In short
AI runs on computers in data centres that use electricity, water and hardware, so every request has a small environmental cost that adds up across billions of users. Teachers do not need technical expertise to teach this well: a few reliable facts, the difference between efficiency and sufficiency, and good habits modelled in class go a long way.
Key points
- Training builds a model once; inference is every time someone uses it. For popular tools, the energy of use keeps adding up for as long as the model is in service.
- Google reports 0.24 Wh for its median Gemini text prompt and OpenAI about 0.34 Wh for an average ChatGPT query; both are company figures, measured in different ways.
- Generating images uses far more energy than generating short text, and general-purpose models use more than task-specific ones.
- Shorter prompts, shorter answers and smaller models can cut energy substantially, according to a 2025 UNESCO and UCL report.
- A school AI policy can cover efficiency and sufficiency alongside data protection, age limits and assessment.
What do teachers need to know first?
Three ideas carry most of the topic: AI runs on physical computers that use electricity, water and materials; small costs per request add up across billions of requests; and both doing things more efficiently and doing fewer unnecessary things help. Everything else is detail you can look up in the glossary.
How does AI use energy?
AI uses energy twice: once to build a model and again every time someone uses it. Both happen in data centres, buildings full of specialised computers.
- Training is the one-off process of building a model by showing it huge amounts of data. For large models it takes months of computing. Meta's model card for its Llama 3.1 family, for example, reports 39.3 million GPU hours of training and 11,390 tonnes of CO₂-equivalent on a location-based measure.
- Inference is using the trained model: each question, answer or image. Each use is small, but a popular tool is used billions of times, so the energy spent on use keeps growing for as long as the model is in service.
- The building around the computers also uses energy, mainly for cooling. Industry measures this with PUE.
Language models read and write in tokens, chunks of text that are often parts of words. OpenAI's rule of thumb for English is that one token is about four characters, or about three-quarters of a word. More tokens in and out means more computing, which is why long answers cost more than short ones.
What facts can I use in class with confidence?
These figures come from named, public sources. Quote them with the source and the year, because the numbers change quickly.
| Fact | Source |
|---|---|
| Data centres used about 415 TWh of electricity in 2024, about 1.5% of world electricity use | IEA, Energy and AI, 2025 |
| That is expected to more than double to about 945 TWh by 2030, slightly more than Japan uses today | IEA, 2025 |
| Data centre electricity use has grown by about 12% a year since 2017 | IEA, 2025 |
| Google's median Gemini text prompt used 0.24 Wh and about 0.26 mL of water (roughly five drops) | Google, 2025 |
| Google reports a 33-fold cut in energy per median prompt over one year | Google, 2025 |
| OpenAI says an average ChatGPT query uses about 0.34 Wh, like an oven running for a little over a second | OpenAI, 2025 |
| In Ireland, data centres used 22% of all metered electricity in 2024, up from 5% in 2015, more than all urban homes | Central Statistics Office, Ireland, 2025 |
| Generating 1,000 images with the least efficient model tested used 11.49 kWh, as much as 522 smartphone charges | Luccioni, Jernite and Strubell, 2024 |
| Shorter prompts and answers can cut energy by more than half; smaller task-specific models by up to 90% in tests | UNESCO and UCL, 2025 |
Two cautions when using them. The Google and OpenAI figures are company self-reports with different methods, so they are not a like-for-like comparison. And the per-prompt figures are for text; images, video and "reasoning" modes that think through many steps use more.
How can I use AI in school responsibly and efficiently?
Use AI where it saves real time or adds real value, and use it in a way that wastes as little as possible. The habits that save energy are mostly the same ones that get better results.
- Ask whether AI is the right tool. A worksheet you already have, a search, or a calculator may do the job. This is sufficiency.
- Write one clear prompt rather than five vague ones. Say who the output is for, how long it should be and what format you need.
- Ask for short answers when short is enough. The UNESCO and UCL report found shorter prompts and responses can cut energy by more than half.
- Choose a smaller or simpler tool for simple tasks, and avoid image or video generation unless it serves the lesson. See choosing the right model.
- Reuse good outputs. Save prompts and resources that work and share them with colleagues instead of regenerating them.
- Turn off AI features you do not use in office and learning software where settings allow.
The use AI efficiently page has more detail for adults.
What should a school AI policy say about green AI?
A school AI policy usually covers safeguarding, data protection and assessment. Adding a few lines on efficiency and sufficiency costs little and models the thinking you want pupils to learn.
Points most school policies already need (England's Department for Education guidance, updated 2025, covers these; check your own country's equivalent):
- Do not put pupils' or staff's personal data into generative AI tools unless the tool and its use have been approved and protected.
- Follow the age limits set by AI tools. UNESCO's 2023 guidance also discusses minimum ages for classroom use.
- Staff remain responsible for checking AI-generated content for accuracy and suitability.
- Set clear rules for AI in homework, coursework and exams, in line with your exam boards.
- Do not allow pupils' original work to be used to train AI models without permission.
Green AI points to add:
- Use AI where it clearly helps learning or saves staff time, not by default.
- Prefer tools and suppliers that publish information about energy, emissions and where their data centres are. The sustainable AI procurement page lists questions to ask.
- Avoid switching on AI features across the school that nobody has asked for.
- Include AI's environmental cost in staff training and in pupils' digital literacy lessons.
- Review the policy each year, because both the tools and the evidence change quickly.
What do pupils often get wrong?
Pupils, and adults, tend to swing between "AI uses no energy, it's just on my phone" and "one prompt boils the ocean". The truth sits between, and the gap is a good teaching opportunity.
| Misconception | Better explanation |
|---|---|
| "AI happens on my phone, so it uses hardly any power." | Most chatbots run in distant data centres. Your phone sends the request and shows the answer. |
| "One prompt uses loads of energy." | A typical text prompt uses a fraction of a watt-hour. The issue is scale: billions a day, plus images and video. |
| "If it gets more efficient, the problem is solved." | Cheaper, faster AI tends to get used more, which can cancel savings. This is the rebound effect. |
| "AI is only chatbots." | AI also covers spam filters, recommendation systems, weather forecasting and medical imaging, many of them small and efficient. |
| "Green AI means not using AI." | Green AI means using AI well: efficiently, and where it adds value. |
| "Data centres run on electricity, so there is no water use." | Many data centres use water for cooling, and some regions are short of water. |
| "The cloud is clean because it's in the sky." | "The cloud" is buildings, cables and computers on the ground, using real power and land. |
Researchers have also shown that efficiency gains in computing have often been followed by more total use rather than less (Freitag and colleagues, 2021). That is why sufficiency matters alongside efficiency, and why the efficiency is not sufficiency article is worth reading.
What to do next
- Try one activity from the lesson ideas: the paper tokens activity needs no technology.
- Share Green AI for pupils with a class.
- Bookmark the glossary and the FAQ for quick answers to pupils' questions.
- Raise the green AI policy points above at your next policy review.
- Look at the courses directory for professional development and for pupils thinking about university.
Frequently asked questions
How much energy does one AI prompt use?
Company figures for a typical text prompt are a fraction of a watt-hour: Google reports 0.24 Wh for its median Gemini text prompt and OpenAI about 0.34 Wh for an average ChatGPT query. Longer answers, images, video and reasoning-heavy models use more.
Should pupils use AI chatbots in school?
That is a decision for each school, within national guidance and the age limits the tools themselves set. England's Department for Education asks schools to follow those age limits, protect personal data and supervise pupil use.
Do I need to be a computing specialist to teach green AI?
No. The core ideas are about energy, scale and choices, which any teacher can explain. The glossary on this site defines the technical terms in plain English.
Is it true that AI uses lots of water?
Data centres use water, mainly for cooling, and the amount depends heavily on the site and cooling method. Google reports about 0.26 mL for its median Gemini text prompt, roughly five drops, but totals across all users and all facilities are much larger and less well reported.
Sources
- Energy and AI, International Energy Agency (2025)
- Measuring the environmental impact of delivering AI at Google scale, Google (arXiv:2508.15734) (2025)
- The Gentle Singularity, Sam Altman (OpenAI) (2025)
- Power hungry processing: watts driving the cost of AI deployment?, Luccioni, Jernite and Strubell (arXiv:2311.16863) (2024)
- AI large language models: new report shows small changes can reduce energy use by 90%, UNESCO (2025)
- Llama 3.1 8B model card, Meta (Hugging Face) (2024)
- What are tokens and how to count them?, OpenAI Help Center (2025)
- Data centres metered electricity consumption 2024, Central Statistics Office, Ireland (2025)
- Generative artificial intelligence (AI) in education, Department for Education (2025)
- Guidance for generative AI in education and research, UNESCO (2023)
- The real climate and transformative impact of ICT: a critique of estimates, trends, and regulations, Freitag et al., Patterns (2021)
Related
Green AI lesson and activity ideas
Nine low-cost green AI activities for ages 9 to 18: energy estimates, paper tokens, data centre siting, debates and a rebound game. No paid tools.
Green AI for pupils
A short, friendly guide for young people: what green AI is, why AI uses energy and water, smart habits, and how to find out more.
Use AI efficiently: everyday habits
Everyday habits that cut the energy of using AI: clear prompts, fewer regeneration loops, care with images and video, caching and turning off defaults.
How to write a green AI policy
A step-by-step guide to writing an organisation's green AI policy, with a template outline covering use, models, measurement, buying, claims and review.