Environmental Cost of AI
Artificial intelligence can be useful, but it requires electricity, water and physical computer equipment. Here is a straightforward look at its environmental cost—and why the complete picture is not always clear.
When you ask an AI chatbot a question, your request is processed by powerful computers in a data centre.
These computers use electricity, produce heat and need cooling. Building them also requires metals, water and other materials.
AI therefore has a genuine environmental cost. However, the size of that cost varies considerably, and AI may also help us use energy and other resources more efficiently.
The figures on this page were last reviewed on 18 September 2026.
Why does AI use so much electricity?
AI models run on specialised computer chips that can perform enormous numbers of calculations very quickly.
Electricity is used both to train an AI model and to run it once it becomes available.
Training an advanced model can involve thousands of computer chips working for weeks or months. After training, more electricity is needed every time somebody asks a question, generates an image or uses another AI feature.
One request may use relatively little energy. The larger concern is scale: popular AI services can handle hundreds of millions of requests.
According to the International Energy Agency's Energy and AI report, data centres used approximately 1.5 per cent of the world's electricity in 2024. This includes websites, streaming, cloud storage and other internet services—not just AI.
The IEA expects data-centre electricity use to roughly double by 2030, reaching around 3 per cent of global electricity consumption. AI is expected to be the main reason for that increase.
The Guardian provides an accessible summary of why electricity demand from AI data centres is growing.
Does using AI produce carbon emissions?
The answer depends largely on where the electricity comes from.
A data centre powered by wind, solar, hydroelectric or nuclear energy will generally produce fewer emissions than one relying heavily on coal or gas.
This means the same AI task may have a different carbon footprint depending on:
- Where the data centre is located
- The local source of electricity
- The efficiency of its computers
- The time of day
- The amount of renewable energy available
Technology companies are investing in renewable and low-carbon power. However, buying enough renewable electricity to match annual use does not necessarily mean a data centre runs on renewable energy every hour of the day.
The IEA expects emissions from data-centre electricity use to grow, although they are likely to remain a relatively small part of total global emissions.
The concern is not that AI is the largest cause of climate change. It is that its energy use is growing rapidly when emissions need to fall.
Why does AI use water?
Computer chips become hot while they work, so data centres need cooling systems.
Some use air. Others use water, some of which may evaporate while carrying heat away from the equipment.
Water can also be used indirectly by the power stations that generate the electricity.
The amount required varies according to the design of the data centre, the local climate and the source of its electricity. This is why claims that every AI question uses a particular amount of water can be misleading.
Location matters too. A data centre using water in a wet region may have less impact than one using the same amount in an area already facing shortages.
Questions have been raised in the UK about the water required by planned AI data centres in Oxfordshire, where future water supplies are already a concern.
What about the computer equipment?
AI's environmental cost begins before a data centre is switched on.
Servers, chips, batteries and cooling equipment require materials such as copper, aluminium, silicon, lithium and rare-earth elements. Extracting and processing these materials can use energy and water, damage habitats and create pollution.
AI equipment is also replaced as newer and more efficient chips become available. Some older equipment can be reused or recycled, but some becomes electronic waste.
The United Nations' Global E-waste Monitor reported that the world produced 62 million tonnes of electronic waste in 2022. Only 22.3 per cent was officially recorded as properly collected and recycled.
AI hardware is only one part of this much wider problem, but the rapid expansion of data centres will increase demand for new equipment.
The United Nations Environment Programme recommends looking at AI's entire life cycle—from mining and manufacturing through to operation and disposal.
How much impact does one AI question have?
There is no single reliable figure.
The energy required depends on:
- The AI model being used
- The length of the question and answer
- Whether it creates text, images or video
- The computers running the model
- The efficiency of the data centre
- The source of its electricity
Generating a short text response will normally require less computing power than creating a detailed image or video.
AI providers do not all publish the same environmental information, making direct comparisons difficult. Technology is also changing quickly, so figures can soon become outdated.
Treat precise claims about the cost of a single AI prompt with caution. They may apply to one model under particular conditions, but not to every AI service.
Can AI help the environment?
AI also has potentially useful environmental applications.
It can help to:
- Predict the output of wind and solar farms
- Balance electricity supply and demand
- Improve weather forecasts
- Detect leaks in water systems
- Monitor pollution and deforestation
- Reduce waste in factories
- Plan more efficient transport
- Develop improved batteries and materials
The IEA estimates that widespread use of existing AI applications could help reduce global emissions. However, this is a possibility rather than a guaranteed saving.
An AI system only provides an environmental benefit if people act on its findings and the savings are greater than the resources used to build and run it.
Should I stop using AI?
For most people, occasional AI use is unlikely to be the largest part of their environmental footprint.
Heating a home, driving, flying and buying new products will usually have a much greater personal impact than asking an AI assistant a few questions.
Nevertheless, billions of individual requests add up. You can use AI more thoughtfully by:
- Giving clear instructions so fewer attempts are needed
- Editing a useful result instead of repeatedly generating new ones
- Avoiding large numbers of images you will never use
- Choosing a smaller or faster model for simple tasks
- Using an ordinary search or calculator when AI is unnecessary
The largest improvements must come from AI companies, data-centre operators, electricity providers and governments. They decide how the equipment is powered, cooled, manufactured and eventually recycled.
The simple answer
AI has a real environmental cost.
It uses electricity and water, requires physical equipment and contributes to electronic waste. Its impact is growing, although estimates vary and AI remains only one part of the world's much larger environmental challenge.
AI may also help us manage energy, transport, water and other resources more efficiently.
The fairest conclusion is that AI is neither environmentally harmless nor automatically an environmental disaster. Its eventual impact will depend on how it is powered, how efficiently it is built—and whether we use it for things that are genuinely worthwhile.
Related: Back to Safety
