What does a prompt cost?

1 · When a young person asks

"How much water does AI use?"

The honest answer: nobody knows exactly. And that is a good place to start.

💧 Where the water goes

AI runs in data centres. Many of them cool their servers with water, and part of that water evaporates.

🏢 Why the numbers differ

It depends on the model, the data centre, the climate and the time of year. And companies decide what they publish.

🔁 "But isn't it a closed loop?"

Some data centres cool with a closed loop that reuses the same water. They use much less water on site — but they need more electricity for cooling, and making electricity often uses water too, at power plants somewhere else. Many large data centres still cool by evaporation: that water goes up as steam and comes back as rain, but not to the same town's taps. So the honest answer is: it depends on how it is built. That is why it is worth asking.

Li et al., Communications of the ACM (on-site and off-site water) · UNU-INWEH 2026

500 ml

for roughly 10–50 answers from GPT-3, depending on when and where it runs

Li et al., Communications of the ACM

0.26 ml

for one median Gemini text prompt — about five drops, says Google

Google, August 2025

Energy is a little easier to estimate

Text prompt ~0.3 Wh

One image ~1.5 Wh

5-sec video ~940 Wh — about 3,000 text prompts

Epoch AI and MIT Technology Review · Hugging Face · Luccioni et al.

Try the AI energy calculator →

🌍 One year of ChatGPT

uses about as much water as 500,000 people need at home in sub-Saharan Africa.

UN University (UNU-INWEH) 2026

⚡ AI is one fifth

AI uses about one fifth of data-centre electricity. The rest is everything else we do online: video, photos, games, the cloud.

UNU-INWEH 2026

Why not just give the number?

A number alone makes a young person feel guilty about one prompt, while the big decisions are made somewhere else. In the 2000s, the oil company BP made the term "carbon footprint" popular to move attention from companies to individuals. Let's not do their work for them.

Say instead: "Nobody knows exactly. Let's find out together."

What young people in Europe say

40,000 young people answered

UNICEF asked young people in 21 countries in Europe and Central Asia (UNICEF & ETF, From Awareness to Action, 2024).

47.7 %

say there are not enough opportunities to express their views on environmental policies and climate action. Another 19.3 % don't know what opportunities exist.

→ You can show them.

61.3 %

have little or no confidence that their government will keep its climate promises. 20.3 % don't know what was promised.

→ Finding out is a good first task.

68.5 %

get climate information mainly from social media; 51.1 % from schools.

→ Check sources together.

14.6 %

name business and industry as responsible for tackling climate change.

→ Zoom out: who builds, who pays, who decides?

Will technological fixes alone solve the climate crisis? Young people are split.

30 % yes30 % not sure40 % no

→ AI is not only a problem or only a solution. Ask what it is used for.

2 · The three bags

Not "use less AI"

That tells her what to stop and gives her nothing to carry. Here are three bags to carry with her. Each bag is one session you can run.

Your carbon handprint

A footprint is the harm you cause. A handprint is the good you help others do. Youth work has a big handprint: when you help a group of young people ask, zoom out and find another way, the change does not stay with you — it spreads.

Bag 1

Ask why

Don't answer with a number. Ask back:

"Why do you ask?"

Her question is often not about water. It is about caring. Researcher Maria Ojala: worry is not the problem — worry is caring. Hope that grows from action helps young people; hope that says "everything will be fine" does not.

Session plan · 45 min

Ages 12+Any group size

10 minEveryone writes one question they have about AI and nature. No wrong questions.
10 minIn pairs: Why did you ask that? What do you care about behind the question?
15 minAsk an AI the same question. Then check the answer together: Where does this come from? Who could know better? What does the AI not say?
10 minCircle: one thing I care about, one thing I want to find out.

You need: paper, one device with an AI chat, the question cards.

Why do you ask?

What worries you?

What would you like to be different?

Who could know the answer?

How could we check it?

Bag 2

Zoom out

One question is small. The building behind it is not. Four questions for any data centre:

What is it for? Who pays? Who decides? Whose river?

🗺 Map: data centres planned in Finland

"But the water comes back." It does — as rain, somewhere else. Not to the same town's taps, because much of it evaporates. Even when some water is returned, large withdrawals can strain rivers and groundwater (UNU-INWEH 2026).

Session plan · 45 min

Ages 13+Groups of 3–4

10 minFind the nearest data centre on a map. Search "data centre" + your region, or use a public data-centre map.
See all 160 data-centre projects in Finland on one map →
Examples from Finland: EK's map of green investments (in English; choose Theme → Data center) and a Yle article about data-centre locations (in Finnish).
20 minIn groups, find answers to the four questions. Use news, the company's page, the town's page. Write down what you could not find. That is also a result.
15 minShare: What surprised you? Which question was hardest to answer, and why?

You need: devices with internet, a big paper with the four questions.

Warm-up idea: make memes about AI's environmental cost — see Meme-tivism under Stories.

Bag 3

Ask for another way

Remember the ozone hole? We didn't stop cooling. We changed the gas. The fridges kept working.

Data centres don't have to cool themselves with drinking water. There are other ways. None of them is free. But it's a choice. Help young people ask the people who make the choice:

"Could you do it another way?"

They don't always listen. That's why we ask together, and in public.

Session plan · 45–90 min

Ages 12+Small groups

15 minTake photos of one problem in your own street or area. (Young people in Colombia did this: trash, standing water.)
15 minAsk an AI for three strengths, three weaknesses and five questions about your idea for fixing it.
15 minFind out what your town or government has promised on climate. Then write one question to someone who decides — with the question builder.
optionalSend it. Agree who follows up, and when.

You need: phones for photos, one device with an AI chat.

3 · Write one question

One question to someone who decides

Pick, write, copy. Then translate it into your own language and send it.

Copied!

Your question

💡 A question from a group is harder to ignore than one from one person.

4 · Stories

What people did

What they did, what changed — and what you can borrow.

Montevideo, Uruguay · 2023

The salty taps

The city almost ran out of fresh water; tap water turned salty. At the same time, Google planned a data centre that would cool itself with 7.6 million litres of drinking water a day — the daily household water of 55,000 people. People took to the streets:

"No es sequía, es saqueo."
It's not drought, it's plunder.

What changedGoogle redesigned the data centre so it would not cool itself with drinking water.

BorrowThe slogan. Name the problem in five words.

DW · Mongabay

Muhos & Kajaani, Finland · 2026

Forest first, questions later

Google cut down over 500 hectares of forest — about 700 football fields — at its planned data-centre sites before the environmental assessment was finished. The authority asked Google to stop the work.

What changedThe work has to wait for the assessment, and Google has to answer questions. Will they continue? Probably. That is what asking does: it makes them answer in public.

BorrowFollow your local news and ask: was the assessment done first?

Yle, 8.10.2026

Germany · 2021

Young people went to court

Young people took the climate law to court — and won. The Constitutional Court said pushing emission cuts past 2030 puts the burden on the young, and ordered the law to be fixed.

What changedThe law had to be fixed.

BorrowYoung people have rights here. The UN says states must listen to children in climate and environment decisions (General Comment No. 26, 2023).

Neubauer et al. v Germany, Federal Constitutional Court

Cartagena, Colombia · 2026

Cameras, AI and the people who decide

Twelve teenage citizen scientists in El Pozón photographed environmental health problems in their own streets: trash, standing water. They used AI to test their ideas — three strengths, three weaknesses, five questions — and took them to the people who decide.

Outcome38 decision-makers came. They committed to bins inside the school, lids for the open holes the students had drawn with AI, and dengue workshops. Follow-up was short and informal, so lasting change isn't proven. AI worked best as a thinking aid, not as the decider.

BorrowThe whole method — it's Bag 3.

De la Vega-Taboada et al., JMIR AI 2026

Quilicura, Chile · 2026

A town became a "human AI"

Quilicura, near Santiago, is a data-centre hub in a long drought. For one Saturday, residents ran Quili.ai: a chatbot answered by people, not machines. The project says it answered over 25,000 prompts from 68 countries.

OutcomeNo measured change. It made the hidden water cost of AI visible around the world.

BorrowPlayful, public, together. quili.ai

AP, 31.1.2026 · Corporación NGEN

London & Berlin · 2024–2026

Meme-tivism

Workshops where people make memes about the environmental cost of AI, with prompt cards and meme templates. Humour makes a heavy topic easier to talk about — and powerlessness easier to name.

BorrowA meme session as a warm-up for Bag 2.

Sinem Görücü & Georgia Panagiotidou · Climate Action Tech

More examples from around the world: airesistlist.org

More for youth work

80 Perspectives on Artificial Intelligence

Reflections on AI for those working with young people. A guide by Jarno Alastalo, made with City of Helsinki Youth Services, Into and the Finnish Society on Media Education.

2025 · free PDF · CC BY 4.0 · English, Finnish and Swedish

Get the free guide →

Sources

UN University (UNU-INWEH) 2026: Environmental Cost of AI's Energy Use: Carbon, Water and Land Footprints
Li, Yang, Islam & Ren: Making AI Less "Thirsty", Communications of the ACM — doi.org/10.1145/3724499
Google 2025: Measuring the environmental impact of AI inference (median Gemini text prompt 0.24 Wh, 0.26 ml) — via MIT Technology Review
George et al. 2023: Water Consumption by ChatGPT — doi.org/10.5281/zenodo.7855594
O'Donnell & Crownhart 2025: We did the math on AI's energy footprint, MIT Technology Review — technologyreview.com
Bertazzini et al. 2025: The Hidden Cost of an Image — arxiv.org/html/2506.17016v1
Luccioni et al. 2023: Energy Use of Text-to-Image Models — arxiv.org/abs/2311.16863
Berthelot et al. 2024: Gen-AI Environmental Estimates — doi.org/10.1016/j.procir.2024.01.098
Epoch AI: AI energy estimates — epoch.ai · Our World in Data — ourworldindata.org
Google Sustainability Report 2024 · Microsoft Sustainability Report 2024
AI energy calculator — csy.fi/ai-energy-calculator
UNICEF & European Training Foundation 2024: From Awareness to Action — unicef.org
Ojala 2012: Hope and climate change, Environmental Education Research · Ojala 2023: Hope and climate-change engagement, Current Opinion in Psychology
Lundy 2007: 'Voice' is not enough, British Educational Research Journal
UN Committee on the Rights of the Child, General Comment No. 26 (2023)
Neubauer et al. v Germany, Federal Constitutional Court, 2021
De la Vega-Taboada et al. 2026: A New Model for Youth-Driven Community Change, JMIR AI — ai.jmir.org/2026/1/e79464
Yle 8.10.2026 and 13.10.2026: Muhos and Kajaani
Montevideo 2023: DW; Mongabay
AP 31.1.2026: A chatbot entirely powered by humans — via WTOP · quili.ai
Meme-tivism — Climate Action Tech · re:publica 2026
Alastalo: Musta laatikko (BP and the carbon footprint)

Jarno Alastalo

Techno-philosopher · Speaker · Author

"I made this talk with AI too: the code, some pictures, the voices. I don't know what it cost the planet. But the stories, the choices and the slowness were mine."

jarnoalastalo.com jarno@csy.fi