AI and the Environment: What You Need to Know

Awdur: Andrew Collins; Amser Darllen: 5 munud
Mae hwn yn adnodd agored. Mae croeso i chi ei gopïo a'i addasu. Darllenwch y telerau.
Os hoffech gymorth pellach gyda'ch her ddigidol, trefnwch sesiwn am ddim gyda DigiCymru

Few topics in the AI conversation generate more debate than its environmental impact. Headlines range from alarming to dismissive, and almost every source has some kind of bias. The truth, as is often the case, sits somewhere in the middle.

This is a subject we at ProMo Cymru give careful consideration to. Our AI Statement acknowledges that AI has a real environmental cost, and that we do not yet have all the answers about how to address it, but we are committed to learning, being honest about the trade-offs, and sharing what we find out.

How much energy does an AI prompt actually use?

The honest answer is, not very much for a typical text-based interaction. Google published research in 2025 measuring the environmental impact of its Gemini AI across energy, emissions, and water. Their findings suggest the avergae text prompt uses around 0.24 watt-hours of energy, emits roughly 0.03 grams of CO2, and consumes about 0.26 millilitres of water – around five drops. They compared this to watching TV for less than nine seconds.

Technology blogger Andy Masley has done a thorough analysis of ChatGPT’s environmental footprint, reaching similar conclusions. His research suggests that a single prompt uses roughly the same energy as leaving a wireless router on for three minutes, and that it would take around 1,000 prompts to increase your daily energy use by 1%.

It is worth noting that both of these sources have a stake in the narrative. Google is one of the largest AI providers, and Masley is a self-described AI optimist. Neither makes them wrong, but it is worth keeping in mind when reading the numbers.

If you want to assess your own AI footprint, EcoLogits calculator lets you estimate the energy usage of different AI models yourself, which is a useful way to get a more neutral sense of the picture.

So why does AI get such a bad press?

A few reasons. First, the aggregate numbers are genuinely large. AI data centres consume significant amounts of electricity, and that demand is growing fast. Second, much of that electricity still comes from carbon-intensive sources. Third, the physical construction of AI infrastructure – the hardware and the housing of new data centres – carries a substantial upfront environmental cost that is harder to quantify and often left out of per-prompt comparisons.

There is also a meaningful difference between models. A simple question sent to a lightweight model uses a fraction of the energy of a complex multi-step task run on the most powerful model available. This distinction matters, and it is one that most of the alarming headlines tend to flatten.

The Newid programme – the Welsh Government-funded digital support initiative delivered in partnership by ProMo Cymru, WCVA, and Cwmpas – covers AI governance and environmental impact in its training for third sector leaders. One of the key points made in that training is exactly this; different models have very different footprints, and making conscious choices about which tool you use for which task is one of the most practical things an organisation can do. Newid runs this training a few times a year, so if you’re interested in attending, check newid.cymru/tag/training-and-support to see when the next session is running.

The bigger picture

Individual prompt use is unlikely to be a significant part of any organisation’s environmental footprint. The more meaningful questions are at a systemic level – how AI infrastructure is powered, how hardware is manufactured, and whether AI’s potential to optimise energy use across other sectors actually gets realised. The International Energy Agency has suggested that AI applications could deliver substantial carbon reductions by 2035 if adopted widely across the economy, potentially offsetting data centre emissions several times over. That is a hopeful projection, but it comes with real uncertainty.

None of this means environmental considerations should be ignored. The National Lottery Community Fund, in their own AI guidance, makes the point that AI tools require large amounts of energy and water, and encourages organisations to use AI mindfully, only where it will clearly help. That feels like a reasonable position to us.

Practical things your organisation can do

You do not need to stop using AI to take the environment seriously. Here are some straightforward things worth considering:

Match the model to the task: A lightweight model is usually sufficient for summarising a document or drafting an email. You do not need the most powerful model available for every task, and smaller models use significantly less energy.
Avoid unnecessary repetition: Sending the same prompt five times hoping for a better result uses more energy than taking a moment to write a clearer prompt in the first place.
Use the EcoLogits calculator: It gives you a rough sense of the footprint of different models and tasks, and is a useful tool for building awareness across your team.
Have a conversation in your organisation: A simple internal discussion about when and how AI is being used, and what models are appropriate for different tasks, is a good starting point for responsible use.
Check out ProMo Cymru’s AI Statement: We have shared our own thinking and the tools we use at promo.cymru/ai-statement, including a link to the Newid AI Risk Assessment Tool which can help your organisation think through its approach.

The environmental impact of AI is a real issue, and one that the sector is still figuring out. But it is also one where thoughtful, informed use makes a genuine difference. The goal is not to stop using AI, it is to use it well.

If you would like support thinking through your organisation’s approach to AI, get in touch with our team to learn how we can support you.

Get In Touch

Wedi'i gomisiynu gan Catalyst