The energy behind AI and the AI behind energy

AI is creating a new source of electricity demand through data centres and high-performance computing. At the same time, it can help energy companies improve forecasting, maintenance, system visibility, and resilience. The opportunity for utilities, large energy users, and investors lies in understanding both sides of the issue, rather than treating AI as either a simple demand shock or a simple efficiency tool.

A new pressure on the energy system

Over recent years, energy markets have already had to absorb a series of structural changes. Electrification has gathered pace, renewable generation has expanded, infrastructure has aged, and demand patterns have become less predictable. AI now adds another layer to that challenge.

This is not because AI is the largest source of global electricity demand. It isn’t. The issue is that AI-driven demand is growing quickly, appearing in concentrated locations, and often moving faster than the energy infrastructure that’s required to support it.

The International Energy Agency (IEA) expects global electricity demand to grow by an average of 3.6% a year between 2026 and 2030, adding roughly 1,100 terawatt hours (TWh) of demand each year. To put that in perspective, that is enough electricity to power some small-to-mid-sized countries for hundreds of years. This growth is being driven by several factors, including industry, electric vehicles, air conditioning, heat pumps, and data centres. Data centres, as one of the newest yet fastest growing sources of electricity demand, are therefore part of a much wider story: the world is becoming more electricity intensive. 

Energy for AI

The power requirements of AI are most visible in data centres. These facilities house the servers, storage, networking equipment, and cooling systems that underpin modern digital services. As AI workloads grow, more high-performance computing capacity is required, with greater density and more demanding power profiles.

The sheer scale of AI’s seemingly ever-growing power surge cannot be ignored. Gartner estimated that data centres consumed 448 TWh of electricity in 2025 and the IEA expects this to rise to 945 TWh by 2030 (~111% increase), which would be around 3% of the total predicted electricity consumption in 2030. That is still a limited global share, but the rate of growth is notable: data-centre electricity consumption is expected to grow much faster than electricity demand from most other sectors.

The pressure is particularly visible in some markets. In the United States, the Department of Energy reported that data centres consumed about 4.4% of total US electricity in 2023, with that share projected to reach approximately 6.7% to 12% by 2028. The exact path remains uncertain, but the direction of travel is clear enough for utilities, regulators, and large users to plan against it. 

Why local impact matters more than global share

The global numbers can understate the practical challenge. Data centres are not spread evenly across energy systems. They tend to cluster where there is fibre connectivity, available land, favourable planning conditions, customers, tax incentives, and access to power. That concentration can create local constraints even when the global share of electricity demand appears modest.

This is where the issue becomes more operational than theoretical. A data-centre project can be developed in a few years. A new transmission line in an advanced economy can take four to eight years to build, while wait times for essential components to grid infrastructure such as transformers and cables have increased drastically in the last few years. The mismatch between digital infrastructure timelines and energy infrastructure timelines is one of the central tensions created by AI growth.

For businesses, this changes the nature of site selection. Power availability cannot be treated as a background assumption. A site with strong commercial logic may be difficult to develop if grid capacity, connection timing, water availability, or local community acceptance is uncertain.

What is happening now?

The conversation is already moving beyond whether there is enough electricity in aggregate. The more difficult questions are where the power will come from, who pays for the necessary infrastructure, and how costs are allocated fairly.

In some systems, large new loads may improve utilisation of existing assets. In others, they may require significant investment in generation, substations, transmission, storage, or distribution reinforcement. The impact on prices will depend on local conditions, tariff design, and whether large users can operate more flexibly.

Transparency is therefore becoming more important. In the European Union (EU), as part of the revised Energy Efficiency Directive (EED), significant data centres are now subject to monitoring and reporting obligations covering energy performance, with a European database collecting and publishing information relevant to energy performance and water footprint. This reflects a broader direction of travel: as data centres become more important to local power systems, stakeholders will expect clearer evidence of energy use, efficiency, and environmental impact.

However, the data is still imperfect. In an event held around the French AI Action Summit, the OECD (Organisation for Economic Co-operation and Development) and the IEED (Institute of Electrical and Electronics Engineers) stressed the difficulty of getting reliable estimates of data-centre energy use and of identifying how much should be attributed specifically to AI. They also noted that inference (the use of trained AI models) may become a larger environmental and energy issue as AI applications scale. For corporate users, this means claims about AI-related energy use should be handled with care and supported by transparent assumptions.

AI for energy

There is, however, another side to the story. AI is not only a source of additional demand. It is a tool that can help energy companies manage the more complex systems required to power it.

Electricity networks are becoming more decentralised and data-rich. Renewable generation is more weather-dependent. Demand is becoming more variable. Assets are ageing in many markets, while customer expectations around reliability remain high. In that environment, better use of data can make a practical, and perhaps essential, difference.

AI can improve forecasting, helping utilities anticipate demand, renewable output and weather-related disruption. It can support predictive maintenance by identifying early signs of abnormal asset performance. It can help control rooms assess scenarios more quickly and improve visibility across networks, generation assets, and grid-edge devices (equipment that sits between the power grid and end users, helping to manage electricity flow intelligently, i.e. smart metres).

Theoretically, the value gained from using AI in this way is far from insignificant. The IEA estimates that AI-based fault detection can reduce outage durations by 30% to 50%. It also estimates that remote sensors and AI-based management could unlock up to 175 GW of transmission capacity without building new lines, if applied effectively. These are not automatic outcomes, but they show why AI should be viewed as part of the operating toolkit for the sector.

The balance utilities need to strike

For utilities, the AI challenge is therefore two-sided. On one hand, they face new connection requests, more concentrated loads, and pressure to invest at pace. On the other, they have an opportunity to use AI to improve their own planning, maintenance, and system management.

This balance will be difficult to achieve if AI is treated purely as a technology project. The stronger approach is to view it as an operating model issue. Utilities will need better data from large customers, clearer processes for managing connection queues, and more disciplined capital prioritisation. They will also need governance that ensures AI-supported decisions are reliable, explainable, and subject to human accountability.

The most effective applications are likely to be practical rather than dramatic. Better outage prediction. More targeted maintenance. Clearer network visibility. Stronger demand forecasting. Faster scenario analysis. These improvements may not attract the same attention as new generation projects, but they can materially improve resilience and capital efficiency.

What large energy users need to consider

For large energy users, particularly data-centre operators and power-intensive businesses, energy strategy is becoming part of growth strategy. It is not enough to secure a site and assume the power will follow.

Procurement is also changing. Annual renewable electricity matching remains common, but some large users are looking at more granular approaches that match electricity demand with carbon-free supply on an hourly basis. Sustainable Energy for All describes 24/7 carbon-free energy as an approach in which every kilowatt-hour of consumption is matched with carbon-free electricity every hour of every day. This stands in contrast to the traditional route of “annual renewable matching”, which is to say buying enough renewable energy credits over a year, and instead pushes for matching energy demand with clean energy in real-time. This will not be practical or necessary for every organisation, but it illustrates how expectations around credible procurement have become more sophisticated.

The reputational dimension also matters. Where power systems are constrained, large users may face scrutiny over whether their projects support or strain local energy systems. Collaboration with utilities, system operators, policymakers, and local communities will only become increasingly important.

What investors should watch

In the investor space, AI creates opportunity across the energy value chain, but selectivity is essential. Growth in electricity demand may support investment in grids, flexible generation, storage, power equipment, energy management software, and efficiency services. Yet not every AI-related opportunity is equal and they will not always create durable value.

The fundamentals still matter. Location, connection rights, permitting risk, customer credit quality, technology maturity, and local market design will determine whether a project can move from theme to return. In many cases, the more attractive opportunities may be those that solve bottlenecks rather than those that simply sit near a fashionable trend.

The next chapter

AI will not remove the need for physical energy infrastructure. Nor is data-centre growth alone likely to define the future of electricity demand. The next phase will be shaped by how well utilities, users, investors, and regulators manage the interaction between demand growth, grid capacity, affordability, and operational intelligence.

Where there is balance, however, there is a risk of falling, and if demand runs ahead of infrastructure, the result may be delays, higher costs, and lower public confidence. If AI is deployed in energy operations without the right data, skills, and governance, it may create new risks rather than reduce existing ones.

However, there is also a genuine opportunity. Used well, AI can help the energy sector become more responsive, efficient, and resilient. Managed poorly, it may add pressure to systems that are already constrained.

The organisations that will be best placed are those that recognise both realities. They will plan for AI-related demand with greater discipline, use AI in operations with appropriate oversight, and make investment decisions based on practical constraints rather than broad assumptions. In an increasingly electrified economy, that balance may become one of the defining tests of energy leadership.

Contact us

Complete this form and an RSM representative will be in touch.