By 2030, the world’s data centres may use about as much electricity in a year as the entire nation of Japan uses today. Not a Japanese city. Not one Japanese industry. All of Japan. 

The comparison comes from the International Energy Agency’s April 2025 report, Energy and AI, which projected around 945 terawatt-hours by 2030, slightly more than Japan’s electricity use. In April 2026, the agency updated that central projection to 950 TWh.

When the original report appeared, IEA executive director Fatih Birol put it plainly: “Global electricity demand from data centres is set to more than double over the next five years, consuming as much electricity by 2030 as the whole of Japan does today.”

The number that stops you cold

Scale is hard to feel. A hyperscale AI-focused data centre can, by IEA analysis, use as much electricity as 100,000 households. One building. And the plan, as Birol framed it, is not for one. “There is no AI without energy – specifically electricity,” he told world leaders in February 2025. “Thousands of data centres are set to be built in the next five years.”

The starting point makes the jump clearer. Global data centres used about 485 terawatt-hours in 2025. Reaching 950 by 2030 would mean almost doubling in five years. At that total, data centres would account for around 3% of global electricity demand.

Why data centres got so hungry so fast

The IEA points to AI as the most important force behind the surge. Its updated outlook says electricity consumption from AI-focused data centres will triple between 2025 and 2030. Birol calls this a big shift: “AI is one of the biggest stories in the energy world today – but until now, policy makers and markets lacked the tools to fully understand the wide-ranging impacts.”

The pace is the part that gets me. Data centre electricity demand grew by 17% in 2025, while electricity use at AI-focused data centres surged 50%.

And the expansion is concentrated. The IEA’s 2025 modelling found that the United States and China together could account for nearly 80% of global data centre demand growth through 2030. This is not demand spread thinly across the planet. It is a handful of places building very hard, very fast.

Where the power is supposed to come from

So where does 950 terawatt-hours actually come from? In its detailed supply modelling, the IEA projected that renewables would meet nearly half of the additional demand through 2030. Natural gas and coal together were expected to meet more than 40%, while nuclear power would become more important towards the end of the decade and after 2030.

Whether the grid can keep pace depends on choices no one has finished making. As Birol frames it, “AI is a tool, potentially an incredibly powerful one, but it is up to us – our societies, governments and companies – how we use it.”

The circle no one has squared

There is a tension here I cannot shake. Measured per task, AI is becoming sharply more efficient. The IEA says energy use per AI task has recently been falling by at least an order of magnitude annually.

And yet total demand keeps climbing.

That is not necessarily a contradiction. It is an old pattern with a new face, the one economists call Jevons’ paradox: make something cheaper to use, and people may use so much more of it that total consumption rises anyway.

A paper by Alexandra Sasha Luccioni and colleagues puts the worry directly, arguing that “rebound effects undermine the assumption that improved technical efficiency alone will ensure net reductions in environmental harm.” In plain terms, making the technology more efficient may not cut its total environmental cost if cheaper use leads to much more use. 

Which leaves us here. The same technology being sold partly on its promise to make energy systems smarter and leaner is driving one of the fastest demand surges the sector has seen. Efficiency is winning per task and losing in total. As far as I can tell, the outcome is not fixed yet.