The environmental impact of AI
Artificial intelligence is being rolled out across public services at unprecedented speed. From the NHS to local councils, our members are increasingly being asked to use AI tools in their daily work. Government is pushing hard for rapid AI adoption, often without proper consultation or consideration of the consequences. As a union, we're thinking carefully about how AI will affect our members and the public services they deliver. One crucial but often overlooked aspect is AI's environmental impact. If we're serious about tackling the climate crisis, we need to understand what this technology is doing to our planet - and fast. This briefing explains the environmental costs of AI, why they matter, and what we can do about it.
Why is AI bad for the environment?
AI's environmental footprint is massive and growing rapidly.
Energy consumption
AI uses huge amounts of energy – from the training of systems to the processing of individual user requests. Training a single large AI model can emit as much carbon as five cars over their entire lifespans, while a ChatGPT query uses 10 times more electricity than a Google search Global data centre electricity demand could more than double by 2026, potentially exceeding Japan's total annual consumption. In the UK, data centres could account for 6% of national electricity use by 2030.
Water usage
AI data centres use water for cooling, placing significant pressures on local water resources and often drawing from drinking water supplies. A single 100MW data centre can consume 2 million litres of water a day - enough for 60,000 people. AI could consume 4.2-6.6 billion cubic metres of water globally by 2027 - equivalent to roughly half the UK's annual water consumption.
This is particularly concerning in the UK, where we're already facing a projected 5 billion litre daily water deficit by 2055.
Hardware and construction impacts
The physical infrastructure for AI technologies and data centres also has significant environmental impacts.
Microchips need rare earth elements, often mined in environmentally destructive ways, while data centres produce electronic waste containing hazardous substances like mercury and lead Other environmental impacts like "embodied carbon" from constructing data centres (steel, concrete, cooling systems) are huge but rarely considered.
The scale of growth
The number of data centres worldwide has surged from 500,000 in 2012 to 8 million today. This growth shows no signs of slowing.
What’s the UK government doing?
Despite designating data centres as "Critical National Infrastructure" – alongside our energy and water systems - the government has introduced no environmental safeguards or mandatory reporting requirements on AI and its infrastructure. The UK is sleepwalking into an environmental disaster.
But isn't AI good for the environment? Yes, it can be. AI has genuine potential to support environmental goals:
• Climate modelling - helping scientists understand and predict climate patterns
• Pollution detection - identifying harmful pollutants in lakes and rivers, tracking methane emissions • Biodiversity conservation - analysing complex ecosystem data
• Optimising systems - making energy grids and buildings more efficient However, most of these beneficial uses rely on predictive AI, not the energy-intensive generative AI (like ChatGPT, Google Gemini and Copilot) that's driving the current boom.
The environmental cost of AI expansion far outweighs these benefits - unless we act now to regulate it. Tech companies want us to believe we need AI first and everything else comes later, including our climate goals. This is backwards. We can't bet our planet on unproven technology while ignoring solutions we know already work.
Injustice of impact AI's environmental impacts follow the same pattern as other forms of exploitation: those who benefit are not those who suffer.
Global injustice
Most AI infrastructure is in the Global North yet the environmental damage - from mining rare earth minerals to water depletion - disproportionately affects the Global South. Communities on the frontlines of the climate crisis have no say in AI development, but bear its consequences.
UK injustice
Data centres are being built in areas already classified as “seriously water stressed” - including regions served by Thames Water, Affinity Water, Southern Water. The first government “AI Growth Zone” in Culham, Oxfordshire is seven miles from a planned reservoir meant to address water shortages. Local communities face water rationing while data centres get priority access as “Critical National Infrastructure”.
Worker exploitation
AI relies on poorly paid human labour to train models - often in exploitative conditions. From mining to data labelling, the AI supply chain is built on hidden human and environmental costs. This isn't just about carbon. It's about power, justice and who gets to decide our technological future.
What can we do about it?
1) Lobby government to make sustainability the priority
The message is simple: AI only if it's sustainable. We need to demand:
• Mandatory reporting of energy, water use and emissions by all data centres
• Environmental impact assessments before any new AI infrastructure is approved
• Integration of AI demand into national water resource planning
• Penalties for greenwashing and misleading sustainability claims
• Government leading by example - publishing annual digital sustainability reports
2) Get our employers to think critically about AI Not every task needs AI.
Challenge your employer to ask:
• Is AI actually necessary for this task? Often simpler, less harmful solutions exist
• What's the environmental cost? Demand transparency from suppliers
• Have we consulted workers? Implementation should involve those who'll use it
• What are the alternatives? Consider less resource-intensive options
3) Use AI carefully ourselves If you do use AI tools:
• Question whether you need it - could a search engine work instead?
• Use local models where possible - they consume far less energy than cloud-based services
• Avoid AI for images - generating images uses far more energy than searching for existing ones
• Stop when it's “good enough” - the last 2-3% of accuracy takes half the energy
• Choose smaller, efficient models over large general-purpose ones
4) Ask questions and stay informed
You have every right to ask:
• What AI tools am I being asked to use?
• What's their environmental impact?
• Has my employer assessed this?
• Are there alternatives?
Crucially, we all need to have confidence to engage on this issue. This is is new territory for everyone. Staying curious, critical and informed helps us make better collective decisions. The climate crisis demands urgent action. Unregulated AI expansion undermines everything we're trying to achieve.
We need to act now - as workers, as union members, as citizens - to ensure technology serves people and planet, not just profit. For more information or to get involved in UNISON's AI policy work, contact policy@unison.co.uk
Quick links
Existing members
Sign up to the Green UNISON newsletter
Express interests to get involved as a UNISON green rep at work