Europe's power grid is absorbing a double wave: the electrification of demand and the rise of variable renewables. Congestion cost Germany 2.78 billion euros in 2024 alone, and the European Commission puts the grid investment needed by 2030 at 584 billion euros. AI does not replace the cables, but it gets far more out of them: according to the IEA, applications already available could unlock up to 175 GW of transmission capacity without building a single new line and save up to 110 billion dollars a year. From wind forecasting to Fluvius's digital meters, here is what AI changes for the grid, seen from Belgium.
Demand is climbing again everywhere. Data centres consumed about 415 TWh of electricity in 2024; the IEA projects around 945 TWh in 2030, a doubling driven first of all by AI itself. Add to this electric vehicles, heat pumps and the electrification of industry: the European Commission expects EU electricity consumption to rise by about 60% between 2023 and 2030.
Supply, for its part, is becoming variable. Wind and solar produce according to the weather, not according to demand. And the infrastructure is ageing: about 40% of Europe's distribution grids are more than 40 years old. The Commission estimates the grid investment needed by 2030 at 584 billion euros.
The cost of doing nothing already shows in the accounts: in Germany, congestion management (redispatch, reserve plants, curtailment of renewables) cost 2.78 billion euros in 2024, down 17% on 2023 but still structurally high. Building new lines takes ten years or more; making better use of existing ones takes a few months. That is exactly AI's playing field.
Every grid balances in real time. The better you forecast, the fewer costly reserves you mobilise and the less green output you curtail. As early as 2019, Google DeepMind showed that a neural network trained on turbine history and weather forecasts, applied to 700 MW of wind in the United States, could announce output 36 hours ahead and raise the value of wind power by about 20%.
Belgium is not a bystander. Since November 2018 the RMI has run a storm forecasting tool for Elia, the transmission system operator, providing 15-minute wind and output forecasts for every wind farm in the Belgian offshore zone. Work published in 2025 in Advances in Science and Research shows that fine parameterisation of the farms combined with a neural network further improves forecasts in the Belgian North Sea. This ties in with our analysis of AI applied to weather forecasting: AI weather models, tens to thousands of times faster than physical models, make these forecasts accessible at a far lower cost.
On the demand side, the same techniques forecast consumption by substation, by neighbourhood, even by building, integrating weather, calendar and behaviour. This is the basis for everything else: without good forecasting, neither flexibility nor congestion management works.
The key figure comes from the IEA's Energy and AI report: broad adoption of existing AI applications could save the power sector up to 110 billion dollars a year by 2035 and unlock up to 175 GW of transmission capacity without pulling new lines. Enough to connect a good share of the projects waiting in the connection queues.
The second reservoir sits with consumers. The entry condition is metering: by the end of 2025, about 80% of Flemish customers, or 2.95 million electricity meters, had switched to the Fluvius digital meter; the final phase of the rollout, launched in early 2026, runs until mid-2029. With dynamic tariffs and the capacity tariff, every shifted kilowatt-hour now has a visible price.
AI turns this signal into action: charging electric vehicles when electricity is abundant, modulating heat pumps, steering home and industrial batteries. Aggregated by the thousand, these small assets form virtual power plants able to offer reserve to the grid operator. For a company, flexibility becomes a revenue line: shifting a process by a few hours, valuing a chiller or a battery on the balancing markets, shaving the peak to cut network costs.
Three projects set the tone. First, Fluvius's digital metering, the foundation of residential and SME flexibility. Then Elia and the RMI's offshore forecasting, refined by neural networks, essential for integrating the Belgian North Sea wind zone. Finally, the Princess Elisabeth Island, built by Elia 45 km off the coast: the world's first artificial energy island, designed to connect the new offshore farms and the interconnections towards the United Kingdom and Denmark, with heavily instrumented operation.
For Belgian companies, the message is concrete: quarter-hourly metering data exist, the flexibility markets are open to mid-sized players too through aggregators, and the grid operators publish their capacity maps. Those who structure their energy data now will be the first to monetise their flexibility.
AI applied to the grid is a critical system. The European AI Act classes AI that is a safety component in the management and operation of critical infrastructure, including electricity, as high risk: risk management, data quality, human oversight and documentation become mandatory from 2 December 2027 (the Annex III high-risk deadline deferred by the Digital Omnibus, final since 29 June 2026), as detailed in our AI Act guide. The NIS2 directive imposes, in parallel, a cybersecurity baseline on energy operators. And there is a paradox to keep in mind: AI helps the grid, but its data centres also load it, a balance we quantify in the energy impact of AI. In the control room, AI proposes, the operator decides.
Quarter-hourly meters, sub-metering, BMS, own production: inventory what is measured, at what granularity, and who has access.
A per-site forecasting model is the first useful deliverable: it feeds energy purchasing, peak shaving and battery sizing.
Cooling, compressed air, fleet charging, batteries: quantify what can be shifted, over what duration, at what comfort cost.
Value a few hundred kW on the balancing markets or through the capacity tariff, and measure the real revenue over 6 months.
Automate the steering, contract it, document AI Act and NIS2 compliance, and track the indicators over time.
A grid fitted with sensors, communicating meters and software that measure and steer the flows in near real time. AI forecasts output and demand, optimises the operation of existing assets and aggregates the flexibility of thousands of small devices.
Up to 110 billion dollars a year for the power sector and up to 175 GW of transmission capacity unlocked, according to the IEA's Energy and AI report. AI fault detection also cuts outage time by 30 to 50%.
By the end of 2025, about 80% of Flemish customers (2.95 million electricity meters) had a Fluvius digital meter. Elia has relied since 2018 on an RMI tool for 15-minute forecasts for every offshore farm, improved with neural networks, and is building the Princess Elisabeth energy island 45 km off the coast.
Yes: as a safety component of a critical infrastructure, it falls under the high-risk systems of Annex III of the AI Act, with obligations applicable from 2 December 2027 (deadline deferred by the Digital Omnibus), in addition to the NIS2 directive for cybersecurity.
Molderez Consult SRL helps Belgian companies structure their energy data, build reliable forecasts, quantify the flexibility they can mobilise and frame the AI Act and NIS2 compliance of their systems.
Discuss my projectTransparence : cet article a été rédigé avec l'aide de l'intelligence artificielle, puis relu par Molderez Consult SRL.