Modernizing the Power Grid for AI and Rising Demand With SAP

Can electricity grids built for an earlier era support AI data centers, expanding manufacturing, electric vehicles, severe weather, and rising customer expectations at the same time?

In this episode of Tech Talks Daily, I speak with Mark Hollis, utility executive advisor at SAP, about the pressures reshaping the utility industry and the practical choices available to leaders today. Mark spent over 15 years at Duke Energy before moving to SAP, where his work gives him visibility into utility organizations across North America.

Mark describes a combination of load growth, disruption, long construction timelines, and regulation. Data centers are receiving much of the attention because of the electricity required by AI, but he argues that they are only part of the story. Manufacturing is returning to parts of North America, transport and heating are becoming increasingly electrified, and utilities must prepare for wildfires, hurricanes, winter storms, and other events that affect generation and delivery.

The obvious response is to produce additional electricity, but every option comes with physical and commercial limits. Wind and solar contribute to the generation mix, although output depends on conditions. Small modular nuclear reactors could support future demand, but commercial deployment takes time. Batteries can store electricity and release it later, but they do not generate the power they hold. Customer programs can also reduce pressure at busy periods, including arrangements that allow a utility to adjust connected thermostats by a few degrees.

This makes modernization a portfolio of decisions rather than a single bet. Utilities must decide how to divide capital among generation, transmission, resilience, customer systems, and new technology. The people who understand existing processes are often the same employees needed to design and implement replacements. At the same time, information technology and operational technology are becoming increasingly connected, forcing companies to reconsider how business functions share information and how technology decisions support outcomes across the enterprise.

AI creates another tension because it contributes to electricity demand while also offering utilities new ways to work. Mark says most utilities he meets are cautiously optimistic. Their questions include where to begin, whether the value is proven, how long adoption will take, and whether poor data must be fixed before useful work can start. He warns against choosing the hardest problem first or judging a business process by the forgiving standards people accept from consumer AI tools.

His advice begins with the business problem. Automating an inefficient process can make it more expensive and harder to correct. Utilities should define what they need to improve and why, establish connected data with the right business context, set clear quality requirements, and retain human review where errors could affect customers, safety, finance, or regulatory obligations.

Mark brings this to life with several utility AI use cases. AI could summarize customer interactions across field service and contact center systems, allowing the next employee to understand what happened previously. It could review billing exceptions during unusually hot or cold periods and support earlier customer communications when consumption is likely to produce a much higher bill.

He also discusses using AI to summarize lengthy rate-case rulings before approved changes enter billing systems. During outages, an AI system could help dispatch crews by considering skills, equipment, parts, certifications, location, safety, and customers with medical needs. A human dispatcher could then review and approve the recommendation rather than building the complete schedule manually.

The opportunity is real, but so are the limits. Utilities operate regulated infrastructure where reliability, safety, auditability, and public trust cannot be treated as optional features. Where should the industry begin, and which use case offers the right combination of low effort, meaningful impact, and manageable risk? Listen to the conversation and share your thoughts with me.