Is AI Really Using All Our Water? A Student’s Question Worth Taking Seriously
Is AI Really Using All Our Water? A Student’s Question Worth Taking Seriously
A student recently told me that she did not want to use AI because of the amount of water it uses.
That stopped me in my tracks.
Not because she was wrong to worry. In fact, I was quite impressed. Most students worry about whether AI will write their homework, destroy revision skills or produce a suspiciously polished essay about Macbeth. This student was thinking about the hidden environmental cost of the technology.
And she has a point.
AI does use water. Data centres use electricity, electricity generation can involve water, computer chips have a water footprint, and some data centres use water for cooling. The “cloud” is not really a cloud. It is a building full of hot machines, pipes, cooling systems, cables, backup power and planning applications.
But as with many green issues, the truth is not as simple as:
“AI uses water, therefore AI is bad.”
The better question is:
How much water does AI use compared with households, food production, golf courses, industry and ordinary data centres in the UK?
First, We Need to Be Careful With the Word “Use”
Water statistics can be slippery. Rather like trying to hold a wet fish while explaining exam technique.
There are three different meanings.
Water withdrawal means water taken from a river, reservoir, borehole or public supply. Some of it may later be returned.
Water consumption means water effectively lost from the local system, usually through evaporation or incorporation into products.
Embedded water means water used somewhere else in the supply chain: electricity generation, chip manufacturing, farming, food processing, clothing and so on.
This matters because one litre of water in a shower, one litre evaporated from a cooling tower, one litre sprayed on a golf green and one litre used to grow imported food are not environmentally identical.
The Household Baseline: We Already Use a Lot
In England, average daily household water use was 136.5 litres per person per day in 2024–25. Under dry-year adjustment, that becomes 140.3 litres per person per day. So a family of four is easily using around 546 litres per day before we have even mentioned AI, golf courses or industry.
The same Environment Agency analysis reported that water companies in England abstracted 14,637 million litres per day, while total distribution input was 13,946 million litres per day. Leakage was still a major issue: 2,617 million litres per day, or almost 19% of water supplied.
That means one of the biggest “water users” in England is not a student asking ChatGPT to explain photosynthesis. It is water leaking out of pipes before anyone gets to use it.
How Does AI Compare With Other Water Users?
Here is the important context.
| Water user or sector | Approximate water figure |
|---|---|
| One person in England | 136.5 litres/day |
| Household of four | about 546 litres/day |
| Typical surveyed English data centre below techUK threshold | under 27,400 litres/day |
| Large 100 MW hyperscale data centre estimate | about 6.85 million litres/day |
| England golf courses using direct abstraction | 10 million litres/day average |
| England golf courses using direct abstraction in summer | 24 million litres/day peak |
| British food industry | about 900 million litres/day |
| England water leakage | 2,617 million litres/day |
Approximate daily figures. These are not perfectly like-for-like because some are household use, some are sector consumption, and some are estimates for cooling or irrigation.
Values are shown in litres per day. The typical data-centre figure is the upper threshold of 10,000 m³ per year converted to a daily value.
Data Centres: Not All Are the Same
One of the most useful UK sources is the 2025 techUK report on data-centre water use in England, produced with Environment Agency involvement. It found that among surveyed English commercial data-centre sites, 51% used waterless cooling systems, 64% used less than 10,000 cubic metres of water per year, and 89% either measured water use or used systems that did not require water for cooling.
That figure of 10,000 cubic metres per year sounds large, but it is about 10 million litres per year, or roughly 27,400 litres per day. That is not trivial, but it is tiny compared with food manufacturing, national leakage or agricultural irrigation.
However — and this is important — small and medium data centres are not the whole AI story.
The Government Digital Sustainability Alliance report warns that there is still no complete, reliable public dataset for total UK data-centre water use. It also states that a 100 MW hyperscale data centre could consume around 2.5 billion litres per year, which is about 6.85 million litres per day.
That is the difference between a modest computer room using little cooling water and a huge AI-capable facility operating at industrial scale.
Golf Courses: Greener Than They Look, or Thirstier Than We Admit?
Golf is an interesting comparison because it is visible. You can see the sprinklers. You can see the green grass in a dry summer. A data centre, by contrast, hides behind fences, walls and marketing language about “the cloud”.
England has around 2,200 golf courses. The National Framework for Water Resources says about 60% use mains potable water for irrigation, while the rest rely on direct abstraction. Of the courses that directly abstract water, around 820 licences, irrigation consumption is about 10 million litres daily on average, rising to 24 million litres daily in summer.
That does not make golf evil. Courses can also provide green space, trees, rough grass, ponds and wildlife corridors when managed well. But it does show that AI is not the only modern activity with a hidden water question.
Food and Industry: The Giant in the Room
The British food industry is a much larger water user than most people imagine. One review of water use in the UK food and drink sector reports that the British food industry consumes about 900 megalitres per day — that is 900 million litres per day.
This includes water for washing, cleaning, processing, cooling, steam, hygiene and manufacturing. If we want to talk honestly about water, we cannot only talk about AI prompts. We also need to talk about food waste, meat consumption, processed food, packaging, cleaning systems and supply chains.
The National Framework also identifies other important water-using sectors, including food and drink, chemicals, mineral products, agriculture, spray irrigation and power generation. In some regions, future non-public-water demand is expected to be dominated by power generation or agriculture; in others, industry is the bigger factor.
What About One AI Question?
This is where the internet often gets overexcited.
A widely quoted research paper estimated that GPT-3 could consume the equivalent of a 500 ml bottle of water for roughly 10 to 50 medium-length responses, depending on when and where the model is deployed. The same paper estimated that training GPT-3 in Microsoft’s US data centres could directly evaporate 700,000 litres of freshwater, with a larger total footprint when indirect water use is included.
That is a useful warning, but it should not be treated as a universal law of physics.
A prompt served in a hot, water-stressed region using evaporative cooling is different from a prompt served in a cooler region, with waterless cooling, closed-loop systems or renewable electricity. Different models, hardware, data centres and electricity mixes produce different results.
So I would not tell a student:
“Every AI question uses exactly half a litre of water.”
That is too simplistic.
I would say:
“AI has a real water footprint, but the amount depends on the data centre, the cooling system, the electricity source, the model and the location.”
The Problem With AI Is Scale
One student using AI carefully to revise biology is not the main water problem.
Millions of people using AI casually, constantly and wastefully may become part of a much bigger infrastructure problem.
That is where the concern becomes serious. AI increases demand for data centres. Data centres increase demand for electricity. Some cooling systems increase local water demand. And the UK is already facing long-term pressure on water supplies from population growth, climate change, drought resilience and environmental protection.
The UK Government designated data centres as critical national infrastructure in 2024, reflecting their importance to modern life, public services and the economy. That means they are not going away. The question is whether they are built and operated responsibly.
The Fair Answer to My Student
So, was my student wrong to worry?
No.
She was asking exactly the kind of question we should be encouraging students to ask:
Where does the technology come from?
What resources does it use?
Who benefits?
Who pays the environmental cost?
Is there a better way to do it?
But I would not want her to stop there.
Refusing to use AI completely may not be the most effective environmental action if the rest of life continues unchanged: long showers, food waste, leaking toilets, disposable fashion, imported water-hungry crops, unnecessary car journeys and devices left running all day.
A better response is:
Use AI thoughtfully. Do not use it lazily. Demand transparency from AI companies. Ask where data centres are built, how they are cooled, whether they use drinking water, whether they report water use, and whether they are located in water-stressed areas.
That is a stronger green argument than simply saying, “AI uses water, so I will never use it.”
Practical Rules for Greener AI Use
Here are my suggested rules for students, teachers and businesses.
1. Use AI when it genuinely helps
Using AI to explain a difficult concept, generate revision questions, improve accessibility or support learning can be worthwhile.
Using AI to produce endless disposable content that nobody reads is much harder to justify.
2. Ask better prompts
A vague prompt often produces a poor answer, followed by ten more prompts trying to repair it. A clear prompt saves time, energy and water.
Instead of:
“Explain biology.”
Try:
“Explain osmosis for GCSE Biology using a potato practical, three key terms and one exam-style question.”
3. Do not use AI as a thinking replacement
If AI stops a student thinking, it is bad education.
If AI helps a student understand, practise, question and improve, it can be useful.
4. Push companies to publish water data
We need clearer reporting on water use by data centre, not just vague global promises. The most useful information would include location, cooling method, water source, water consumption, local water stress and whether recycled or non-potable water is used.
5. Keep the bigger water picture in mind
Fixing leaks, reducing food waste, improving industrial efficiency, harvesting rainwater, using grey water sensibly and designing water-efficient buildings may save far more water than simply telling students never to use AI.
Conclusion: The Cloud Has Pipes
The most important lesson is not that AI is evil.
The most important lesson is that modern life hides its infrastructure.
We tap a screen and forget the server.
We open a tap and forget the reservoir.
We eat a sandwich and forget the irrigation, washing, processing and transport.
We walk across a green golf course in August and forget the sprinklers.
We ask AI a question and forget the heat, electricity and cooling behind the answer.
So yes, AI uses water.
But the answer is not panic. The answer is literacy.
Environmental literacy. Digital literacy. Scientific literacy.
A student who asks, “How much water does AI use?” is already doing something valuable. She is refusing to treat technology as magic.
And that may be the most important green skill of all.
Comments
Post a Comment