As the AI revolution gains momentum, there are increasing calls for measures to slow escalating environmental costs.
Globally, data centre power demand is increasing four times faster than all other sectors, according to the International Energy Agency.
Most studies say generative AI models – which create text, images, audio and video from prompts – consume five times more energy than traditional computing methods.
Some say it could be significantly higher, with concerns including:
- The computational power required to train generative AI models requires large amounts of electricity, leading to increased carbon emissions if this comes from non-renewable sources
- Each use of an AI tool requires a significant amount of energy, and
- Large amounts of water are needed for cooling generative AI’s hardware.
Ketan Joshi, an Oslo-based climate analyst, says on a basic level asking an AI chatbot a question consumes far more energy than finding the answer via a simple web search, adding the comparison: “You might still get the shopping done, and that single trip alone may not even look all that bad in terms of cost or emissions, but what happens when that’s all of your trips, and when all of society starts doing this?”
AI’s global carbon footprint – including electricity, carbon dioxide, water use, chip production, and the waste of old hardware – was estimated as being between 32.6 and 79.7m tonnes of CO2 emissions in 2025.
It is difficult for utilities, regulators and data centre developers to manage, with demand accelerating and the transmission infrastructure likely to take years to plan, permit and build.
Water use has been calculated as 312.5 to 764.6bn litres as data centre servers become hot when they process large workloads, so many facilities use water-based cooling systems to stop equipment from overheating.
Microsoft reported a 23 per cent increase in water consumption in 2023, and the company linked the challenge in part to new technologies, including generative AI, which needs specialised hardware, and making that takes metals, minerals, energy, water, and global supply chains. Some parts depend on mining and refining materials that can damage land, pollute water, and increase the lifetime emissions of AI hardware before the model even runs.
The report “Artificial Intelligence: Impact on human relationship and society”, published in the House of Lords Library, says over half of adults and young people in the UK now use generative artificial intelligence (AI), which can produce human-like content and dialogue.
As Europe races to catch up with China and the United States, Brussels hopes necessary computing power will be in place with a threefold increase in data centre capacity within seven years.
However, the European Environment Agency, which provides independent information on sustainability issues to policymakers, says the data centre boom could threaten the EU’s green goals, warning that it “presents a growing challenge to achieving climate neutrality”.
Director Leena Ylä-Mononen said: “If this is going to explode, to grow exponentially, we are going to have major issues around electricity sufficiency — the demand will be growing too fast and starts to compete with other industry sectors.
“I hear already some concerns from the industry side — do we have enough energy, especially green energy? And of course it should be fossil-free energy we use to produce the energy for the data centres.”
AI can, of course, become more sustainable, but companies need to cut energy waste, use cleaner power, design better chips, and stop treating hardware as disposable. Data centres can also cut emissions when they use solar, wind, or other low-carbon power, but they need better cooling systems. Repairing, reusing, and recycling equipment can also reduce e-waste and lower mining demand.
It’s early days but as AI technology grows, so will pressure on companies to prove their green credentials. The time to act is now.
