AI2Day

Your ChatGPT question uses about as much electricity as a kettle boiled for one second

AI data centres ate roughly 0.5% of the world's electricity in 2025. That sounds small until you look at where those data centres actually are.

AI2Day NewsdeskAI-assistedPublished Updated Editor: Lee Brown3 min read
Illustration: a darkened operations center with multiple glowing monitors displaying abstract network topology maps
Illustration made with AI. Not a photograph of the events described.
Share

Key points

  • The International Energy Agency estimates global data centres used around 485 terawatt-hours of electricity in 2025, roughly equal to Germany's entire annual electricity output.
  • AI-focused data centres accounted for about one-third of that total, meaning AI consumed approximately 0.5% of the world's electricity in 2025.
  • A single text query to a large language model such as Google Gemini uses around 0.24 watt-hours of electricity, less energy than a kettle uses in one second.
  • In Virginia, data centres consume more than one quarter of the state's total electricity supply.
  • The IEA projects data centres could reach 3% of global electricity demand by 2030, with AI facilities driving most of that growth.

Every time you ask ChatGPT to write a work email or help plan dinner, a server somewhere burns a tiny slice of electricity to answer you. Google, one of the first companies to publish this kind of detail, released figures in 2025 showing a standard text query to its Gemini model uses around 0.24 watt-hours. That's less than what a kettle uses in a single second.

So individual queries are cheap. The bigger question is what happens when hundreds of millions of people ask questions all day, every day.

According to the International Energy Agency (IEA), global data centres, the large warehouse-style buildings packed with servers that run our digital services, used about 485 terawatt-hours of electricity in 2025. That's roughly what Germany generates in a whole year, and about 1.5% of all electricity produced on Earth.

Not all of that is AI. Email, Netflix, online banking and streaming all live in data centres too. AI-specific facilities account for roughly one-third of the total, putting AI's own share at around 0.5% of global electricity in 2025. Meaningful, but smaller than most people assume.

Should I feel guilty about using AI tools?

Probably not about your individual queries, but geography matters more than the global headline. Data centre demand is heavily concentrated in a handful of places, and that's where the real pressure lands. Ireland's data centres already account for more than 20% of its national electricity supply. Virginia's share is more than one quarter. Even if the worldwide figure looks manageable, local grids in those spots are already feeling the strain.

We first reported on the collision between AI power demand and strained grids on 21 July 2026, when a BloombergNEF forecast suggested AI data centres could consume a fifth of all U.S. Electricity by 2035. This IEA snapshot shows why that trajectory matters at street level, not just in aggregate.

Growth is the other pressure point. The IEA's base-case forecast puts data centres at 3% of global electricity by 2030, with AI facilities responsible for most of the increase. Some analysts think even that figure is too conservative.

One thing worth knowing: the energy figures above cover what happens on the server. The electricity your phone or laptop uses to run an AI app is counted separately and isn't included here.

Your occasional ChatGPT question isn't going to tilt the energy grid. The pressure comes from the sheer scale of infrastructure being built, not from anyone hitting send.

Latest Explained stories

Illustration: A close-up, top-down view of a neat physical desktop with stacks of paper being methodically sorted
Explained

AI Agents Could Soon Tidy Up Their Own Memory Mid-Task

A new system from Apple ML Research lets an AI agent decide for itself when to compress its own working memory, stopping it from forgetting important details at the worst possible moment.

Oct 12
Illustration: Soft natural light from the left
Explained

AI Translation Judges Struggle With the Same Errors Human Reviewers Do

A new study from Apple ML Research finds that large language models are only a partial replacement for human graders when checking the quality of machine-translated text, with both groups stumbling on the same hard cases.

Oct 12
Illustration: a single yellowing citrus leaf on a weathered wooden surface, next to a small circuit board
Explained

A $16 AI Training Run Built Six Models Nobody Had Made Yet

A Hugging Face tool called ML Intern let one developer train, evaluate and publish six specialist AI models in a couple of days, spending less than the cost of a restaurant meal.

Oct 8
Illustration: A vast server room with long rows of glowing blue and white rack-mounted computers receding into the distance
Explained

Apple researchers built a smarter way to train AI agents. Here is why the boring details matter.

A new training method from Apple ML Research teaches AI agents to focus on the lessons they have not learned yet, skipping exercises that are too easy or too hard. It's quietly important for anyone who cares about where general-purpose AI is heading.

Oct 7
© 2026 AI2Day