The Intergovernmental Panel on Climate Change (IPCC) is a United Nations body that assesses science related to climate change and provides advice to policymakers. Its advice has led to the establishment of widespread governmental initiatives, including from the UK government, designed to reduce reliance on fossil fuels that are impacting the global climate. 

While this move to more sustainable sources of energy is underway, concerns have been raised by the media, and our own staff and students about the environmental impact of using AI. Our individual use of AI, and expectation of an ‘always on’ AI/internet experience, drives technology companies to require ever-more-hungry data centres that require large amounts of energy and water (to cool them down).  

The complexity of AI 

There is a lack of transparency in the AI industry when it comes to environmental impacts and the whole process of AI development from mining precious metals to the water consumption of data centres is complex. Even the environment impacts of a single prompt is affected by the model you use, the length of your prompt, where you are in world and even the time of day it’s processed.  

However, there are studies that look at carbon emissions and water consumption of prompting an AI chatbot, like ChatGPT. Note that AI chatbots are a small part of the AI industry, with AI for content recommendations, business analysis, targeted ads etc. forming a much larger part.  

Multi-purpose, general AI systems (like ChatGPT) use more energy than task-focused AI systems (Luccioni et al, 2024) and it is the potential scale of general AI-use that is likely to pose challenges to our energy infrastructures at a time when we are seeking more sustainable alternatives. We should also consider our use of AI within the context of our use of other digital services, such as video-conferencing, gaming, using Netflix or streaming. These activities also involve significant energy usage. 

A typical prompt (less than 100 words) 

Here are some calculations from current research and thinking. A typical query using GPT-4o is between 0.3Wh and 3Wh, which in the UK is between 0.04g and 0.4g of carbon dioxide equivalent (CO₂e) (for the USA this is between 0.12g and 1.2g). This is the same as using a laptop for 3 minutes or playing a gaming console for 1 minute.  

Producing an image with stable diffusion 3 is about 0.6wh or 0.08g CO₂e in the UK.  

The water consumption for a prompt is between 10 – 25ml of water. This is slightly less than downloading a phone app and a fraction of the average UK shower, which is about 13 litres a minute. 

The number of data centres worldwide has surged from 500,000 in 2012 to over 8 million, with energy consumption doubling every four years, and the expansion of AI contributing to this growth (Li et al 2024).  

The development of AI and the use of LLMs and genAI is driving a significant increase in energy use. 

The Mistral Example  

In July 2025 AI company Mistral (based in France) released a statement showing their environmental impacts. The impacts of their model producing 1 page of text were 1.14g CO₂e, and 50ml of water. They also showed that training their model + 18 months of usage resulted in 20,400 tonnes of CO₂e emissions, and 281 000m3 of water consumed (in May 2025 Mistral had 7.6million site visits). If we compare this to the average UK person over a year, we find that it is the same carbon emissions as about 2,500 people and the water consumption of about 5000.  

AI for good 

It should also be noted that AI offers wide-ranging potential solutions to environmental challenges too, such as suggesting innovative energy solutions, improved prediction of weather patterns, improving water treatment, or enhancing efficiency and reducing emissions or water leaks from existing systems. 

Bite-sized task  

Step 1 – learn 

Read the article below and browse the optional extra resources. They capture the complexity of understanding AI and its energy use: 

We did the math on AI’s energy footprint. Here’s the story you haven’t heard. | MIT Technology Review 

Optional extra reading: 

Power Hungry Processing: Watts driving the cost of AI Deployment? 

Why using ChatGPT is not bad for the environment – a cheat sheet 

Is using AI for a task more efficient than a human? 

Step 2 – do  

Use this calculator to find out your own carbon footprint – WWF Footprint Calculator 

Step 3 – reflect  

• Reflect on the articles above. They capture some of the complexity involved in talking about AI’s impact on the environment. Did anything surprise you?  
• Consider the kinds of questions the WWF Calculator asked you. What might this tell you about your own environmental impact? How do you feel about this? 
• Consider the other digital tools that you use, or other technology-heavy activities you engage in e.g. gaming. What might be their environmental impacts? 
• How might you talk to students about AI and its environmental impact?

References: 

Li, P. et al. (2024). Making AI less “thirsty”: uncovering and addressing the secret water footprint of AI models. ArXiv. [Preprint]. https://arxiv.org/abs/2304.03271. 

Luccioni, A., Jernite, Y., Strubell, E. 2024. Power Hungry Processing: Watts Driving the Cost of AI Deployment? ACM Conference on Fairness, Accountability, and Transparency (ACM FAccT ’24), June 3–6, 2024, Rio de Janeiro, Brazil. https://doi.org/10.48550/arXiv.2311.16863  

This article also draws upon information from Professor Simon Kemp, Deputy Director of the Sustainability and Resilience Institute, and the UOSM 2043 module Global Sustainability Challenges.  

© 2025. This work is openly licensed via CC BY-NC-SA