The Data Centre Boom Has a Power Problem

The global race to build artificial intelligence is producing an unexpected contest for something far older than computing: electricity. Microsoft, Google, Amazon and Meta are investing in vast data centre infrastructure, but their ability to expand increasingly depends on transmission lines, substations, transformers and the willingness of electricity networks to accommodate another enormous customer. A company can finance a building, order servers and secure land long before the local power system is ready to serve it. That gap is becoming one of the defining business constraints of the AI economy.

A data centre is often presented as a collection of servers processing information. Its physical demands are much broader. Processors require uninterrupted electricity, cooling equipment removes heat, backup systems protect against interruptions, and network connections link the facility to customers around the world. The more powerful the computing equipment becomes, the more demanding the supporting infrastructure can be. The commercial value of a data centre therefore depends not simply on the machines inside it, but on whether an entire network of utilities and suppliers can support those machines reliably.

In Northern Virginia, one of the world's largest concentrations of data centres, the industry has become a major factor in electricity planning. The region's appeal originally rested on fibre connectivity, proximity to major internet infrastructure, skilled workers and an established ecosystem of operators. Those advantages attracted further investment, creating a cluster whose scale now places pressure on the infrastructure that helped make it successful. For electricity utilities, a new data centre is not merely another commercial connection. It can represent a substantial and persistent addition to demand, requiring investments whose consequences extend well beyond the facility itself.

The International Energy Agency estimated that data centres consumed approximately 415 terawatt-hours of electricity globally in 2024, roughly 1.5% of worldwide electricity consumption. Its 2025 analysis projected that consumption could more than double to approximately 945 terawatt-hours by 2030. These figures do not mean that every planned data centre will be built, or that electricity demand will follow a perfectly predictable path. They do establish the scale of the challenge: computing infrastructure is becoming a significant consideration in decisions about electricity generation, networks and industrial investment.

The economics become especially complicated because electricity infrastructure and computing infrastructure operate on different timescales. Technology companies can deploy new generations of processors relatively quickly. Building major transmission connections, expanding substations and obtaining approvals for electricity infrastructure can take considerably longer. A delayed connection can leave an otherwise completed development unable to generate its expected revenue. Our earlier examination of grid connections shows how these delays affect projects throughout the energy economy. The resulting costs may fall on the developer, the utility, other electricity customers or some combination of them, depending on the regulatory arrangements and commercial contracts involved.

Ireland offers a particularly revealing example. Dublin's established technology ecosystem helped attract major data centre investment, but the concentration of facilities also raised difficult questions about electricity demand and grid capacity. Irish authorities have had to balance the economic advantages of digital infrastructure against the practical limits of the power system. The issue is not whether data centres contribute to the economy. It is whether the wider infrastructure can absorb their growth without creating unacceptable consequences for other users or national energy objectives.

The same tension appears in emerging digital markets, although the starting conditions differ. In Kenya, South Africa and Nigeria, expanding demand for cloud services, financial technology and local data processing creates opportunities for data centre operators. Yet the commercial proposition must account for electricity reliability, backup generation, energy costs and network connectivity. Our reporting on African electricity networks illustrates why access and reliability cannot be assumed. A facility in Nairobi or Lagos competes not only on its computing specifications but on the reliability and total cost of delivering a continuous service.

Cooling adds another layer of complexity. Traditional air-based cooling systems face growing challenges as computing equipment becomes denser and more power-intensive. Liquid cooling can remove heat more effectively from high-performance equipment, but it introduces additional engineering, maintenance and investment requirements. Companies supplying cooling systems, power distribution equipment, backup batteries and electrical controls consequently occupy increasingly important positions in the data centre value chain. Maintaining data centre equipment also depends on the specialist technicians whose shortages can stop expensive machinery elsewhere in the economy. The visible commercial race may involve AI models and semiconductor chips, but substantial opportunities also sit with the businesses that make their operation physically possible.

Water introduces a further local consideration. Some cooling technologies use water directly or indirectly, creating questions about availability in places already experiencing pressure on supplies. Water requirements vary considerably according to the facility's design, climate and cooling technology, so data centres cannot all be treated as having the same environmental footprint. Nevertheless, local authorities must consider how industrial demand interacts with residential consumption, agriculture and other economic activities. A project that looks efficient on a global corporate balance sheet can still produce difficult trade-offs within the community hosting it.

Energy sourcing has become part of the competitive response. Technology companies have entered agreements supporting renewable electricity, explored nuclear power arrangements and invested in approaches intended to reduce the emissions associated with computing. These arrangements can help support additional generation or manage long-term exposure to electricity prices. They do not automatically resolve the physical problem of delivering sufficient power to a particular facility at the moment it is needed. The supply chain for electricity infrastructure, including transformers and high-voltage equipment, can be as decisive as the electricity purchase contract itself.

This distinction matters for public policy. Governments want investment, construction activity, technical employment and the economic benefits associated with digital infrastructure. Utilities must maintain reliable electricity supplies. Regulators must decide how infrastructure expansion should be financed. Local communities may welcome investment while questioning its effects on land, water and electricity costs. Each participant sees a different part of the same project, and decisions made in one part can transfer costs or constraints to another.

The relationship between data centres and the electricity grid also changes how locations compete. A site with inexpensive land and favourable tax treatment may prove less attractive than one with reliable access to substantial electrical capacity. Regions that can coordinate power generation, transmission infrastructure, planning and digital investment may gain an advantage over places that pursue data centre developments without addressing the infrastructure they require. The availability of electricity becomes an economic development asset in its own right.

There is a further uncertainty in the business model. Data centre developers and utilities must make long-lived investments based on expectations about future computing demand. Yet artificial intelligence is evolving rapidly. More efficient processors, changing model architectures, different patterns of AI use and shifts between training and inference could alter the amount and location of electricity required. Building too little infrastructure risks restricting growth; building too much risks leaving customers with expensive assets that are not fully used. The challenge is therefore not simply predicting how much AI will expand, but determining who carries the financial risk when forecasts prove wrong.

This is why the data centre boom cannot be understood solely as a technology investment story. It is also a story about industrial geography, electricity regulation, construction capacity, capital allocation and the distribution of infrastructure costs. The companies that control computing capacity may command enormous market attention, but their expansion depends on a network of less celebrated businesses and institutions. Where power interruptions remain common, backup systems become a separate economic burden rather than a simple precaution.

The next phase of the digital economy will be shaped as much by decisions in electricity control rooms, planning offices and engineering workshops as by developments in AI laboratories. Computing power can be purchased, but dependable electrical capacity must be planned, financed, connected and maintained. The data centre boom has made that distinction impossible to ignore.