AI's Power Bill: By 2030, Data Centres Could Use as Much Electricity as Japan
One ChatGPT question uses about 0.34 Wh. But data centres could double to 950 TWh by 2030 – Japan's yearly use. AI's power bill in 4 numbers, explained.
Watch the Short
The 60-second version is on YouTube. This article is the full story.
New Shorts every week on tech trivia, AI and gadgets.
Subscribe to TheTechPranaKey takeaways
- One average ChatGPT question uses about 0.34 watt-hours, according to OpenAI – roughly an oven running for a second.
- The world's data centres used about 485 TWh in 2025, around 1.5% of all electricity. AI-focused ones grew 50% in that one year.
- The IEA expects data centres to reach about 950 TWh by 2030 – roughly double, and about as much as Japan uses in a year.
Every time you ask an AI chatbot a question, a computer somewhere in a data centre does the work – and that computer runs on electricity. One question uses almost nothing. Billions of them, plus everything else data centres do, add up to one of the fastest-growing power bills on the planet.
The Short above counts it down in four numbers. Here’s what each one means.
Number 4: 0.34 watt-hours – one ChatGPT question
In June 2025, OpenAI’s CEO Sam Altman wrote that an average ChatGPT query uses about 0.34 watt-hours of electricity. In his comparison, that’s roughly what an oven uses in a little over one second, or a high-efficiency light bulb in a couple of minutes.
That’s tiny. Two caveats, though:
- It’s OpenAI’s own figure, and it hasn’t been independently checked.
- It’s an average text question. Making images or video, or running long “reasoning” tasks, can use far more.
Number 3: 485 TWh – all data centres in 2025
Zoom out from one question to every data centre in the world. These buildings run everything online: cloud storage, email, video streaming, banking, gaming and AI.
According to the International Energy Agency (IEA), data centres used about 485 terawatt-hours (TWh) of electricity in 2025. That’s around 1.5% of all the world’s electricity, and roughly what Germany generates in a year.
It’s been rising fast: from about 360 TWh in 2023 to 416 TWh in 2024, then 485 TWh in 2025.
Not all of that is AI. The IEA estimates that AI-focused data centres used about 155 TWh in 2025, with the rest going to traditional workloads.
Number 2: +50% – AI’s growth in a single year
Here’s the number that shows where things are heading. Electricity use at AI-focused data centres jumped by about 50% in 2025 alone.
AI chips are hungry. Training a large model means running thousands of chips flat out for weeks, and every chatbot answer, generated image and AI search result needs more computing afterwards.
Number 1: 950 TWh – data centres by 2030
In its 2026 report, Key Questions on Energy and AI, the IEA projects that data centres will use about 950 TWh by 2030. That’s roughly double 2025, and about 3% of the world’s electricity.
For scale, the IEA’s earlier report described its similar 945 TWh projection as roughly as much electricity as Japan uses in a year.
And the AI-focused data centres? They’re expected to more than triple, to around 465 TWh by 2030.
Where is all this power going?
The growth is concentrated in two countries. The IEA expects the United States and China to account for almost 80% of the growth in data-centre electricity demand to 2030.
- United States: about 224 TWh in 2025, projected to reach about 426 TWh by 2030. Data centres already use about 5% of US electricity.
- China: about 117 TWh in 2025, projected to reach about 277 TWh.
Data centres also cluster in a handful of places, which can strain the local grid even when the national share looks small.
Where will the electricity come from?
The IEA expects a mix. Renewables should supply the largest single share of the new demand, but natural gas and nuclear each add a large amount too. In the US, slow grid connections are pushing some data-centre builders to set up their own gas-fired power plants on site.
How sure are these numbers?
They’re projections, not certainties. The IEA runs several scenarios for 2030:
| IEA scenario | Data centres in 2030 |
|---|---|
| Lift-Off (faster AI growth) | ~1,008 TWh |
| Base Case | ~945–950 TWh |
| High Efficiency | ~868 TWh |
| Headwinds (slower growth) | ~833 TWh |
Even the slowest scenario sees data-centre electricity use rise by well over 60% from 2025.
So, is AI worth the power bill?
There’s no single answer. A single question costs about as much energy as a second of oven time, and the IEA also argues that AI could help the energy sector run more efficiently. But the totals are growing fast, and the power has to come from somewhere.
Quick recap:
- 0.34 Wh: one average ChatGPT question (OpenAI’s figure)
- 485 TWh: all data centres in 2025, about 1.5% of world electricity
- +50%: growth at AI-focused data centres in 2025
- 950 TWh: the IEA’s 2030 projection – roughly Japan’s yearly use
Your turn: is AI worth the power bill? Tell us in the YouTube comments.
Sources
- IEA: Key Questions on Energy and AI (2026), executive summary
- IEA: Energy and AI (April 2025), executive summary
- Our World in Data: How much energy do data centers and artificial intelligence use?
- Industrial Info: IEA data centers pose challenge to global power industry
- Scientific American: AI will drive doubling of data center energy demand by 2030 (10 Apr 2025)
- Simon Willison, quoting Sam Altman's 'The Gentle Singularity' (10 June 2025)

