AI & the future of energy and utilities

AI combines weather models, satellite images of clouds, readings from wind turbines and the past output of solar parks. From these it forecasts how much power will come in per quarter-hour and per area, from an hour to several days ahead. Grid operators see sooner where the grid will fill up, traders base their positions on them and park owners schedule maintenance for a windless day. The latest weather models are themselves built on AI: they compute a global forecast in minutes where conventional models need hours, so a forecast can be refreshed more often each day. The next step is forecasts per street and per minute that also show how certain they are, and that drive switching and trading directly.

Grid operators often apply a fixed load limit to overhead lines, based on cautious assumptions about the weather. In cooler weather or with more wind, the conductors can carry more current. Models use measurements of conductor temperature and sag, plus the hourly weather forecast, to work out how much power can safely flow through a line, and AI makes those forecasts more precise. That frees up room in a grid that is full on paper, while new connections take years to arrive. According to the International Energy Agency, such technology could release up to 175 gigawatts of extra transmission capacity worldwide. The next step moves from one line to the whole grid: AI also manages the load on underground cables, transformers and substations dynamically, and reallocates the headroom every quarter-hour.

Drones fly along high-voltage lines, taking photographs and thermal images. Image recognition picks out a cracked insulator, a joint running too hot or a tree growing too close to the conductor. Sensors in substations measure vibration, temperature and gases in transformer oil, and AI recognises the patterns that come before a failure. If something does go wrong, AI pinpoints the fault faster. According to the International Energy Agency, that can cut the length of power outages by 30 to 50 percent. The next step: drones that fly the grid on their own from a docking station, and robots that inspect substations day and night, so that field engineers head out with a clear target.

An agent is AI that completes a multi-step task by itself. Grid operators receive thousands of questions every day about a connection, a house move or an outage in the street, and an agent can follow such a request from start to finish. In pilots, agents now support customer service; development is moving towards linked data checks, assessment of grid capacity, scheduling of field engineers and updates to the customer. That leaves staff with time for the customer who depends on a medical device at home, or the business owner who has been waiting for months. The next step: the customer's agent and the grid operator's agent agree a connection, a flexible contract or a charging slot between them.

Sensors measure pressure, flow and sound in the pipe network. AI recognises the pattern of a leak, often before any water surfaces, and shows which section of pipe it is in. At the wastewater treatment plant, AI adjusts aeration and chemical dosing to the actual level of pollution, which can lower the plant's energy use. When water is reused for industry and farming, AI combines the sensor readings and flags a change in quality before it reaches the process. The next step is a pipe network that plans its own maintenance, with robots that inspect and repair pipes from the inside.

AI is driving fast growth in the power demand of data centres. According to the International Energy Agency, data centres used about 1.5 percent of the world's electricity in 2024, and that consumption is expected to more than double by 2030. Locally the effect is larger: a single campus can draw as much power as a city. AI can also be part of the answer. Computing that is not urgent, such as training a model, can move to hours with plenty of sun or wind. AI runs the cooling more efficiently, and waste heat can feed a heat network. Tech companies have also signed contracts for power from small modular reactors that have yet to be built. In time, such a plant could stand right next to the data campus.

AI links solar panels, home batteries, charging points, heat pumps and backup generators into a virtual power plant that keeps recalculating how much capacity it has available. Together they can supply power, delay consumption and help keep the grid in balance. AI predicts how much flexibility each installation offers and allocates the tasks within the terms agreed with the owners. Virtual power plants already sell services to the grid commercially in several places, and the number of connected households is growing. Owners stay in control: if you want your car fully charged by seven, you set that. Before long they will be large enough to replace some reserve power stations, and a grid operator will call on them neighbourhood by neighbourhood to relieve a congested grid.

Fed with live measurements, a digital twin shows how new connections, changing demand and lost links affect the load on cables and substations. A grid planner can see what a new factory or a neighbourhood full of heat pumps does to the grid, now and in ten years. AI makes the model faster and fills in missing data, such as consumption in streets without sensors. It can also run thousands of possible futures, with more or less solar power, electric vehicles and industry. That lets the grid operator see years ahead where the grid will be under strain. The next step is a digital twin of the entire energy system, with electricity, heat, gas and water in one model, in which grid operators and governments run scenarios together up to 2050.

Grid operators and energy companies that work AI-first let AI do the repetitive and predictive work that people do not need to do. Forecasting, planning, sorting alerts, checking applications and searching inspection images: AI does all of it quickly and without tiring. A field engineer assessing a situation on site, a colleague deciding during an outage who gets help first, an adviser explaining to a customer why they have to wait: that is where human judgement carries weight. Around 2040, much of planning and inspection could run without human intervention. People then focus on exceptions and on balancing the claims of customers, safety and growth.

When AI works on the grid, the jobs of many people change, and in five to ten years roles that barely exist today could carry real weight. The supervisor of a grid that runs itself sets the goals and limits within which AI switches, and steps in when the system meets something it does not know. A trainer teaches grid AI models to deal with rare situations, from an ice storm to a heatwave. An operator runs a fleet of inspection robots and drones that check substations and lines day and night. And a broker decides what customers' agents and the grid's agents may agree with each other. The work of the field engineer and that of the data analyst increasingly overlap.

During a storm or heatwave, thousands of alerts can arrive in a short time. AI groups alerts with the same cause, points to the likely location of the fault and proposes switching actions to get as many customers as possible back on supply quickly. The operator weighs what the system does not know, such as a care home running on backup power or an engineer still working in a substation, and decides. The next step is a control room in which AI carries out routine switching itself, while the operator keeps control of exceptions, goals and limits. One operator then oversees a much larger part of the grid.

Leadership in energy and water means going first in a change that affects everyone. Leaders take the initiative on AI and new technology and show where the company wants to go, with a picture of the future concrete enough for a grid planner and a field engineer. They bring their people in to help build it: the operator in the control room, the engineer in the substation and the partners in the region. They listen to their staff's ideas and use them to shape a direction people want to follow together. That is how leaders build tomorrow's energy system with their people, even when some things remain uncertain.

In its own future academy, an organisation teaches what it needs now, without waiting for schools and universities to offer that knowledge. Staff who have been around for years learn to work with AI, agents and new technology; new colleagues learn from their first week how work in the organisation is changing. Leaders learn to steer a team of people and AI. The course uses examples from the organisation's own work, and the programme adapts as that work changes.

In an ecosystem, start-ups, scientists, partners and governments work together on problems too big for one organisation. One party's decisions affect another party's planning, and how well they plan together determines how quickly they move forward. In more and more places they meet to coordinate their plans and develop new solutions together. Across borders, too, governments and organisations plan large projects jointly and exchange models and experience. AI can test the joint plans before anyone invests.