AI infrastructure is beginning to change the economics of electricity in a way that extends beyond the question of whether enough generation capacity exists. The more consequential shift may be that data centres are emerging as unusually large, concentrated and persistent buyers of power at a time when almost every major part of the economy wants more electricity. That changes the competitive equation. Factories are electrifying production. Transport is adding electric vehicles. Buildings are increasing their reliance on electric heating and cooling. Households continue to add appliances and digital services.

Semiconductor manufacturing and battery production are also creating new industrial loads. At the same time, AI data centers can bring enormous demand to a single location and operate around the clock. The International Energy Agency expects global electricity consumption to grow at an average annual rate of 3.6% between 2026 and 2030, with industry, electric vehicles, air conditioning and data centres among the principal drivers. Data center consumption alone could more than double from 2024 levels to about 945 terawatt-hours by 2030. The interesting question, therefore, is not simply where the next megawatts will come from. It is who will be able to secure them, at what price and for which economic purpose.

AI Is Becoming a Serious Buyer in the Electricity Market

Electricity has rarely behaved like a conventional technology input. A software company can add servers, storage or cloud capacity relatively quickly, but electricity depends on generation, transmission, distribution and local system conditions that move on very different timelines. AI changes that relationship because compute expansion can translate directly into electricity demand. Accelerated servers, which are strongly associated with AI workloads, are projected by the IEA to grow their electricity consumption by about 30% annually through 2030 in its base case. 

That creates a new commercial dynamic. A large data center does not merely consume power; it can create a substantial, long-duration electricity load that can support investment in generation and related infrastructure. The customer can also provide utilities and power developers with a large source of electricity demand over an extended period, depending on the project’s operating profile and contractual arrangements. The customer is attractive precisely because the demand can remain high for years.

Other electricity users have different requirements. A manufacturer may need power to expand a plant. A transport system may require electricity across charging networks. Households may consume more during heat waves or cold weather. These demands do not necessarily arrive with the same commercial certainty as a hyperscale computing project. The result could be a market in which electricity increasingly flows toward the users capable of making the strongest economic commitment.

Electricity Could Become a Competitive Advantage for AI

The strategic advantage in AI may increasingly depend on access to power alongside access to chips, land and networking capacity, particularly as large data-centre projects require substantial electricity connections and new generation capacity. Recent developments in U.S. power markets illustrate why. PJM, the country’s largest electricity grid operator, recently faced a capacity shortfall in its 2028-29 auction as demand increased, particularly from data centers. The operator has responded with additional procurement mechanisms and efforts to facilitate bilateral arrangements between generators and large electricity users. Those developments point toward a more direct commercial relationship between large electricity users and power suppliers.

PJM’s bilateral matchmaking process is designed to connect new generation resources with large loads and allow parties to negotiate contract structures, including tenure and pricing.  This could also change how companies evaluate locations. A region with cheap land but limited power may become less attractive for data-centre development than a location with stronger access to electricity and a clearer path to additional generation and grid capacity. In that environment, electricity availability becomes an economic asset that can influence where computing clusters emerge. The consequence is subtle but important: electricity availability can become an economic consideration in determining where new data-centre capacity is developed, particularly where grid connection timelines and local supply constraints affect project schedules. They may increasingly compete to shape the supply arrangements around it.

Efficiency Will Matter, but Efficiency Does Not End the Competition

There is an obvious counterargument. AI hardware, software and data-centre infrastructure are also becoming more efficient, and the IEA’s scenarios explicitly model improvements in these areas as a major factor determining future electricity demand. That trend is real, and it could materially change the demand curve. The IEA expects efficiency improvements to offset some of the additional electricity requirements created by rising AI adoption. Its scenarios also show how dramatically the long-term outcome could vary depending on hardware efficiency, software optimization, AI uptake and infrastructure constraints. But efficiency does not necessarily translate into lower total electricity consumption.

If efficiency improvements lower the electricity required for individual AI services, the resulting reduction in energy use could be offset in part by stronger adoption and greater use of AI services. The IEA’s scenarios demonstrate how changes in AI uptake and efficiency can produce substantially different levels of future data center electricity demand. The same phenomenon has appeared across other technologies: efficiency can reduce the cost of consumption while encouraging more consumption. For electricity markets, that distinction matters more than the headline efficiency figure. The relevant question is not how many joules an AI task saves, but how much additional demand those savings unlock.

The Next AI Infrastructure Race May Be About Economic Priority

The emerging competition for electricity does not require an energy crisis to become significant. It only requires enough competing demand for electricity to become more valuable at specific locations and times. That is already becoming plausible. The IEA expects global electricity demand to grow substantially faster during 2026-30 than it did over the previous decade, with data centers joining industrial activity, cooling, electrification and electric vehicles as major sources of incremental demand. The implication for AI infrastructure is uncomfortable because it removes the assumption that every desirable computing project can simply obtain the power it needs.

The winning system may not be the one with the largest cluster of accelerators. It may instead be the one that can secure sufficient electricity at a cost that allows the resulting compute capacity to operate economically. The AI boom is therefore making electricity access an increasingly important consideration for computing infrastructure. Data centers remain a relatively small share of global electricity demand, but their concentrated growth can have a much larger effect on individual power systems and locations. That is a much harder question, and it may ultimately determine how far the next phase of compute expansion can go.