Jump to a section
The short answer

Two things are true at once. A typical text question to a chatbot uses a tiny amount of energy, roughly a second of microwave use by one estimate, so your own everyday use is a very small part of your footprint. But the AI industry as a whole is growing fast: data centres' electricity use is expected to roughly double by 2030, and many are being built in places short of water. Video generation uses far more than text.

The short answer: it depends what you mean

Most arguments about AI and the environment talk past each other because they're about two different things: the cost of one person asking one question, and the cost of an industry building enormous data centres around the world. The first is small. The second is big and growing. Both matter, in different ways.

One question: smaller than you'd think

In August 2025, Google published its own estimate for a typical (median) text prompt in its Gemini app: 0.24 watt-hours of electricity, 0.03 grams of carbon dioxide equivalent, and 0.26 millilitres of water, about five drops. Google compares the energy to watching TV for less than nine seconds, and says the footprint of that median prompt fell 33-fold over the previous year.

Two important caveats: Google says the figures haven't been independently verified, and they cover text prompts only.

The environmental researcher Hannah Ritchie, who has looked at these estimates in detail, suggests a typical range of about 0.3 to 3 watt-hours per question, since Google's figure leaves out things like training. Even so, she calculates that ten prompts a day would be roughly 0.03% to 0.1% of an average American's or Briton's daily electricity use.

Pictures and video: a different story

Text is the light end. When MIT Technology Review measured open-source AI models in 2025, a standard image used a few thousand joules, a few seconds of microwave use. A five-second video clip used about 3.4 million joules: more than 700 times a high-quality image, and roughly an hour of microwave use.

So if you want to keep your AI footprint down, the biggest lever is simple: be choosy about generating video.

The big picture: data centres are growing fast

Add up everyone's questions, plus the energy used to build and train the models, and the totals get serious. The International Energy Agency estimates that data centres used about 415 terawatt-hours of electricity in 2024, about 1.5% of the world's total, and expects that to roughly double to around 945 terawatt-hours by 2030. It says electricity use in the specialised servers that run AI is growing by around 30% a year.

The US is at the centre of it. MIT Technology Review reported that US data centres used roughly 200 terawatt-hours in 2024, and that their share of US electricity could rise from 4.4% to about 12% by 2028. The IEA expects US data centre use to rise by around 240 terawatt-hours by 2030 compared with 2024.

What about water?

Data centres also use water, and where they're built matters as much as how much they use. A Bloomberg analysis, reported by TechRepublic, found that more than two-thirds of data centres built since 2022 are in water-stressed areas, including parts of Texas and Arizona.

TechRepublic also cites the IEA estimate that a 100-megawatt data centre can use up to 2 million litres of water a day, about as much as 6,500 households. That's why local communities, not just climate scientists, are paying attention.

Why the numbers are so fuzzy

One reason this debate is so confusing is that the biggest AI companies have shared very little. When MIT Technology Review did its research in 2025, OpenAI and Google both declined to share energy figures for their models, and researchers it spoke to said the available information wasn't enough to make reasonable projections. Google has since published its per-prompt estimate, but without independent checking.

So treat any single number, including the ones on this page, as an estimate rather than a fact carved in stone.

What you can actually do

If you care about this, and plenty of people reasonably do, these habits make the most difference without giving up the useful stuff:

  • Use AI where it genuinely saves you time, rather than as a toy you regenerate over and over.
  • Go easy on video generation, which uses far more energy than text.
  • Ask a good question first time, so you need fewer follow-ups.
  • Keep perspective: Hannah Ritchie's analysis suggests everyday chatbot use is a small part of most people's footprint.
  • If the bigger picture worries you, that's a question for companies and governments about where data centres are built and how they're powered, and it's fair to ask it.

How to ask an AI chatbot a good question (the four-part trick)

Sources

These sources support the guide. Tool settings, prices and policies change, so dated source boxes show when those details were checked.

AI helps with research and drafting. Dan checks the sources and is responsible for what is published.

How we check