In short
These are short, sourced answers to the questions people most often ask about AI and the environment: how much energy AI uses, whether it is getting better or worse, and what individuals, organisations and governments can do. Each answer links to a fuller page where there is one.
Key points
- The IEA estimates data centres used about 415 TWh of electricity in 2024, about 1.5% of the world's total, and expects more than double that by 2030.
- A typical text prompt uses a fraction of a watt-hour on company figures; images, video and long agentic tasks use much more.
- Efficiency gains are real and fast, but the rebound effect means total use can still grow.
- Measurement is patchy: most per-prompt figures are self-reported and not directly comparable.
These are short answers to the questions people ask most often about AI and the environment. Every figure comes from a named public source, listed at the end of the page, and the numbers change quickly, so check the year. Where a question deserves more than a few sentences, follow the links to the guides, good practice pages and articles. Technical terms are explained in the glossary.
If your question is not here, ask us.
Frequently asked questions
What is green AI?
Green AI is developing and using artificial intelligence in a way that keeps its energy, carbon, water and material footprint as low as possible, and using it where it adds real value. It covers both efficiency (doing the same with less) and sufficiency (asking whether AI is needed at all).
How much electricity does AI use?
Nobody publishes an exact figure for AI alone. The International Energy Agency estimates that all data centres used about 415 TWh in 2024, around 1.5% of global electricity, and names AI as the most important driver of growth.
Will AI's energy use keep growing?
On current trends, yes. The IEA expects data centre electricity use to more than double to about 945 TWh by 2030, slightly more than Japan's total consumption today, and data centres to account for around a tenth of global electricity demand growth to 2030.
How much energy does one ChatGPT or Gemini prompt use?
OpenAI says an average ChatGPT query uses about 0.34 Wh, and Google reports 0.24 Wh for its median Gemini text prompt. Both are company figures using different, partly unpublished methods, so they are not a like-for-like comparison, and they cover text rather than images or video.
Does AI use a lot of water?
Data centres often use water for cooling, so AI does too. Google reports about 0.26 mL for its median Gemini text prompt, roughly five drops; OpenAI says about 0.000085 gallons per average ChatGPT query. The bigger concern is local: large data centres built in water-stressed areas.
Is training or using AI worse for the environment?
Both matter. Training is a large one-off cost; Meta reports 11,390 tonnes of CO₂e (location-based) to train its Llama 3.1 family. Inference happens every time anyone uses a model, so for widely used tools it keeps adding up for as long as the model is in service.
Is generating images worse than generating text?
Usually, yes. A 2024 study found that generating 1,000 images with the least efficient model tested used 11.49 kWh, as much as 522 smartphone charges, while the most efficient text model used as much as 9% of one charge for 1,000 runs.
What is the carbon footprint of a data centre?
It depends mainly on the electricity it uses and how clean that electricity is. The IEA estimates that electricity use by data centres causes about 180 million tonnes of CO₂ a year, rising to 300 million tonnes by 2035 in its base case.
Is AI getting more efficient?
Yes, quickly. Google reports a 33-fold cut in the energy of its median Gemini text prompt over one year. But efficiency makes AI cheaper and easier to use, so total use can still rise, which is the rebound effect.
What is the rebound effect?
It is when an efficiency gain makes something cheaper, so people use more of it and some or all of the saving is lost. The economist William Stanley Jevons described the extreme case in 1865, when more efficient steam engines increased Britain's coal use.
Can AI help fight climate change?
It can. The IEA says widespread adoption of existing AI applications could cut emissions by far more than data centres emit, around 5% of energy-related emissions in 2035, but warns that rebound effects could undercut this and that AI is 'not a silver bullet'.
How can I use AI more sustainably?
Ask whether you need AI for the task, write one clear prompt instead of several vague ones, ask for shorter answers, and avoid generating images or video you do not need. A 2025 UNESCO and UCL report found shorter prompts and responses can cut energy use by more than half.
Are smaller AI models better for the environment?
For many tasks, yes. The UNESCO and UCL report found that smaller, task-specific models could cut energy use by up to 90% in its tests, and research has found general-purpose generative models far more costly than task-specific ones for the same job.
What is a token?
A token is the chunk of text a language model reads or writes, often part of a word. OpenAI's rule of thumb for English is about four characters, or three-quarters of a word; more tokens means more computing.
What is PUE?
Power usage effectiveness is a data centre's total energy use divided by the energy its computing equipment uses. A PUE of 1.0 would be perfect; the Uptime Institute's 2025 survey found an industry average of about 1.54.
Is running AI on my own device greener than the cloud?
Sometimes. Running a small model locally avoids network and data centre overheads and keeps data private, but large data centres can be more efficient per task, and millions of small devices add up. It depends on the model, the task and the hardware.
Is it greener to use AI or do the task myself?
It depends on where you draw the boundary: your time, your computer, rework when the AI is wrong. For a quick fact or a simple sum, a search or a calculator is usually the lighter choice.
How is AI's energy use measured?
Ideally by metering the hardware while it runs, but many published figures are estimates rather than measurements. Tools such as CodeCarbon and EcoLogits estimate energy and emissions, and the Green Software Foundation's Software Carbon Intensity method, also ISO/IEC 21031, gives a standard way to report carbon per unit of work.
Do AI companies have to report their energy use?
In the EU, the AI Act requires providers of general-purpose AI models to document the model's known or estimated energy consumption for regulators. Public, comparable reporting for every AI service is not yet required in most countries.
Is there an energy label for AI?
Not an official one yet. Projects such as Hugging Face's AI Energy Score rate models on energy use for common tasks, and researchers and standards bodies are working on label-style benchmarks.
What is greenwashing in AI?
It is making an AI product or company sound more environmentally friendly than the evidence supports, for example claiming 'carbon-neutral AI' based only on offsets. In the EU, Directive (EU) 2024/825 bans generic environmental claims to consumers that cannot be proved.
Why are data centres built where they are?
Operators look for reliable power, grid connections, cool climates or water for cooling, land, connectivity and customers nearby. Concentration can strain local grids: in Ireland, data centres used 22% of all metered electricity in 2024, up from 5% in 2015.
Should schools teach green AI?
It fits naturally into existing computing, science, geography and citizenship lessons, and helps pupils become thoughtful users of AI. Our schools section has a teachers' guide and lesson ideas that need no paid tools.
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)
- Llama 3.1 8B model card, Meta (Hugging Face) (2024)
- 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)
- W. Stanley Jevons, The Coal Question, 1865, Yale University, Energy History (1865)
- What are tokens and how to count them?, OpenAI Help Center (2025)
- Uptime Institute global data center survey 2025, Uptime Institute (2025)
- CodeCarbon, CodeCarbon (2025)
- EcoLogits, GenAI Impact (2025)
- ISO/IEC 21031:2024 Software carbon intensity (SCI) specification, International Organization for Standardization (2024)
- Annex XI: technical documentation for providers of general-purpose AI models, European Commission, AI Act Service Desk (2024)
- AI Energy Score, Hugging Face (2025)
- Directive (EU) 2024/825 on empowering consumers for the green transition, European Union (2024)
- Data centres metered electricity consumption 2024, Central Statistics Office, Ireland (2025)