Over the last two years, manufacturers have learned how to move Physical AI from pilot to production. Energy and utilities are now at the same starting line. They face grid-scale demand pressure and they have the chance to skip the expensive lessons and go straight to what works.

Earlier this year, I wrote about how Agentic AI, Physical AI and sovereign AI are converging to reshape the plant floor. The World Economic Forum’s Global Lighthouse Network, which now spans 238 of the world’s most advanced industrial sites, found that most successful transformations combine multiple technology domains rather than deploying them in isolation. That finding matters even more today. The question I hear most often from energy and manufacturing leaders is simple: does the same playbook work on the other side of the fence, on the grid, across the pipeline network and at the substation?

The short answer is yes, with important adjustments. The architecture can travel, but the priorities have to change. And the timing is now urgent. A leading consulting firm’s 2026 Power and Utilities Industry Outlook projects that US peak electricity demand will grow by roughly 26% by 2035, with data center load alone potentially reaching 176 gigawatts, about five times the 2024 level. Grids must absorb that growth as they modernize and no credible plan to do both leaves AI out.

One maturity model, two different destinations

The three-stage journey I described for manufacturing applies just as well to energy and utilities: digital visibility first, then predictive insight, then autonomous action. The difference lies in what autonomy is asked to do.

In a factory, autonomy serves throughput, quality and yield. In energy and utilities, it sits much closer to the core of the business: keeping the grid stable, the pipeline intact and the lights on for communities with no tolerance for downtime. A manufacturing plant can absorb a short line stoppage. A substation or compressor station cannot absorb a comparable failure without cascading consequences.

The research reflects this. Gartner’s 2026 Hype Cycle for Physical AI describes a “trust hurdle” across asset-intensive industries. Production use of industrial AI agents is forecast to rise sharply by the end of the decade, yet operators are keeping humans and deterministic systems in control of physical actuation. For now, agents mostly serve in advisory and read-only roles.

This should shape how every energy and utility leader sequences investment. The first question is where AI can fail most safely. Grid monitoring, substation inspection, pipeline integrity checks and field service dispatch are natural early candidates. They strengthen human judgment on distributed, hard-to-reach assets without putting autonomous control in the critical path on day one.

Where the Physical AI playbook must flex

Three differences stand out when manufacturing lessons are applied to energy.

• Asset distribution: A factory is a bounded, instrumented environment. A utility network covers thousands of miles of distributed, aging and often unmonitored assets. Edge and connectivity strategies must be designed for intermittent connections, not built on the assumption that everything is always online.
• Regulatory exposure: Energy and utilities face far more safety, environmental and grid reliability regulations than most discrete manufacturers. Every autonomous recommendation needs an audit trail that a regulator will accept.
• Consequence of error: A missed defect on a production line is a quality cost. A missed anomaly on critical energy infrastructure is a safety and public trust event. The bar for acceptable accuracy is simply higher.

These concerns are real. A research study on utility resilience found that customer outage costs reached roughly $121 billion in a single recent year, nearly double the seven-year average, even as utilities raised spending on advanced resilience technologies by 38% over two years. The same study found that the biggest obstacle to scaling AI and geospatial intelligence across utility operations was not the technology. It was data silos and legacy systems that block enterprise-wide adoption.

IDC’s 2026 Energy and Utilities Industry-Specific Tech and Innovation Survey, covering more than 500 utilities worldwide, reaches a similar conclusion from another angle. Utilities have plenty of ambition. What they lack is the data architecture, talent and organizational readiness to turn pilots into production. A shared Physical AI architecture is designed to close exactly that gap.

None of this means energy and utility leaders should move more slowly than manufacturers did. They should move deliberately, sequencing use cases by risk and consequence rather than by which demo most impresses the boardroom.

The case for one architecture, not two

The stronger argument and the one I make to clients directly, is that manufacturing and energy organizations should resist building two separate Physical AI stacks just because their assets look different.

The underlying architecture is the same whether the asset is a stamping press or a substation transformer: edge compute, computer vision, sensor fusion, digital twins and an orchestration layer that lets agents act on physical assets. Building it once and configuring it twice costs far less than building it twice from scratch. It also gives a diversified industrial enterprise a single point of governance for AI acting on physical infrastructure. That governance question is one every board is starting to ask: who is accountable when an autonomous system makes a call on a physical asset and how is that decision reviewed?

The plants and grids that get this right will not be the ones with the most advanced model. They will be the ones with the clearest answer to: who trusts this system and why?”

What clients keep telling us

In conversations with energy and manufacturing leaders this year, one tension keeps surfacing. Leaders are excited about what is possible and equally firm that autonomy is not the goal in itself. Safety, reliability and uptime are the business imperatives. Autonomy is one way to get there and only where the risk profile allows.

That is why I believe the goal for any industrial or energy organization should be “more trusted,” not “more autonomous.” If a control room operator or field technician cannot question, explain or override a recommendation, they will not adopt it, however accurate the model. Explainability is how adoption happens, not a feature added at the end.

How HCLTech can help turn the playbook into operating reality

Clients do not need another AI point solution. They need engineering, operational technology, enterprise data and AI connected into a governed pathway that runs from visibility to prediction and, where the risk case supports it, to assisted action. HCLTech helps clients build that pathway across three layers.

1. Build the trusted data and asset foundation. Unify operational, geospatial, maintenance and enterprise data. Connect distributed assets through edge and IoT patterns and create digital representations of critical equipment. This gives AI the context it needs to support decisions across plants, grids, substations, pipelines and production facilities.
2. Industrialize high-value, human-in-the-loop use cases. For utilities, these include predictive asset performance, outage prediction and restoration, smart meter analytics, non-revenue loss detection, image and geospatial analytics and AI-assisted field service. For oil and gas, they include asset integrity management, predictive maintenance, pipeline and facility monitoring, intelligent asset tracking, digital twins and worker safety. The design principle is the same in both: augment operators first and automate only when controls, evidence and operating confidence are mature.
3. Scale with governance built into the architecture. Embed explainability, role-based decision rights, cybersecurity, model monitoring and auditable workflows into the orchestration layer. HCLTech then helps move proven use cases beyond pilots through engineering-led transformation, platform integration and managed operations, without creating a separate governance model for every asset class.

For a utility, that pathway could link asset health models to outage forecasting, crew dispatch and restoration workflows. For an oil and gas operator, it could link sensor and inspection data to integrity risk, maintenance planning and field execution. In both cases, the value comes from closing the loop between insight and operational response while keeping people accountable for high-consequence decisions.

Where to start: A practical sequencing framework

For energy and manufacturing leaders weighing their next move, this four-step sequence reflects what is working across our client programs and the industrial sites covered in this year’s research.

A four-step framework for sequencing Physical AI across manufacturing and energy

1. Map one architecture across the portfolio. Inventory the edge, data, computer vision and orchestration capabilities already funded separately in manufacturing and in energy. Consolidate them into a single reference architecture before either side buys another point solution.
2. Sequence use cases by consequence of error, not visibility. Rank candidates by what happens if the model is wrong, not by which team is loudest or which demo is most compelling. Start where AI can fail safely: monitoring, inspection and advisory use cases on distributed assets.
3. Build the audit trail before you need it. For every autonomous or semi-autonomous recommendation, define upfront who reviews it, how often and what evidence a regulator or safety auditor would expect to see. Retrofitting governance after deployment is much harder.
4. Design for the operator, not just the model. Every Physical AI use case should be able to explain, in the frontline operator’s own words, why its recommendation deserves trust. If it cannot, it is not ready to scale, whatever its accuracy.

Organizations that work through this sequence honestly will move faster over the next eighteen months than those chasing the most sophisticated pilot. The Physical AI playbook transfers across manufacturing and energy. What does not transfer automatically is the discipline to sequence it by risk, govern it once and earn trust one decision at a time.

Let’s continue the conversation

If your organization is deciding how to sequence Physical AI investments across manufacturing, utilities or oil and gas assets, HCLTech can help. We assess portfolio readiness, prioritize use cases by value and consequence of error, define a shared edge, data and AI architecture and put in place the governance needed to scale. The work begins with a focused roadmap that ties asset reliability, field productivity, safety and resilience to measurable operating outcomes.

Sources: World Economic Forum, Global Lighthouse Network: Rewiring Operations for Resilience and Impact at Scale (2026); Gartner, Hype Cycle for Physical AI (2026) and 2026 Magic Quadrant for Global Industrial AIoT Platforms; Deloitte, 2026 Power and Utilities Industry Outlook and From Silos to Synergy: How Utilities Are Integrating AI and Geospatial Intelligence for Resilience (2026); IDC, Energy and Utilities Industry-Specific Tech and Innovation Survey (2026); HCLTech, Energy and Utilities Industry Solutions; HCLTech, Data & AI for Utilities; HCLTech, Oil and Gas Technology Solutions; HCLTech, Engineering Excellence for Oil and Gas Operations.