The digital grid toolkit

Where grids cannot be built fast enough, the answer lies in delivering more energy, securely, with the existing networks. Existing grids typically contain meaningful untapped capacity, spare capacity that varies by location and time. Not all of it is usable, because it must be weighed against power quality, system flexibility, stability and resilience, but it can be unlocked by improving how the system is planned and operated.

Three complementary tools can unlock the untapped grid capacity: digitalisation, grid-enhancing technologies (GETs) and artificial intelligence (AI), collectively referred to here as the “digital grid toolkit”. This section introduces what each is and how they relate; Chapter 2 gives a fuller technical treatment of each.

Digitalisation is the foundation. Better sensing, data and connectivity, for example through supervisory control and data acquisition (SCADA) and advanced metering, allow operators to see their networks far more fully and control assets remotely, enabling them to route flows, manage congestion, and support voltage and stability actively.

GETs build directly on that foundation. GETs are hardware, software or hybrid solutions that cost-effectively raise the capacity, flexibility, stability or resilience of electricity grids by revealing latent headroom and system limits, redirecting flows or making existing assets more controllable. This report focuses on GETs that offer the greatest potential for digitalisation and AI and that are deployed and controlled directly by grid operators, such as dynamic ratings, advanced power-flow control, topology optimisation and fast corrective controls, rather than the wider set of technologies (including advanced conductors, voltage uprating1 and storage) sometimes included under the GET label in the literature.

AI adds a further, cross-cutting layer. It spans a spectrum from rules-based systems through machine learning to generative and, increasingly, agentic AI, capable of planning and acting across several steps with limited prompting. AI has been used by network operators for two decades, but what is different about the current wave of AI solutions is that new AI architectures can process large, diverse data streams, graphics processing units (GPUs) and cloud computing have sharply improved the speed and cost of model training and inference, and utilities now hold more operational, asset, weather, imagery and customer data than in previous technology cycles. AI can therefore support operators managing the growing grid complexity in unprecedented ways.

None of these substitutes for continued grid expansion. Digital, GET and AI-enabled solutions unlock capacity more quickly and, in many cases, at lower cost, but traditional buildout remains indispensable; the two should be used together to prioritise, defer or target new infrastructure where appropriate.

Better use of the existing grid, at both transmission and distribution level, can be understood as three layers, each resting on the layer below, each with a different balance between human judgement and AI, and sharing one goal: enabling higher system utilisation, in a secure way.

At the base is the operational security criteria layer, which determines how risk is assessed and operating margins are set. Traditionally, grids have been planned and operated with deterministic security margins: networks were sized to cope with the expected highest demand hour of the year, while also accounting for a defined list of likely outages that could reduce network capacity. A risk-based approach instead asks how often, and by how much, the grid could be pushed further without increasing the chance of a fault, thereby replacing implicit margins with explicit, quantified decision criteria. This is a fundamental change, requiring new planning and operational methods, but it is where the highest long-term returns for utilisation, reliability and resilience are likely to sit. The operational security layer sets a ceiling on what the two layers above it can safely achieve.

The IEA’s Electricity 2026 report found that the largest gains from GETs arise precisely when they are combined with a risk-based approach to grid operation. AI and digital tools can monitor, forecast and advise on limits, but secure operation still rests on validated models, clear rules, operator authority and the ability to act under contingencies.

The network asset layer above draws more from the assets in service, through GETs, including automation and decision-support tools, and keeps assets available through predictive maintenance, early warning of developing faults and faster restoration; here AI helps optimising and predicts, while the operator validates and supervises.

The flexibility layer at the top matches demand-side resources, batteries and EVs, whether behind or in front of the meter, to spare capacity in time and place; co‑ordination can rely on market or price signals, but flexibility can also be delivered without an explicit signal, for example through pre-agreed operating protocols. Here AI forecasts and co‑ordinates, and can act within sub-seconds, while the operator sets rules, products and market design.

Across all three layers, the respective roles of human and AI are expected to evolve, but human oversight remains structural: operators stay accountable for system security even as AI takes on a larger operational role, and sustaining that accountability requires operators to retain sufficient understanding, tools and authority to intervene at any point, a theme this report returns to when discussing trust in AI-based decision-support tools.

Because the layers draw on the same physical headroom, gains overlap rather than add up; as the network asset layer pushes assets harder, the binding constraint can move from thermal capacity to stability, so operational security margins must keep pace. This is why capacity-unlocking measures need to be assessed against the operational security framework that makes them usable, and why regulators and operators need transparent methods for quantifying uncertainty, approving risk-based limits and validating remedial actions before new headroom is reflected in planning, markets or real-time operation.

This report concentrates on the two lower layers: the network itself and how it is operated, since the physics and operating practices involved are broadly common across countries and improvements transfer relatively easily. This is where the IEA's work with energy regulators can most directly help operators use existing capacity better.

While the flexibility layer is an equally important route to higher utilisation, it is more specific to each market's design and involves many actors beyond the network operator; it is therefore the focus of complementary IEA work and is treated here as context.

The opportunity to make better use of grids arises across all timeframes: long-term planning, operational planning over horizons ranging from several weeks ahead to day-ahead, and real-time operation. Each rests on different assumptions as uncertainties narrow as the system approaches real time, but the same underlying transition applies throughout: from static, worst-case assumptions towards methods that represent a wider set of conditions, including evolving climate conditions, more realistically. The digital grid toolkit can add value at every timeframe, but only delivers that value where it is actually used. A tool proven and used in real-time operations, for instance, but absent from the planning studies conducted months or years ahead, never influences the decisions made on the basis of those studies.

Real-world deployments show measurable benefits across the digital grid toolkit, including higher transfer capacity, lower congestion costs, faster connection studies, improved forecasting and greater reliability.

These examples show that the digital grid toolkit can deliver measurable gains when applied to a clearly identified system need and integrated into routine network processes. The size of the benefit, however, depends on the local constraint being addressed and on whether the technology is treated as part of planning and operations. This is why deployment should start from a system-needs assessment: identifying the system constraints, and then selecting the appropriate combination of digital, operational and physical measures.