AI is thirsty. But have we been ignoring what else is drinking our water?
AI has a water problem.
There’s no point pretending otherwise.
Data centres require water. The electricity powering them can require water. Semiconductor manufacturing requires water. And as artificial intelligence becomes embedded into everything from Google searches and customer service to healthcare, advertising and the software we use every day, that demand is going to grow.
Some estimates suggest global AI could be responsible for 4.2–6.6 billion cubic metres of water withdrawal annually by 2027.
That is an enormous number.
And understandably, it has generated some fairly alarming headlines about AI, water shortages and even future “water wars”.
But whenever I see those numbers, I keep coming back to another question:
What are we comparing them to?
Because AI didn't suddenly invent humanity's water problem.
We have been operating extraordinarily water-intensive industries for decades, and some of the biggest are sitting right in front of us on supermarket shelves and in our wardrobes.
First, let's put 6.6 billion cubic metres into context
Agriculture is already, by an enormous margin, the world's biggest user of freshwater.
According to the UN Food and Agriculture Organization, agriculture accounts for approximately 72% of global freshwater withdrawals.
Not AI.
Not households.
Not showers.
Agriculture.
And within agriculture, animal production has a substantial footprint of its own.
A major global assessment by researchers Mesfin Mekonnen and Arjen Hoekstra estimated the total water footprint of animal production at approximately:
2,422 billion cubic metres of water every year.
Compare that with the projected 4.2–6.6 billion cubic metres of annual water withdrawal associated with global AI in 2027.
On the surface, animal agriculture's number is roughly 370–580 times larger.
But; and this is important - that isn't an entirely fair apples-to-apples comparison.
Because not all "water use" means the same thing
This is where environmental statistics become a little less headline-friendly.
Water footprint studies generally distinguish between three types of water:
Green water is rainwater stored in soil and used by plants.
Blue water is surface water and groundwater — rivers, lakes, reservoirs and aquifers.
Grey water is a calculated volume representing the amount of freshwater required to dilute pollution to acceptable standards.
Around 87% of animal agriculture's estimated 2,422 billion cubic metre footprint is green water.
In other words, much of that giant number represents rainfall falling on pasture or crops that eventually become animal feed.
That isn't equivalent to a data centre drawing potable water from a municipal water system in a drought-affected region.
And that distinction matters.
But it doesn't make the animal agriculture issue disappear.
The same study estimated that approximately 6.2% of animal production's footprint was blue water.
That works out to roughly 150 billion cubic metres of surface and groundwater annually.
Even that figure is dramatically larger than current projections for AI.
And around 98% of the entire water footprint of animal production comes from producing animal feed, rather than animals simply drinking water.
That's an important part of this conversation that rarely seems to make it into a headline.
The beef example is particularly confronting
Globally averaged, producing one kilogram of beef has been estimated to have a total water footprint of approximately:
15,400 litres.
For comparison:
sheep meat: approximately 10,400 L/kg
pork: approximately 6,000 L/kg
goat meat: approximately 5,500 L/kg
chicken: approximately 4,300 L/kg
eggs: approximately 3,300 L/kg
cow's milk: approximately 1,000 L/kg
Again, those figures include green, blue and grey water, so don't interpret "15,400 litres" as 15,400 litres of drinking water being poured onto a cow.
It isn't.
But the comparison becomes particularly interesting when we look at nutrition.
Researchers found the average water footprint per calorie of beef was around 20 times that of cereals and starchy roots.
Per gram of protein, beef's footprint was approximately six times that of pulses.
So this isn't simply an unavoidable consequence of producing food.
What we choose to produce — and consume — matters enormously.
And then there's cotton
Food isn't the only familiar industry with a serious thirst.
Globally, producing one kilogram of cotton has been estimated to require around 10,000 litres of water across its total water footprint.
The cotton required for a single pair of jeans has been estimated at around 8,000 litres.
Again, location matters enormously.
Cotton grown largely using rainfall is a completely different water proposition from irrigated cotton grown in a water-stressed region.
And that's actually the same problem we need to talk about with AI.
AI's water problem isn't just about how much. It's about where.
A billion litres of water doesn't have the same environmental consequence everywhere.
Water consumed in a region with abundant renewable water supplies is very different from water withdrawn from an already stressed aquifer or municipal drinking supply.
And this is where criticism of AI infrastructure is absolutely justified.
A 2026 review of AI's water footprint noted that two-thirds of data centres built since 2022 were located in water-stressed regions.
That's concerning.
Data centres can use freshwater directly for cooling, while the electricity supplying those centres creates another indirect water footprint. Semiconductor manufacturing adds another layer again.
And unlike rain falling on a field, some of this demand can compete directly with water required by communities.
So I'm certainly not arguing:
"Animal farming uses lots of water, therefore AI is fine."
It isn't.
I'm arguing something different.
We need to stop discussing sustainability in silos
AI is the new thing.
And new things make fantastic villains.
A server farm is futuristic, enormous and slightly terrifying.
A steak isn't.
A cotton T-shirt isn't.
An irrigated field isn't.
We've culturally normalised the environmental costs of industries that have existed for generations.
AI doesn't yet have that privilege.
That doesn't mean we should stop asking difficult questions about AI.
We should be demanding water-efficient cooling systems, recycled and non-potable water use, renewable energy, responsible data-centre locations and far greater transparency from technology companies.
If an AI company wants to build a giant data centre in an area already experiencing serious water stress, communities should absolutely be asking questions.
But those questions should be part of a much bigger conversation.
Because agriculture already accounts for roughly 72% of global freshwater withdrawals.
Animal production has been estimated to represent around 29% of agriculture's total water footprint.
Around 12% of global groundwater and surface-water consumption for irrigation has been estimated to go towards producing animal feed.
And industries such as cotton have significant footprints of their own.
Against that backdrop, talking as though AI has suddenly arrived to drink the world's water supply dry misses a rather uncomfortable amount of context.
AI absolutely needs to become more efficient
And it probably will.
That's one of the interesting differences between emerging technology and mature industries.
There is enormous commercial pressure to make computing more efficient.
Waterless and closed-loop cooling technologies are developing. Data centres can increasingly be located according to climate and water availability. Wastewater and reclaimed water can replace potable supplies in some facilities. Computing workloads can potentially be shifted geographically or temporally depending on electricity and water conditions.
None of those things means the problem solves itself.
Regulation, transparency and pressure matter.
But AI's water footprint is not a fixed physical law.
It is partly an infrastructure-design problem.
And that gives us something to work with.
So, is AI going to cause the world's water wars?
Water scarcity is real.
Climate change is real.
Growing populations, agricultural demand, industrial development, deteriorating infrastructure and increasingly unpredictable rainfall are all placing pressure on freshwater systems.
AI is now another rapidly growing demand that needs to be managed within that system.
But reducing the conversation to:
"I used ChatGPT, therefore I am destroying the world's drinking water"
isn't particularly useful.
Especially if we're unwilling to examine the much larger systems we've been participating in for decades.
The better question isn't:
Does AI use water?
Of course it does.
The question is:
What activities are we collectively using water for, where is that water coming from, how much of it is actually being consumed, and are the benefits worth the environmental cost?
Ask that of AI.
Ask it of beef.
Ask it of dairy.
Ask it of cotton.
Ask it of energy.
Ask it of every major industry.
Because if we're genuinely worried about the future of the world's freshwater supply, we need considerably more than a new villain.
We need context.
And we need the numbers.
