The Hidden Power Bill of AI: Where All That Electricity Goes

Every time you ask a chatbot something, a machine somewhere gets a little warmer. You never see it, and you never pay for it directly, so it's easy to forget it exists. But somebody is paying, and the bill is getting big enough that grid planners have started to lose sleep over it.

The Scale

Start with the scale. Data centers used about 1.5 percent of the world's electricity in 2024. That sounds small, and honestly it is.

The interesting part is the direction. In 2025, their electricity demand grew 17 percent, more than five times faster than overall global demand. Over the same year, tech companies spent over $400 billion on AI infrastructure.

Where does it go next? The IEA expects data center demand to more than double by 2030, to around 945 TWh. For a sense of size, that is more than Japan's total electricity use today. Put differently, the world's data centers would use about as much electricity as an entire industrialized country.

A Lopsided Race

It's lopsided, too. The US and China account for almost 80 percent of the growth. In the US, data centers are projected to drive nearly half of all growth in electricity demand. The IEA even projects that by 2030, American data processing will use more electricity than making steel, cement, aluminium and chemicals combined.

Think about that for a second.

Servers outconsuming heavy industry.

Not a Doom Story, yet

Now the part that rarely makes the headlines: this is not a doom story, at least not yet. Data centers account for roughly a tenth of global demand growth to 2030, less than air conditioning, industrial motors, or electric vehicles. The AI power problem is real, but it is concentrated. It hits hard in a few places, like specific regions of the US, and barely registers in most of the world.

That concentration is the real story. A grid doesn't care about global averages. When a single campus asks for as much power as a small city, the local utility has to build, and building takes years. Meanwhile the chips arrive in months. That mismatch in timing is probably the most underrated bottleneck in the whole AI race.

Who Gets Power First

There's also a quieter question about who gets power first. Countries and regions that can offer cheap, fast, reliable electricity will attract the data centers. Places that can't will watch the investment go elsewhere. Energy used to be a background input to the digital economy. Now it's starting to decide where the digital economy lives.

I don't think there's a clean conclusion here, and I'd distrust anyone who claims one. The optimistic view says AI will help make grids smarter and offset its own footprint. The cautious view says efficiency gains tend to get eaten by more usage. Both can be true at once.

What I do know is this: next time an AI tool answers you in two seconds, remember there's a power plant somewhere that had to be ready for it.

Source: IEA, Energy and AI (2025)