Understanding the Environmental Impact of Generative AI

By Robert Ball, Climate Change Coordinator at GCVS

Three quarters of people working in the voluntary sector now use generative AI in their day-to-day work. The technology arrived quickly and has been built into many of the digital tools we rely on, which makes it hard to choose whether or not to use it.

Like any tool, AI comes with benefits and limitations, and it is important to understand the effects on our energy supply and risks to our work. Alongside issues such as accuracy, fairness, bias, and privacy, we also need to discuss the serious environmental impact of AI. In short, generative AI uses large amounts of energy and water, and it is extremely difficult to know just how much.

Before exploring this in more depth, we encourage organisations to take part in the Third Sector Human Rights and Equalities Project’s email course on AI and ethics, which offers a clear introduction to these issues.

Image by Arno Senoner from Unsplash

How AI Uses Energy and Water

Generative AI tools work by processing huge amounts of information and generating new content through complex mathematical “averaging”. Training an AI model on data uses energy, then generating outputs consume energy again. These tools can now produce convincing emails, summaries, reports, images and video, but behind the scenes, each request involves significant computational effort and resources.

A traditional web search provides links to pages with the most relevance. Generative AI takes every possible answer and predicts a likely response. Looking up shops in Pollokshields, for example, now usually includes an AI-generated summary guessing what you want to know. Generating that summary increases energy demand.

Data centres perform this work continuously. They require electricity to run and chilled water to prevent overheating. The environmental impact is hard to measure because emissions vary wildly, and companies do not always publish figures. A single AI response could be processed in a data centre in Missouri, where electricity is heavily coal-powered, or in the UK, where renewable energy is more common. Factors such as time of day, weather and grid demand all play a part.

Recent findings from the Scottish AI Alliance’s People’s AI Panel on AI and Climate: Pulse Report also highlight growing public concern about the environmental footprint of artificial intelligence. The report reflects strong community support for clearer environmental reporting from AI companies and greater accountability around energy and water use. You can read the full report online: People’s AI Panel on AI and Climate: Pulse Report

Why Environmental Costs Matter

The environmental impact of AI as a technology has been well summarised elsewhere, such as this article from MIT which, amongst other things, says that:

“Globally, the electricity consumption of data centers rose to 460 terawatt-hours in 2022. This would have made data centers the 11th largest electricity consumer in the world, between the nations of Saudi Arabia (371 terawatt-hours) and France (463 terawatt-hours), according to the Organization for Economic Co-operation and Development.” 

The cost-of-living crisis adds another layer of concern. Due to poor infrastructure and pricing, 34 per cent of people in Scotland are currently in fuel poverty. Water shortages across Scotland in 2025 showed how vulnerable our resources are, and this becomes more common as we get more heat waves and our demand grows.

We all know the world’s climate has changed, bringing more extreme weather to Glasgow. The more wasteful our systems are, the worse this will get. Progress has been made worldwide to drive it down, but maintaining that progress requires careful stewardship of our energy and water resources.

We say this to highlight the problems in putting a resource-intense technology into everything we do. AI has its uses: in Scotland, for instance, AI is used in warning systems for floods and wildfires. Queen Mary University of London is using waste heat from AI computer servers to heat campus buildings, and other examples of tools for good are emerging; however, this is very different from using AI automatically for routine work.

Links to Cost of Living

Energy used by AI tools still comes from the grid and water from the same source as communities, driving up demand and potentially increasing household bills. In 2025, the UK Government announced plans to make the UK a global leader in AI. The data centres currently in the Scottish planning system will double our energy demands, and research from the University of Glasgow shows that Scotland’s existing data centres use over 13 million litres of tap water each year. If companies generate their own power, this can also carry risks, depending on the method used.

Five years ago, a Google search used around 0.3Wh of energy, roughly 1% of what it takes to boil a kettle. Google now reports that its AI summaries require an additional 0.24Wh, almost doubling the energy used for each query, not including the energy used to train the AI in the first place.

This is only for text on emails and the web. Consider the additional energy required for AI-generated videos or images multiplied across thousands of organisations. This is where the concern lies.

There are currently no reporting standards for AI companies. With no agreed format or obligation to share data, companies can choose which figures to publish – and have lobbied not to. 

Given the climate and cost of living issues, the Glasgow Climate Action Hub believes energy and emissions data from AI companies should be made publicly available.

Are they doing the job?

As we said above, what we call AI produces results through complex mathematical “averaging”. When we input a request, it analyses patters from vast datasets from across the internet (as well as other AI’s) to predict the most statistically likely, contextually relevant response. AI outputs ‘pass’ when their results are convincing to people. False or nonsensical outcomes (“hallucinations”) pass too, especially when tools are trained on AI-generated content - which is being increasingly generated and posted online by users of AI - and they tend to agree with whatever we tell them.

Even when AI produces technically correct answers or high-quality images, the ethical issues remain. A generated artwork, for example, is shaped by the work of a person that was originally fed into the model, including copyrighted material. This is arguably plagiarism.

Using AI Responsibly in Voluntary Organisations

We are not suggesting that voluntary organisations avoid AI altogether. In many cases, it cannot be avoided. Our aim is to provide enough information for organisations to make informed decisions about when and how to use these tools.

When developing or reviewing an AI policy, consider asking:

• Does this task genuinely save time or improve the quality of our work?

• Is the output reliable enough to be trusted?

• Is this a useful tool, or is it novelty over substance?

• Do we understand the environmental cost, and does the benefit outweigh it?

Finally, a reminder that the Third Sector Human Rights and Equalities Project offers a helpful email course on AI ethics, which we recommend to all organisations.

Thinking about AI email course

Our Climate team are lead partners in the Glasgow Climate Action hub and offer a range of support to Glasgow voluntary sector groups and organisations. Read about their help to create a climate action plan.