Artificial intelligence is being rolled out across workplaces and public services at rapid speed. From drafting emails to analysing medical scans, AI tools are increasingly presented as essential to modern work.
But trade unions, among others, have concerns about the pace – and lack of caution – with which society is embracing this advanced technology. One area of chief concern is the environmental cost of AI.
The public sector is legally required to meet net-zero targets, but if you look at the vast physical infrastructure of data centres, and their effect on energy networks and water systems, the environmental footprint is growing quickly.
The environmental impact of AI
Some of the figures already emerging around AI’s environmental impact are striking:
- A single ChatGPT query can use around 10 times more electricity than a Google search
- By 2035, data centres could account for 20% of the UK’s projected total CO2 emissions
- One large data centre can consume up to two million litres of water every day – roughly the equivalent of 6,500 households
- AI’s carbon emissions last year were estimated to be comparable to the entire city of New York
- Freshwater consumption linked to AI infrastructure in 2025 exceeded global bottled water consumption
- AI hardware production relies on toxic electronic waste and mineral mining often associated with labour exploitation.
What exactly is a data centre?
AI often feels intangible to users, however the systems behind it are remarkably material. At the heart of AI’s environmental impact are its data centres, facilities filled with high-performance hardware that store and process enormous amounts of information. These require vast computing power: thousands of servers running continuously, generating an extortionate amount of heat.
To prevent overheating, data centres use a method called direct evaporative cooling, where cold water vapour is blasted across server racks to reduce temperatures. This process is extremely water-intensive, which is why large data centres can consume millions of litres of freshwater every day.
As AI is rolled out across our public services, the demand will rise, as will the resources they consume.
A lack of transparency
According to Foxglove, the non-profit organisation that “fights to make tech fair for everyone”, one of the biggest challenges is the lack of publicly available information about the environmental costs of AI infrastructure.
Tom Hegarty, Foxglove’s head of communications, explains that Foxglove’s investigation into AI’s environmental impact started in earnest around 2024, when the organisation started examining the infrastructure behind generative AI. “The first thing we realised was that there is a real lack of reliable, transparent information about the environmental costs of data centres in the UK,” he says.
Foxglove initially approached data centre developers directly for information about water use and energy consumption. “We got almost nothing back,” Hegarty says, apart from PR-like assurances that environmental impacts were being managed. Foxglove turned instead to Freedom of Information requests to water companies. The results were revealing, but incomplete. Only about half of companies responded with usable figures, and some of the most significant data – including figures from Thames Water – was already years out of date.
“Even the numbers we did receive were almost certainly a massive underestimate,” Hegarty says. “The expansion of data centres since then has been enormous.”
A major planning loophole
Foxglove’s research uncovered another issue: how data centres are classified within the UK planning system. Right now, data centres are typically categorised under B8 planning use, the same classification used for storage warehouses. It sounds minor, but the implications are significant. Warehouses generally have relatively modest energy and environmental impacts, so they are often approved with limited scrutiny. A hyperscale data centre, by contrast, may require huge amounts of electricity and water while producing large amounts of carbon emissions.
“The problem,” Hegarty explains, “is that if you’re a planning officer looking at a list of applications, a data centre can look like just another warehouse.”
As a result, projects involving enormous computing infrastructure can pass through the planning process without the level of environmental scrutiny typically applied to major infrastructure developments such as roads, railways, or power stations.
And in all cases, there is currently no mandatory legal requirement for an environmental impact assessment.
Workers and the climate dilemma
For many workers, the rapid expansion of AI raises uncomfortable contradictions. UNISON policy officer Kate Jones says that union members increasingly recognise the environmental impact of the tools, such as Copilot, that they are being encouraged (if not yet required) to use by their employers.
“People are asking: if every search I do has a significant environmental cost, what does that mean for my own impact, and what does it mean for the net-zero targets in the public sector?” she says.
Jones argues that the general conversation around climate change often focuses too heavily on individual behaviour, which ignores the bigger issue: “The responsibility here isn’t really with individuals. The responsibility lies with the systems that are driving this technology.
“We need AI firms thinking about how they can make sure their products are less environmentally damaging. And we need governments looking at AI infrastructure, as its being proposed, and asking do we need this one, do we need that? And we need employers and governments looking at procurement and asking which companies are going to have more of an environmental impact, which ones won’t.”
A fundamental issue, she says, is not whether AI can be used in any given instance, but whether it should be. “There are incredible applications. AI is really helping with cancer diagnosis, for example, and that’s amazing, that’s not something anyone would argue against. But do we really need to use it to generate every piece of written content? Especially when the results often aren’t even that good?”
What needs to change
Kate notes that “the government is incredibly reluctant to do anything that will slow down AI, and that includes regulation.” But addressing AI’s environmental impact will require action at multiple levels. UNISON’s key demands are:
- Mandatory reporting by data centres of energy use, water consumption and emissions
- Environmental impact assessments for new AI infrastructure projects
- Integration of AI demand into national water resource planning
- Penalties for misleading sustainability claims or ‘greenwashing’
- Public sector transparency through annual digital sustainability reports.
And this is what members can do:
- Collective bargaining and campaigning in their branches for responsible use of AI in the workplace, asking employers how AI fits in with their environmental strategy
- On an individual basis, ask if a task genuinely requires AI
- When using generative AI, use the smallest model possible generative – lighter models use significantly less energy
- Avoid using AI to generate images and videos – it uses a significant amount of energy.
As Kate Jones concludes: “There are so many questions as to whether we need to be using AI, but the environmental one is really important. When the environmental cost is so high, are we using AI where it truly matters – or simply because we can?”
