AI and the Environment
Learn more about the conversations surrounding AI and sustainability.
AI and the Environment
AI and You
As knowledge of generative AI models has grown, so have concerns over the environmental impact of these models. As with all such questions, it is vital to remember that:
- AI tools used at current scale are extremely intensive in terms of both energy and water use.
- AI companies, the energy industry, and governments bear responsibility for the wider environmental impact of these tools as they have the power to make the changes required.
- Because the improvement of energy and water infrastructure, supply, and efficiency are out of our hands, placing the burden of the environmental impact of AI onto individuals is unreasonable and comparatively ineffective.
- Whilst we should absolutely be mindful of our use of AI and its wider impact, we should be realistic about our role too. Be kind to one another when discussing AI use.
Mindful use
Being mindful of our use of AI (just as we may wish to be mindful about driving a car or buying clothes) is something to actively encourage. Here are some things to be mindful of when using AI models:
- Effective prompting: The fewer prompts that it takes to address your query, the less energy and water is used. Click here to see our effective prompting guidance and make your prompts as efficient as possible.
- Consider the models you are using: Vote with your usage! You may, for example, decide to use models with more transparent sustainability tracking and missions.
- Avoid certain tasks where possible: Tasks such as producing videos are far more energy intensive. Try to avoid more intensive tasks where possible.
Environmental impact of AI use
AI tools remain extremely energy intensive. There are several factors which influence this:
A study by MIT Technology Review found that an AI model using 400 billion parameters ("parameters" are what the AI utilises to make its predictions) used around the same amount of energy in producing one text-based query as running the microwave for eight seconds. Now scale that up to millions of uses globally per minute and for image and video generation!
All that energy has to come from somewhere. If energy production (whether in the form of fossil fuels or renewables) is not increased, data centres place additional burden on the energy grid. In some cases, as in the United States, this has caused the energy bills of residents local to the data centres to spike.
Data centres require energy (and water for cooling) around the clock on a non-stop basis in order to provide services to a global market.
There are so many variables in the generation process (data centre is servicing your request, which model you are using, which output you are generating etc) that it can be challenging to estimate energy usage in simple terms.
A BBC article from February 2025 indicates that "a typical data centre can use between 11 million and 19 million litres of water per day, roughly the same as a town of 30,000 to 50,000 people." Whilst not all of this water goes to waste and whilst this does "not take into account recent efficiency improvements or developments in AI", there is no doubt that AI usage is extremely water intensive and regularly uses drinkable water for its cooling.
The possible benefits of AI
Proponents of AI also note the role that it could play in resolving sustainability issues if harnessed by the right people in the right way.
Researchers publishing in Nature have predicted that “advancements in AI in power, transport and food consumption could reduce global emissions of greenhouse gases by 3.2 to 5.4 billion tonnes of carbon-dioxide-equivalent annually by 2035.” Visit Green and intelligent: the role of AI in the climate transition for a summary of that paper from LSE.
Nonetheless, a study from MIT have shown that “deep learning” AI models may struggle more with climate predictions than simpler models due to the complexity and variability of the data.
The same study from Nature predicts that AI could optimise grid management and improve the load factor of solar and wind energy sources by as much as 20%. Visit Green and intelligent: the role of AI in the climate transition to read more about this.
To take one example, AI use in medical settings has demonstrated huge productivity boosts, reducing diagnosis time by “over 90% in tasks related to lesion detection and bone metastasis analysis”. This decrease in diagnosis time can improve healthcare outcomes and reduce the amount of time people spend in treatment. Read more about this in Reducing the workload of medical diagnosis through artificial intelligence.
A recent experiment in California used AI to manage ‘nitrogen and irrigation applications in growing grapes.’ The experiment found a 25% increase in productivity with a 20% reduction in water use.
An article from Oxford’s Saïd Business School notes several possible benefits of AI in biodiversity including monitoring illegal wildlife trafficking, ecosystem restoration planning, and predictive modelling for habitat conservation.
Find support and get involved
Sustainability
For further information on Kent's work on sustainability or to get involved, click the button below.
Wellbeing support
If you are experiencing any anxiety over the environmental impact of AI and wish to speak to someone, please click below to see the available from Student Support & Wellbeing
Home page
Click below to return to the AI home page