AI studies 2026: what utilities and municipal utilities report on AI

What AI in the energy industry means

The studies count differently. PwC asks whether a company has introduced AI solutions, and a pilot is enough. Digital@EVU asks about strategy, skills and which of 30 listed applications are in use. Keep that in mind when you put the figures side by side.

We found no study that puts a euro figure on investments or savings for the industry as a whole.

The studies at a glance

Six pieces of work carry the picture. Two survey utilities, one models processes, one asks customers, and two look at the grid.

PwC: four in five have started

PwC surveyed decision-makers at German utilities of every size, from municipal utilities (Stadtwerke) with fewer than 100 employees to integrated groups with more than 2,000. The study covers 2025; the trade paper ZfK reported the results on 13 April 2026.

79 percent have introduced first AI solutions. 61 percent of those are experimenting, regardless of company size. PwC calls it the PoC dilemma: solutions get tested and never reach regular operation. Not a single company surveyed rates itself as advanced or AI-driven.

That's the real news.

AI goes where it's easy: 29 percent falls to customer service, 21 to support functions, 12 to sales. Generation, grid operation and trading lag well behind, according to PwC. As barriers, 30 percent name technology and data, meaning interfaces and patchy records, 28 a lack of skills, 18 legal and ethical concerns and 16 cultural and organisational resistance.

Two details are easy to miss. IT heads rate their company's AI maturity more cautiously than management does. And around three quarters of the more than 70 vendor solutions mapped run on a hyperscaler cloud. For a utility that falls under NIS2 or the German KRITIS umbrella act, that's no side issue.

The practice examples ZfK quotes from the study come almost entirely from grid and metering: a chatbot that answers 95 percent of customer enquiries on its own, image recognition for around 50,000 meter readings a month at 99.5 percent accuracy, a plausibility check that resolves 90 percent of meter-reading exceptions without manual work. How such cases get pre-sorted is covered in our article on AI for exception cases in market communication and metering.

Digital@EVU 2026: strategy planned, people missing

Digital@EVU has measured the industry's digital maturity since 2016. BDEW, the Swiss association VSE, Kearney and IMP³ROVE presented the 2026 edition on 14 April 2026. More than 110 utilities from seven countries took part, 96 percent of them from Germany, Austria and Switzerland.

BDEW's headline said more than 90 percent work with an AI strategy or plan one. True. The report also shows the split, though: 29 percent have implemented one, 58 are planning one. Of those with a strategy, 24 percent have clear KPIs to measure its effect.

Skills look similar. 56 percent know which digital capabilities they'll need. 29 percent have an HR strategy to build them.

The barriers depend on size. Across all participants, 76 percent name missing data infrastructure, 68 a lack of staff skills and 59 security concerns, in data protection and cybersecurity alike; that's 24 points more than in the 2023 edition. Among small utilities with revenue below 100 million euros, 71 percent put a lack of internal capacity first. At the other end, among the giants with more than five billion euros in revenue, 73 percent have implemented an AI strategy and 91 percent run their own competence centre for AI and data analytics.

And regulation? 56 percent expect the AI Act to have a neutral effect. For utilities with a chatbot facing customers, we think that's too relaxed, because the Article 50 transparency duties have applied since 2 August 2026.

m3 and Fraunhofer IPK: where the potential sits in a municipal utility

Instead of surveying companies, m3 management consulting and Fraunhofer IPK built a model municipal utility and assessed more than 300 typical processes in five areas: grids, procurement, customer service and sales, commercial processes, generation. The study was presented on 30 October 2025.

The result sounds familiar and is still useful. The greatest potential lies in grid and generation, but projects there are demanding and need an organisation with high AI maturity. The easier entry point is customer service and sales, because it needs neither deep technical knowledge nor major restructuring. That matches what PwC actually finds in use. How far grid operators have got with the harder part is in our article on AI load and generation forecasting in the distribution grid.

One figure circulates in the press. Municipal utilities could save 56 percent with AI by 2035, KOMMUNAL reported on 27 October 2025, citing Handelsblatt. It doesn't appear in the publications of m3 and Fraunhofer IPK, and the baseline is not stated. A municipal utility managing director quoted in the same piece called such figures "false precision".

We'd work with the process list, not the percentage.

Simon-Kucher: what customers think

Simon-Kucher's energy customer study 2026, run with Appinio in November 2025 among 1,000 people in Germany, asks a question the industry studies leave out: what happens if the supplier uses AI to control the electricity supply? 12 percent would cancel their contract.

Twelve percent sounds small. In a market where, according to the same study, one electricity customer in three is open to switching, it isn't.

The study doesn't define what "control" means exactly. For sales it's still a useful distinction: AI in the background, in forecasting or billing, is different from AI that visibly switches things at the customer's home. Both need explaining. For chatbots, Article 50 of the AI Act has required since 2 August 2026 that people can tell they're talking to an AI. How utilities use AI in customer service without losing trust is covered in our article on AI in customer service and churn prevention.

dena and VDE: AI in the grid depends on data

For the grid, two older pieces of work look more closely than any survey. In the Data4Grid project, the German Energy Agency (dena), umlaut and 14 grid operators built three proofs of concept with real grid data: scenarios for e-mobility, an assessment of metering points for more grid transparency, and consumption forecasts from smart meter data. The accompanying Fraunhofer IEE report of January 2023 concludes that AI can make grid operation more efficient and reliable if data management, data security and the legal frame are thought through.

In July 2025 VDE ETG published a background paper on AI in grid control technology. It describes a four-stage introduction model that builds on technical safety management, an ISMS under ISO/IEC 27001 and the V-model of software development. AI adds to the toolbox of the people deciding in the control room. Handing over control is explicitly not the goal.

Both pieces predate the 2026 surveys and still fit their finding. In the grid, the data situation decides, and three in four utilities name it as their top barrier in Digital@EVU.

Municipal utility or large utility: what the figures mean for each

None of the studies reports municipal utilities separately. Digital@EVU does split by revenue, though, and the gaps are wide. We read five differences from the figures; the assessment in the table is ours, the percentages come from the studies.

The pilot trap isn't a municipal utility problem. PwC's 61 percent in trials applies to all sizes. What holds things up differs: in a municipal utility it's people, in a group it's decisions.

Which roles are affected

For a utility, AI isn't a business line of its own. It lands in the market role whose process it changes, and that's where the duties sit too.

Triggers with dates

The studies describe a market taking its time. The legal frame no longer does.

If you're piloting today, set up the documentation so that it still holds in 2027.

What we take from it

On one point the studies agree. The bottleneck is the path into operation.

Four steps for the coming months

1. Inventory before strategy paper

   List which AI systems run in your company, including those built into standard software. Without that list, every classification is a guess.

2. Bring one pilot into operation

   One. Pick the one with the clearest benefit and decide who owns it in operation, how it is monitored and what it may cost. This is exactly the step where, according to PwC, 61 percent get stuck.

3. Data before tools

   For the next two use cases, check whether the data exists, is accessible and may legally be used before you choose a product. Metering data falls under the German Metering Point Operation Act (MsbG), customer data under the GDPR. And if the data sits with a service provider, access belongs in the contract.

4. Treat skills as an HR topic

   Training for AI literacy under Art. 4 covers the duty. Bringing pilots into operation takes defined roles and an HR strategy, which only 29 percent have according to Digital@EVU.

How we support utilities along this path, from prototype to governance, is set out on the page AI for utilities: from pilot to operations.

Frequently asked questions