Spotlight on AI and the environment

Many of us are now aware of the environmental impacts of AI, but with new AI data centres threatening to crop up all over the UK, it’s important to realise how these centres will have many adverse effects on our land and water. Right now, data centres are estimated to consume around 415 terawatt hours of electricity annually. This is equal to about 1.5% of the entire world’s energy consumption. To put that in perspective, it is roughly equal to the total yearly electricity use of the UK and Italy combined, and could power around 41.5 million homes for a year.

Water Consumption

AI’s massive use of electricity means that a huge supply of water is needed to cool down the machines. It is currently estimated that for each kilowatt per hour of electricity used by a data centre, 2 litres of water is needed to cool it down. This means that data centres right now are using up billions of litres of water a year. And it’s only estimated to get worse, with AI-related water consumption predicted to equal the basic annual domestic needs of 1.3 billion people by the end of the decade.

It won’t be long until we see AI lead to worldwide water supply shortages. Shortages that will hit the least responsible the hardest. A lack of access to clean water in developing countries can lead to serious diseases, such as cholera. And in countries affected by drought, already worsened by climate change, a lack of water can lead to failed harvests, causing a lack of food and severe malnutrition among deprived communities.

Along with its mass consumption, AI also has a huge output of waste. AI-related infrastructure is predicted to produce 2.5 million tonnes of e-waste annually by 2030. E-waste is incredibly harmful to the environment, causing toxic leaks, damage to important ecosystems and harmful air pollution. And it is not biodegradable, so it will be here for a while.

The Problem Starts Small

The growing problem of AI’s mass energy consumption is often blamed on high scale AI development, however studies have shown that the problem starts with us. Daily usage of AI, such as using an AI chatbot to help with your maths homework or write your email, accounts for roughly 80-90% of the total energy demand of AI.

Chatbots like ChatGPT are becoming so widely-used that a commonly-used AI service is now estimated to process around 2.5 billion prompts per day. And generating an image using AI can require more than 1000 times the energy of a simple text answer. Along with mass energy consumption, AI systems also produce heaps of carbon into our atmosphere, with advanced AI generative systems on their way to becoming a leading source of carbon emissions. In fact, it is estimated that a single use of ChatGPT-4 creates roughly 4.32 grams of CO2, which is then released into the air. 4 grams may not seem like much, until you imagine it 2.5 billion times a day.

Scotland’s Part

The Green Turtle is based in Scotland so what’s happening there? Planning applications for new data centres have suddenly risen in Scotland, with more than 20 large data centres proposed so far. One scheme in Auchtertool, Fife has even been deemed the second largest in the world, showing the massive impact data centres would have on Scotland’s land. Along with concerns over the ecological impacts I have mentioned, many also questioned whether Scotland has the infrastructure to cope with the resulting energy demand from the data centres.

But there is cause for hope. On 16th September 2026, Holyrood voted to pause planning applications for all new AI data centres for up to a year, until the Scottish government has developed a national AI strategy. This is a huge step forward for safeguards around AI and means that the Scottish government must now urgently call in all AI planning applications nearing a decision-date. The government also announced that all large-scale data centres would have mandatory environmental impact assessments. This is a huge step for our environment and means that Scotland’s landscape will be better preserved from AI. There’s still a long way to go, but this is a first great step in the right direction.