AI Data-Centre Boom Is Turning Grid Capacity Into Scarce Infrastructure

Electricity use by data centres is expected to almost double by 2030, intensifying pressure on power systems that often take far longer to expand than the facilities they must serve. UNECE is warning that the widening gap between data-centre construction and grid development is creating new risks around reliability, regulation, infrastructure costs and the environmental footprint of digital growth.

Artificial intelligence is scaling faster than the electricity systems built to support it. Global data-centre power consumption is expected to rise from 485 terawatt-hours in 2025 to 950 TWh by 2030, according to the International Energy Agency, pushing the sector's share to around 3% of global electricity demand.

The primary concern is not simply how much electricity data centres will need, but how quickly that demand is arriving. UNECE warns that major computing facilities can be built within two to five years, while the transmission lines and grid infrastructure required to support them can take more than a decade to plan, approve and construct.

AI Growth Is Colliding With the Slow Pace of Grid Expansion

The mismatch exposes a structural weakness in the current AI investment cycle. Computing capacity can be commissioned relatively quickly as companies expand cloud services and AI workloads, but electricity networks operate on far longer infrastructure timelines.

Global data-centre investment is forecast to rise from around US$800 billion a year in 2026 to US$1.8 trillion annually by 2050. Electricity demand, however, is accelerating much sooner, creating pressure for grid investment to anticipate digital growth years before facilities are fully operational.

Location makes the problem more acute. Data centres tend to cluster where connectivity, land and regulatory conditions are favourable, concentrating very large loads within specific parts of electricity networks rather than spreading demand evenly across national systems.

A country can have sufficient generating capacity overall while still facing severe local constraints. Congestion, connection delays and weak transmission infrastructure can emerge in precisely the regions attracting the most digital investment, turning geography into a critical part of the AI infrastructure equation.

Grid Stability Is Becoming an AI Policy Issue

UNECE's warning goes beyond concerns about electricity consumption. Large AI data centres can create volatile demand that changes rapidly, making it harder for grid operators to balance supply and demand in real time.

Overloaded electricity systems can experience voltage oscillations, unintended disconnections and, in extreme circumstances, cascading failures. UNECE says such stresses are already appearing in some systems that rely heavily on renewable generation, where rapid changes in data-centre demand can be difficult to match immediately with available supply.

The issue is not renewable energy itself, but flexibility. Electricity systems with variable generation need sufficient storage, network capacity, balancing resources and operational visibility to respond when large consumers suddenly increase or reduce their demand.

Several governments are already responding to local pressure. Ireland has introduced restrictions on new data-centre connections in Dublin, while the Netherlands has imposed constraints on where facilities can be built, reflecting growing concern about concentrating large electricity users in grid-constrained areas.

These measures indicate how the AI infrastructure debate is widening. Data centres are no longer being treated solely as technology investments or real-estate projects; they are increasingly becoming decisions about electricity-system planning, network resilience and access to scarce grid capacity.

The Most Difficult Question May Be Who Pays

Expanding the grid is technically possible, but financing it creates a more complicated policy problem. Connecting a major data centre can require substantial investment in transmission lines, substations and other infrastructure, yet there is no consistent framework for determining who should bear those costs.

Data-centre developers, electricity companies and consumers all have different interests. Developers want predictable and timely grid connections, utilities must recover the cost of infrastructure, while regulators face the question of whether households and businesses should subsidise upgrades driven by unusually large new sources of demand.

Uncertainty over cost-sharing can delay investment at precisely the moment when grid expansion needs to accelerate. UNECE has warned that unclear rules can slow infrastructure decisions, leaving power systems exposed to rising demand without a defined mechanism for financing the required upgrades.

Regulators also lack enough information about how some large facilities use electricity in real time. Limited visibility makes it harder to forecast demand spikes, assess local network stress and decide where new infrastructure should be built first.

Clearer connection rules, stronger reporting requirements and transparent cost allocation could therefore become as important to the next phase of data-centre expansion as access to land, chips or capital.

Regulation Is Expanding Beyond the Power Bill

The regulatory challenge is widening further as governments examine the broader environmental footprint of data centres. Electricity consumption remains the most immediate concern, but water use, emissions and pressure on local resources are also moving higher on the policy agenda.

UNECE warns that environmental rules remain fragmented. Different jurisdictions can impose very different standards, potentially encouraging facilities to cluster in locations where regulation is less restrictive even when local infrastructure is already under strain.

Governments face a difficult balance. Data centres bring investment and support the digital economy, but large concentrations of energy-intensive facilities can place additional pressure on electricity systems and surrounding resources if network expansion and environmental safeguards fail to keep pace.

The next phase of AI expansion may thus depend as much on energy policy as on computing technology. Grid-connection rules, infrastructure financing, siting restrictions and real-time reporting requirements could increasingly influence where data centres are built and how quickly they can operate.

Key questions now extend beyond whether enough electricity can be generated. Policymakers must decide where high-demand facilities should be located, how much flexibility grids need, who should pay for new infrastructure and how much information operators must provide about their electricity use.