How Much Electricity Does AI Really Use?

In recent years, a single AI task has needed less and less energy. A lot less.

And yet, in the International Energy Agency's central projection, the electricity used by data centres roughly doubles by 2030.

How can both be true? And how much electricity does AI really use?

Let's look at the numbers. Only the numbers.

This is Chevlume. The signal, not the noise.

Start with the big picture. In 2025, data centres around the world used roughly 485 TWh of electricity, up about 17% in a single year.

It sounds enormous, but global electricity demand is enormous too. Those 485 TWh are only slightly more than 1.5% of the world's electricity demand.

And here's the detail that matters. That 1.5% is all data centres, not just AI.

So the global figure covers data centres as a whole. AI is inside it, and it's growing much faster than the rest.

Because while data centres overall grew about 17% in 2025, AI-focused data centres grew about 50%.

In that one year, data centres added about 70 TWh of demand. Global electricity demand, overall, grew 3%.

Now look ahead. In the IEA's central projection, data centre electricity use reaches about 950 TWh by 2030, roughly double. That would be around 3% of global demand. And in the same projection, consumption by AI-focused data centres triples.

That's the IEA's projection, not a measurement.

Zoom in on one country, and the share gets bigger. In the United States, data centres used 58 TWh in 2014. By 2023, they used 176. That's 4.4% of the country's electricity.

A report by Lawrence Berkeley National Laboratory for the U.S. Department of Energy projects 325 to 580 TWh by 2028, 6.7% to 12% of U.S. electricity.

Which brings us back to the paradox. The IEA says the energy efficiency of AI is improving at a rate unprecedented in energy history. Energy use per AI task has dropped by at least an order of magnitude a year in recent years. That's 10 times less every year.

So why does the total keep rising?

Because more people are using AI. And energy intensive uses, such as AI agents, are on the rise. Growth in use outweighs the efficiency gains.

Each task gets cheaper. There are simply many more tasks, and heavier ones.

So if the global share is small, why are power grids feeling the strain? Because electricity isn't used globally. It's used somewhere.

First, scale. A typical AI-focused data centre uses as much electricity as a hundred thousand households. The largest ones under construction will use 20 times that.

Second, location. In 2024, the United States accounted for 45% of global data centre electricity use. China, 25%. Europe, 15%. And nearly half of U.S. data centre capacity sits in just five regional clusters.

Third, the grid itself. In advanced economies, new transmission lines take four to eight years to build. Wait times for transformers and cables have doubled in three years, and about 20% of planned data centre projects could be delayed.

Grids are the bottleneck.

So how much electricity does AI really use? Here's the signal.

One. Today, all data centres together use slightly more than one and a half percent of the world's electricity. AI is a fast-growing part of that, not the whole.

Two. The direction. The IEA projects about 950 TWh by 2030, around 3% of global demand, with AI-focused use tripling.

Three. The place. Demand is concentrated in a few regions and clusters, on grids that take years to expand.

Small share. Fast growth. Local strain. Those are the numbers to watch.

All sources are in the description. Follow Chevlume.