Artificial intelligence is often discussed as a contest over computing power, investment and innovation. Increasingly, however, its most immediate constraint may be far more physical: whether electricity networks can connect and reliably serve the data centres required to train and operate advanced systems. The rapid expansion of AI infrastructure is creating a growing mismatch between the pace of digital investment and the slower development of generation, transmission and grid capacity.
The United Nations Economic Commission for Europe has warned that data intensive facilities are being constructed faster than the electricity systems needed to support them. This does not mean that AI data centres will inevitably destabilise power networks. It does mean that governments, utilities and technology companies face a planning challenge in which poorly coordinated growth could create local reliability, affordability and infrastructure risks.
Demand is Rising Faster Than Networks
The scale of projected demand illustrates why this issue has moved beyond a narrow technology concern. The International Energy Agency expects global data centre electricity consumption to increase from about 485 terawatt hours in 2025 to roughly 950 terawatt hours by 2030, approaching 3 percent of global electricity demand. Electricity use by AI focused data centres is expected to grow even faster, tripling over the same period.
At the global level, this share may appear manageable. Yet national and regional systems experience demand locally, not globally. A small number of high capacity facilities can be concentrated in specific areas, drawing power comparable to major industrial sites. The IEA notes that AI focused data centres can consume as much electricity as 100,000 households, while the largest facilities under construction may require twenty times that amount.
This concentration creates a central policy dilemma. Data centres can often be developed and connected within two to five years, whereas new transmission infrastructure may take more than a decade because of planning, approvals and construction. The result is a risk that digital capacity becomes available before the surrounding grid can accommodate it.
Reliability Depends on Coordination
The principal concern is not simply higher electricity consumption, but the timing and variability of demand. AI computing loads can change quickly, particularly where training, inference and other intensive tasks coincide. UNECE has cautioned that stressed systems could face voltage fluctuations, unplanned disconnections and, in severe circumstances, wider cascading failures.
Grid constraints may also delay the technology investments themselves. The IEA estimates that, unless current risks are addressed, around one fifth of planned data centre projects could face delays due to grid pressures. It identifies lengthy connection queues, shortages of transformers and cables, and the extended timeline for new transmission as major obstacles.
These pressures are likely to intensify debates about location. Ireland has imposed restrictions on some new connections in Dublin, while the Netherlands has applied constraints on where data centres can be built. Such measures illustrate an emerging shift from treating data centre siting primarily as a commercial decision toward viewing it as an electricity system planning question.
More Read
Who Pays for Expansion?
A further issue concerns the distribution of costs. Large new facilities may require upgrades to substations, transmission lines, generation capacity and storage. Without clear rules, those expenses could be borne by developers, electricity providers, public authorities or, indirectly, other consumers through tariffs.
The IEA stresses that data centres do not necessarily increase electricity prices in every setting. In systems with spare capacity, predictable new demand can improve the use of existing infrastructure. However, in constrained systems, rapid and uncertain load growth can require expensive investment and may place upward pressure on prices if costs and risks are not managed carefully.
Transparent connection rules and cost allocation frameworks therefore matter as much as engineering solutions. They can help determine whether infrastructure investment is timely, whether consumers are protected from avoidable costs, and whether developers receive credible signals about where new projects can be accommodated.
Flexibility Can Change the Equation
The issue is not solely one of building more infrastructure. Data centres could also become more flexible electricity users. The IEA suggests that operators could shift certain workloads, use spare server capacity differently, or deploy onsite batteries and backup generation in ways that support grid operations during periods of stress.
This approach has limits, since interruptions to AI services can be commercially costly. Nevertheless, greater flexibility could reduce pressure at peak times and allow more efficient use of existing networks. It also points to a broader principle: AI infrastructure should be planned as part of the power system, rather than added after electricity decisions have already been made.
A Final Note
The expansion of AI data centres presents both an infrastructure challenge and an opportunity for better energy planning. Its consequences will depend less on whether AI demand grows, which appears increasingly likely, than on how quickly grid investment, regulation, transparency and operational flexibility adapt to that growth. A coordinated approach can reduce reliability risks while preserving the economic and technological benefits that AI infrastructure may offer.
