How AI Is Changing Data Centers and Electricity Demand

How AI Is Changing Data Centers and Electricity Demand

A single AI chatbot request can use more electricity than a traditional web search, although the amount varies significantly by model and task. Multiply that across billions of requests and add the computing power needed to train and run increasingly capable AI models, and the result is a growing demand for electricity that is already affecting data-center planning, power generation and grid infrastructure in some regions.

This article explains how AI is changing data centers, how extra electricity that shift actually requires, where the strain is showing up first and what the industry is doing to keep up.

Why AI Uses So Much More Electricity Than Regular Computing

Traditional data centers mostly handled things like websites, email and business software. These tasks are relatively light on power compared to what AI requires. Training an AI model involves running enormous numbers of calculations across thousands of specialized chips continuously for weeks or months. Once a model is trained every single response it generates, known as inference also consumes electricity and AI-specific chips called GPUs draw more power than the standard processors used in older data centers.

According to the International Energy Agency (IEA) a typical Google search uses 0.3 watt-hours of electricity while a single request to a large AI model can use several times that amount. When that difference is multiplied across billions of queries the additional electricity demand becomes significant even before accounting for the separate much larger power draw needed to train new models in the first place.

The good news is that efficiency is improving quickly. The IEAs 2026 analysis found that power consumption per individual AI task has been falling by least an order of magnitude every year a pace the agency describes as unprecedented, in the history of energy technology.. This efficiency gain is being outpaced by how fast AI usage itself is growing so total electricity demand keeps rising even as each individual task becomes cheaper to run.

The Numbers Behind the Surge

Independent estimates vary somewhat, but the overall trend from major research organizations tells a consistent story.

The IEA's April 2026 report, Key Questions on Energy and AI, found that global electricity consumption from data centres increased by 17% in 2025. Electricity consumption from AI-focused data centres grew even faster, increasing by 50% during the year. The figures highlight how quickly AI is becoming an important driver of data-centre electricity demand.

The IEA projects that global electricity consumption from data centres will roughly double from about 485 terawatt-hours (TWh) in 2025 to around 950 TWh by 2030. That would represent about 3% of global electricity demand by the end of the decade.

This rapid growth is also being supported by enormous investments in AI infrastructure. Major technology companies are increasing spending on data centres, chips and other infrastructure needed to support AI services, with investment expected to remain high as companies expand their computing capacity.

Where the Strain Is Showing Up First

Regional Power Grids

Some regions have become AI data center hubs and local grids feel the pressure directly. Northern Virginia is home to one of the concentrations of data centers in the world. In Northern Virginia the availability for new data center facilities is extremely tight. In Northern Virginia rents for data center space have been rising for years.

Ireland provides one of the clearest examples of the growing impact of data centres on electricity demand. Ireland's Central Statistics Office reported that data centres accounted for 23% of total metered electricity consumption in 2025. The rapid growth has made data-centre connections and electricity infrastructure an important issue for the country's energy system.

The United States Faces a Growing Power Shortfall

In the United States, the rapid expansion of data centres is creating new challenges for electricity generation, transmission and grid connections. In some regions, available grid capacity and the time required to build new infrastructure are becoming important constraints for planned data-centre projects. Access to reliable electricity is therefore becoming an increasingly important factor when companies choose locations for new data centres.

Big Tech Is Turning to Nuclear Power

Faced with the need for round-the-clock electricity several major technology companies have signed long-term deals with nuclear power operators. One known example is Microsoft’s 2024 agreement with Constellation Energy. Under this deal Microsoft agreed to restart Three Mile Island Unit 1 in Pennsylvania. This reactor is separate from the one involved in the 1979 meltdown.

The 20-year deal means Microsoft will buy 100% of the plant’s 835 megawatts of output. Constellation says that amount is equivalent, to the electricity used by 800,000 homes. That matches the power consumed by Microsoft’s data centers across states.

The plant, now renamed the Crane Clean Energy Center received a loan in 2025. It is now expected to return to service in 2027.

Microsoft is not alone. Amazon has signed an agreement tied to the Susquehanna nuclear plant also in Pennsylvania and other major cloud providers have pursued deals involving smaller next-generation reactor designs. These moves reflect growing interest in reliable, around-the-clock power sources for data centres. Solar and wind can provide substantial amounts of electricity, but their variable output means they may need to be combined with storage, transmission, other generation or grid resources when continuous power is required. Nuclear power is one option being considered because it can provide steady electricity.

Cooling Adds Another Layer of Demand

Powering the chips is one part of the picture. Focused data centers produce a lot of heat and keeping servers at safe temperatures needs much more energy and often water. The IEA says computing power and cooling are the two energy-consuming processes inside a modern data center. People see that many operators are shifting toward cooling systems. Liquid cooling systems include to-chip and immersion cooling. Liquid cooling systems can handle the heat density of AI chips better, than traditional air-based systems. Liquid cooling systems can also cut water use a lot compared to cooling methods.. The electricity needed for cooling still remains.

What This Means for Everyday Electricity Users

The amount of power that AI needs is starting to appear in electricity bills and decisions about how to manage the electric grid not just in big company energy deals. In areas where there are a lot of data centers utility companies have had to plan for power lines more electricity generation and in some cases changes to rates to pay for upgrading the grid. Operators who manage the grid in places with lots of data centers have also seen very high electricity use during hot weather showing that the power used by data centers is now big enough to have a real effect on the whole system especially when cooling systems are also using a lot of power.

This has made the energy use of AI a topic of discussion. People living near places where new data centers or nuclear plants are being built have asked questions about electricity costs, water use and how fast new projects are getting approved. At the time people who support this kind of investment say that this wave of spending especially on nuclear power could make the grid more reliable and help get rid of older dirtier power plants. Companies, like Microsoft have said these deals are part of their plans to reduce carbon emissions.

How the Industry Is Trying to Keep Up

Several strategies are emerging as the industry works to meet AI's growing power needs without overwhelming existing grids.

Questions People Often Ask

Does using ChatGPT or a similar AI tool really use that much more electricity than a Google search? Per single request, yes, based on IEA's comparison of a typical search against a typical large-model AI query. But context matters. Individual queries are still a small fraction of total electricity use compared to the much larger amount consumed during AI model training, and per-task efficiency is improving quickly as chips and software get better. The bigger driver of total demand is the sheer scale and frequency of AI usage across billions of users, not any single request.

Is AI actually making data centers worse for the environment? It depends on how the additional electricity is generated. Rising absolute electricity demand from AI data centers is a real trend, and disclosed emissions from some hyperscalers have been increasing even as they buy more renewable energy. At the same time, several major operators are directing new investment toward carbon-free sources like nuclear power specifically to meet AI's around-the-clock demand, which supporters argue could help offset some of the added strain over time. The full environmental picture depends on which power sources end up meeting the new demand in each region.

Will electricity prices keep rising because of AI data centers?In areas with data centers utilities have already had to plan for grid upgrades. Those upgrades can show up as electricity rates over time. I wonder whether prices will rise across the country. That question is more complicated. It depends on grid capacity how fast new generation becomes available and on rules, about how infrastructure costs are shared between data center operators and other electricity customers.

Are smaller AI models a realistic way to reduce this demand?  Efficiency gains are becoming a tool in the industry’s effort to slow the rise in electricity use per AI task. These gains come from building more focused models that handle specific jobs instead of relying on huge all-purpose systems. Still even though these models use power the overall demand keeps growing faster. That’s because AI is being adopted quickly across industries. So while efficient models help slow the climb they haven’t stopped the trend, in energy use. The growth is still happening, a little bit more slowly than it would without these improvements.

The Bottom Line

AI has changed the way data centers operate. Once a small part of the internet’s infrastructure they are now among the fastest-growing users of electricity around the world. This shift is not temporary. We’ve seen a 50% increase in electricity use by AI-focused data centers within one year. Major companies are making multibillion-dollar deals for power to keep up. These developments show that the energy needs of computing have fundamentally changed. It's not a short-term spike. It's a long-term change.

Efficiency improvements are helping reduce the energy required for individual AI tasks, but rapid adoption and the emergence of more energy-intensive AI applications are increasing overall demand. For technology companies, utilities and governments, the challenge is to expand electricity supply and grid capacity while continuing to improve the efficiency of AI hardware, software and data centres..

It affects how electricity grids are planned. It influences how power is priced. It changes how new power plants and transmission lines are built. This trend isn't going away soon. Experts believe it will continue at least through the end of the decade.

Sources

• International Energy Agency (IEA) — Key Questions on Energy and AI https://www.iea.org/reports/key-questions-on-energy-and-ai/executive-summary
• Ireland Central Statistics Office — Data Centres Metered Electricity Consumption 2025 https://www.cso.ie/en/releasesandpublications/ep/p-dcmec/datacentresmeteredelectricityconsumption2025/
• Constellation Energy — Crane Clean Energy Center and Microsoft Agreement https://www.constellationenergy.com/news/2024/Constellation-to-Launch-Crane-Clean-Energy-Center-Restoring-Jobs-and-Carbon-Free-Power-to-The-Grid.html
• Constellation Energy — Crane Clean Energy Center Restart Update https://www.constellationenergy.com/about/locations/crane-clean-energy-center.html
• CBRE — Northern Virginia Data Center Market in 2025 https://www.cbre.com/press-releases/northern-virginia-extends-lead-as-largest-u-s-data-center-market-in-2025
• U.S. Energy Information Administration (EIA) — Commercial Electricity Sales Have Soared in Virginia, Driven by Data Centers https://www.eia.gov/TODAYINENERGY/detail.php?id=67664
• U.S. Department of Energy — Data Centers
• https://www.energy.gov/