Why AI needs so much electricity

Image: Illustration / Digiopedia 

Artificial intelligence is becoming one of the biggest new sources of demand for computing infrastructure. Behind every AI model is a network of data centers filled with processors that consume electricity, along with cooling, networking and power systems needed to keep those machines running.

The growing demand is not simply because AI models are large. People are using AI at enormous scale, while newer applications are becoming increasingly computationally intensive.

AI requires powerful computing

Most modern AI systems rely on specialized processors, including GPUs and other accelerators, that can perform huge numbers of calculations simultaneously.

Training a large model can involve thousands of processors working together for extended periods. Once training is complete, the model still requires computing resources every time someone uses it.

That second stage, known as inference, is becoming increasingly important as AI services are used by millions of people and incorporated into software, search, productivity tools and other products.

The amount of electricity required by an individual AI request can vary considerably. Recent research found that optimized frontier-model inference can use less energy per query than some widely circulated estimates suggest, but long reasoning and agentic tasks can require substantially more computation than ordinary requests. 

The data centers behind AI are getting bigger

AI is driving the construction of increasingly large data centers, and their electricity requirements are becoming significant enough to affect national and regional power planning.

The International Energy Agency estimates that global data-center electricity consumption was about 485 TWh in 2025 and could reach around 950 TWh by 2030. AI-focused data centers are expected to grow faster than the data-center sector overall. 

The IEA says data centers could account for around 3% of global electricity consumption by 2030. That is still a relatively small share of worldwide electricity use, but the demand is highly concentrated geographically, which can make the local impact much larger. 

In the United States, Lawrence Berkeley National Laboratory's 2025 update estimates that data centers could consume 11.8% of total U.S. electricity in 2030 in its reference case, with a modeled range of 9.5% to 15.3%. 

It isn't just the AI chips

The processors performing AI calculations are only part of the electricity equation.

Data centers also need networking equipment, storage, power-conversion systems and cooling infrastructure. The IEA estimates that servers account for around 60% of electricity consumption in modern data centers on average, with the remainder going to other equipment and infrastructure. (

Cooling becomes particularly important as AI systems become more densely packed. More powerful processors generate more heat, requiring increasingly sophisticated systems to keep equipment operating safely.

The U.S. Department of Energy has also highlighted a distinctive characteristic of large AI facilities: thousands of processors can operate together in coordinated cycles, creating large and rapidly changing electrical loads. 

Why AI demand keeps growing

There are two forces working at the same time.

AI is becoming more efficient, with improvements in chips, software and data-center infrastructure reducing the energy required for individual tasks.

But AI use is growing rapidly, and newer applications can require much more computation.

A simple text request is not equivalent to generating a high-resolution video, running a long reasoning process or operating an AI agent through multiple steps. Recent research found that long reasoning queries can consume many times more energy than standard queries. 

This creates a difficult equation for the industry: efficiency improvements can reduce the energy required for individual tasks, but rapidly increasing usage and more demanding applications can offset those gains.

The electricity has to come from somewhere

The growth of AI is increasingly becoming an infrastructure issue rather than just a technology issue.

Data-center developers need access to large amounts of reliable electricity, and connecting new facilities to the grid can take years. The IEA says bottlenecks involving electricity supply, grid connections, equipment and advanced chips are already affecting the pace at which AI infrastructure can expand. 

Recent developments are already changing where companies build data centers. Reuters reported that European developers are increasingly looking toward locations with cheaper and more readily available power as electricity and grid capacity become constraints in major technology hubs. 

The source of that electricity also matters. The IEA expects renewables to supply a significant portion of the additional electricity required by data centers, while natural gas and nuclear power are also expected to contribute. 

Technology companies are investing heavily in new clean-energy capacity. Google, for example, said in its 2026 environmental report that it signed agreements for more than 12 GW of new clean energy in 2025, while also acknowledging that the growth of AI infrastructure is putting additional pressure on electricity systems. 

AI is becoming an energy story

The important point is that AI does not consume huge amounts of electricity simply because an individual chatbot question requires enormous amounts of power.

The larger issue is scale.

Millions of users, increasingly powerful models, AI-generated media, reasoning systems, agents and continuously operating data centers all add to the total computing demand.

At the same time, hardware and software efficiency is improving rapidly. The question for the industry is therefore not simply how to use less electricity per AI task, but whether efficiency improvements can keep pace with the speed at which AI adoption and computing demand are growing.

For now, the trend is clear: the future of AI depends not only on better chips and models, but also on the ability to build enough computing infrastructure and supply it with reliable electricity.