital exists to bring those different kinds of knowledge together. Launched in April 2022, it expanded and renamed an existing Hitachi digital holding organization. Its original home was North America, with a mandate to strengthen the group’s worldwide digital strategy and collaboration across businesses. The origin matters: this was an industrial group reorganizing expertise it already possessed, with Jun Taniguchi as inaugural CEO.

Its own description assigns distinct jobs to the cast. Hitachi Vantara supplies data infrastructure and platforms. GlobalLogic brings software engineering and design. Hitachi Digital Services handles cloud services and IT/OT integration. Energy, rail, and automation businesses contribute experience with physical operations. Hitachi Digital’s proposition is that a customer should be able to combine these capabilities around one operating problem.

This also explains the market position. Large consultancies and systems integrators compete for enterprise transformation budgets; industrial technology vendors compete for factory and infrastructure work. Hitachi straddles those purchasing decisions. Its argument depends on access to the people who understand the assets, alongside people who can build and run software. Customers still have to judge how well the teams actually work together.

The factory is also the classroom

Consider the Singtel collaboration announced in June 2024. The plan paired Hitachi’s industrial AI applications with Paragon, Singtel’s orchestration platform for networks, edge computing, and cloud. Deployment would begin at Hitachi’s Santa Clara research labs, followed by a US factory pilot. Quality assurance, safety, training, and maintenance were among the intended tests. The immediate task was making the applications and connectivity work together.

That sequence is revealing. Hitachi calls its internal testing approach “Customer Zero.” In the Google Cloud alliance, maintenance and inspection work at Hitachi Power Solutions provided a setting for technical validation with Gemini Enterprise. The June 2026 expansion emphasized engineers working directly with customers to uncover site constraints, connect operational equipment with business systems, and incorporate experienced workers’ knowledge. A successful lab demonstration leaves plenty of work for the people on site.

The customer problems can be much larger than a factory fault. In June 2025, Southwest Power Pool and Hitachi announced work to accelerate the studies needed before new generation can connect to the electricity grid. The project combined simulation algorithms, infrastructure, engineering, and energy expertise. Its initial goal was an 80% reduction in analysis times. That figure was a project objective, so it belongs in a different column from an achieved saving.

The price of joining things up

The commercial offering spans consulting, engineering, implementation, infrastructure, and ongoing services delivered through specialist group businesses. Buying it means defining a project and its operating responsibilities. For a customer, the useful economic questions concern downtime avoided, maintenance effort, computing expense, and the continuing cost of integration. A persuasive demonstration does not settle any of those questions by itself.

The investment commitments are substantial, but their scope matters. In June 2024, Hitachi and Microsoft described a projected multibillion-dollar collaboration over three years. Hitachi also stated a group-level plan to invest ¥300 billion, then about $2.1 billion, in generative AI during FY2024. Those announcements describe strategic spending and collaboration. They cannot tell a manufacturer what connecting its particular production line will cost.

Hitachi has acknowledged the difficulty behind the buying decision. Writing about customer conversations and stalled pilots, then-president and COO Gajen Kandiah concluded that “the issues were more than technical.” His proposed principles started with measurable outcomes, continued with systems built for particular industries and data, and included responsible, reliable operation. The implication for buyers is practical: decide what should improve before selecting the machinery that promises to improve it.

“the issues were more than technical.”Gajen Kandiah · Our Position on AI and GenAI

Teach the system where to look

The portfolio now extends further. September 2026 brought HMAX Data Center, Cyber, Data Fabric, and AI Operations. Data Fabric structures operational information and expert knowledge for AI use; the wider expansion addresses security and continuous operation. The direction follows the original factory problem: useful reasoning depends on organized knowledge, reliable systems, and support after deployment.

A reader can copy the discipline without copying the corporate structure. Pick one costly operating decision. Record the baseline. Ask an experienced technician to explain the checks and exceptions. Validate against unfamiliar cases. Where equipment data is unreliable, expertise is unavailable, or site validation is skipped, the same approach has weaker foundations. Start with the valve somebody must actually find.