Synthetic intelligence is reworking the world. From ChatGPT answering on a regular basis inquiries to AI-powered techniques serving to medical doctors, companies, researchers and college students, the expertise is turning into an more and more vital a part of trendy life.
However behind the comfort of asking an AI chatbot a query, producing a picture or producing a report lies an environmental price that many individuals hardly ever think about: water.
Whereas a lot of the dialog about synthetic intelligence focuses on electrical energy consumption and carbon emissions, one other useful resource is quietly turning into a significant concern. Freshwater.
The computer systems powering synthetic intelligence want monumental quantities of electrical energy and generate important warmth. To maintain these machines working safely, the information centres housing them require refined cooling techniques. Relying on the expertise used, a few of these amenities devour substantial portions of freshwater.
And the priority is rising as expertise corporations race to construct extra highly effective AI techniques and increase their information centre infrastructure.
So, is synthetic intelligence consuming all of the world’s clear water? Is our rising dependence on AI placing further stress on an already strained world water provide?
The reply is extra sophisticated than the headlines counsel. Nevertheless, the environmental implications are critical sufficient to deserve consideration.
How Does Synthetic Intelligence Use Water?
To know the issue, you will need to perceive the place synthetic intelligence truly operates.
AI techniques don’t exist fully within the cloud. Behind each chatbot, picture generator and AI-powered software are bodily computer systems housed in monumental amenities generally known as information centres.
These amenities include 1000’s of highly effective pc servers that course of data, retailer information and run advanced calculations.
In contrast to strange workplace computer systems, the high-performance processors used to coach and function superior AI fashions can generate appreciable warmth, notably when 1000’s of them work concurrently.
If the temperature turns into too excessive, the gear can turn into much less environment friendly or endure injury. Consequently, information centres want dependable cooling techniques to keep up appropriate working temperatures.
That is the place water enters the image.
Behind each AI response is a community of highly effective computer systems working inside large information centres.
Many information centres use cooling towers and evaporative cooling techniques. These techniques take away warmth by permitting some water to evaporate. The method helps decrease temperatures, however the evaporated water should be changed.
In some amenities, this implies constantly drawing contemporary water from municipal provides, rivers, reservoirs or different sources.
It is very important perceive that not each information centre makes use of the identical cooling expertise. Some depend on air cooling, closed-loop liquid cooling, recycled water or different techniques that considerably scale back their dependence on freshwater.
However, water stays an vital a part of the environmental footprint of many computing amenities.
AI Does Not Simply Use Water to Cool Computer systems
One of many largest misconceptions about AI’s water consumption is that the one water concerned is the water used inside information centres.
In actuality, its water footprint extends past the buildings themselves. There are three main methods synthetic intelligence contributes to water consumption.
1. Direct water consumption by information centre cooling
That is maybe the simplest a part of the issue to know.
AI servers generate warmth, and a few amenities use water-based cooling techniques to take away it. When evaporative cooling is used, a number of the water disappears into the environment and can’t instantly be reused throughout the similar native water system.
The bigger the computing operation, the better the potential cooling demand. A big AI information centre working constantly can subsequently place appreciable demand on the water sources of the world the place it’s positioned.
Some information centre cooling techniques use evaporative cooling, which consumes water as warmth is launched into the environment.
2. Oblique water consumption by electrical energy technology
Synthetic intelligence requires monumental computing energy, and that computing energy requires electrical energy. Nevertheless, producing electrical energy may also devour water.
Some thermal energy stations use water for cooling, whereas sure types of electrical energy technology have their very own water necessities.
Which means even when an information centre doesn’t use a lot water immediately, the electrical energy powering its servers might have a further water footprint elsewhere.
In different phrases, the environmental price of operating an AI system shouldn’t be restricted to the water flowing by the cooling gear inside its information centre.
In keeping with the Worldwide Power Company’s Power and AI report, world information centres consumed roughly 560 billion litres of water in 2023. The IEA estimates that this might rise to roughly 1.2 trillion litres yearly by 2030 below its base-case projection.
These figures cowl information centres typically, not AI alone.
The place Does the Web Really Stay?
The IEA additionally explains that electrical energy technology and power provide account for a considerable proportion of knowledge centre water consumption.
This demonstrates why assessing solely the water used inside information centres provides an incomplete image.
3. Water utilized in manufacturing pc chips
There’s one other a part of the story that receives even much less consideration. The superior processors utilized in synthetic intelligence techniques should first be manufactured.
Producing semiconductor chips is a particularly advanced industrial course of involving extremely purified water, sometimes called ultrapure water.
This water is utilized in cleansing and rinsing processes throughout chip manufacturing as a result of even tiny particles or contaminants can have an effect on the efficiency of semiconductor elements.
As demand for AI processors will increase, the semiconductor business should produce extra superior computing {hardware}.
This creates a further water demand all through the expertise provide chain.
Due to this fact, the water footprint of AI begins lengthy earlier than somebody opens ChatGPT and kinds their first query.
How A lot Water Does ChatGPT Really Devour?
That is the place the topic turns into notably attention-grabbing.
In 2025, researchers Pengfei Li, Jianyi Yang, Mohammad A. Islam and Shaolei Ren revealed a analysis paper titled Making AI Much less ‘Thirsty’: Uncovering and Addressing the Secret Water Footprint of AI Fashions in Communications of the ACM.
The researchers examined the water footprint related to coaching and working massive language fashions.
Certainly one of their estimates advised that coaching GPT-3, a big language mannequin developed by OpenAI, in Microsoft’s US information centres might devour roughly 5.4 million litres of water in complete when direct and oblique water consumption had been included.
Of this quantity, roughly 700,000 litres represented direct, onsite water consumption.
The researchers additionally estimated that roughly 500 millilitres of water may very well be consumed for each 10 to 50 medium-length responses generated by GPT-3, relying on the situation and working circumstances.
Nevertheless, this shouldn’t be interpreted as a set measurement for each ChatGPT dialog.
Completely different AI fashions have totally different computing necessities. The size and complexity of a request, the {hardware} getting used, the cooling system, the native local weather and the electrical energy supply can all affect the outcome.
A brief textual content response shouldn’t be environmentally equal to producing a posh picture or processing a prolonged AI activity.
The vital level is that AI interactions have an infrastructure price, even when the expertise seems to function effortlessly on a smartphone or laptop computer.
You may learn the unique analysis in Communications of the ACM.
Might AI Devour Billions of Litres of Water within the Coming Years?
The priority turns into even better once we think about how rapidly synthetic intelligence is increasing.
Expertise corporations are investing billions of {dollars} in new information centres, superior processors and computing infrastructure to assist more and more highly effective AI techniques.
As extra folks and companies undertake AI, the demand for computing sources is anticipated to extend.
A extensively cited projection by Li and colleagues estimated that world AI-related water withdrawal might attain between 4.2 billion and 6.6 billion cubic metres yearly by 2027.
To know the size, one cubic metre comprises 1,000 litres of water.
That projection subsequently represents between 4.2 trillion and 6.6 trillion litres of water withdrawals yearly.
It is very important distinguish between water withdrawal and water consumption.
-
Water withdrawal: The whole quantity of water taken from sources akin to rivers, lakes and groundwater. A few of this water might ultimately be returned.
-
Water consumption: The portion that’s not returned to the unique water supply in a readily reusable type, for instance as a result of it has evaporated.
The identical analysis estimated that roughly 380–600 billion litres may very well be consumed by evaporation related to world AI demand in 2027.
These figures are projections, not affirmation that this amount has already been consumed or that it’ll essentially be reached.
However, they spotlight the potential scale of the problem if AI infrastructure continues increasing with out enough consideration to water effectivity.
Why Ought to We Be Nervous When Hundreds of thousands of Folks Already Lack Clear Water?
A very powerful query shouldn’t be merely how a lot water synthetic intelligence makes use of.
It’s the place that water is getting used.
Water shortage stays a critical problem in communities world wide, making the situation of water-intensive industries notably vital.
Water shouldn’t be distributed equally internationally. Some areas have plentiful freshwater sources, whereas others expertise extended droughts, unreliable rainfall, groundwater depletion and critical water shortages.
An information centre positioned in a water-rich area might have a really totally different environmental impression from one working in an space already experiencing extreme water stress. This is the reason the situation of AI infrastructure issues.
When a big information centre is constructed in an space experiencing water shortages, its demand for water can create further stress on current sources.
Native residents, farmers, companies and different industries might depend upon the identical water provides.
This doesn’t mechanically imply {that a} information centre will deprive close by residents of ingesting water. The precise impression relies on its water supply, consumption ranges, native infrastructure, environmental laws and the way water sources are managed.
Nevertheless, the potential for competitors over restricted freshwater sources is a legit concern. In international locations the place entry to wash ingesting water is already a significant improvement problem, the growth of water-intensive industries deserves notably cautious planning.
It might be tough to justify technological progress that locations pointless stress on communities struggling to safe dependable entry to important water provides.
Are Expertise Corporations Doing Something About It?
The rising environmental impression of knowledge centres has not gone unnoticed by main expertise corporations.
Microsoft, Google and different expertise companies have introduced sustainability commitments and investments geared toward lowering the environmental impression of their infrastructure.
Among the approaches embrace:
Water-efficient cooling techniques: Corporations are exploring cooling applied sciences that scale back or eradicate the necessity for steady freshwater evaporation.
Closed-loop liquid cooling: As a substitute of continually consuming new water, these techniques flow into coolant by gear to take away warmth. Relying on their design, they will considerably scale back ongoing water demand.
Recycled and reclaimed water: Some amenities use handled wastewater or recycled water relatively than relying fully on drinking-quality freshwater.
Improved information centre areas: Constructing amenities in cooler climates or areas with extra sustainable water sources can scale back cooling necessities and native environmental stress.
Extra environment friendly AI {hardware}: Growing processors that ship extra computing efficiency with much less electrical energy may also scale back the quantity of warmth that must be eliminated.
Microsoft, for instance, introduced a brand new information centre design in 2024 supposed to make use of zero water for cooling. The corporate said that its chip-level cooling method might save greater than 125,000 cubic metres of water yearly per information centre in contrast with its earlier common water-use efficiency.
Its reported world information centre water utilization effectiveness additionally improved from 0.30 litres per kilowatt-hour in FY2024 to 0.27 litres per kilowatt-hour in FY2025.
These developments are encouraging, however they don’t imply that the whole expertise business has eradicated its water footprint.
The electrical energy provide, semiconductor manufacturing, building and different elements of the availability chain nonetheless have environmental penalties.
Moreover, sustainability commitments should be supported by clear reporting and independently verifiable outcomes.
You may look at Microsoft’s figures in its information centre power and water effectivity report and its 2025 Environmental Sustainability Report.
Is AI Actually Well worth the Environmental Price?
That is maybe probably the most tough query.
Synthetic intelligence has monumental potential to enhance folks’s lives.
It may assist medical analysis, help scientists in analysing advanced data, assist farmers make higher selections, enhance transport techniques and make training extra accessible.
Additionally it is serving to companies automate repetitive duties and enabling researchers to discover options to environmental issues.
Nevertheless, the advantages of AI don’t mechanically justify limitless consumption of pure sources.
Not each AI software offers the identical stage of social worth, and never each computing activity requires probably the most highly effective out there mannequin.
As AI turns into extra deeply built-in into on a regular basis life, expertise corporations should think about whether or not the advantages of specific purposes justify their power and water necessities. There’s additionally a have to make environmental data extra accessible to the general public.
Shoppers can see how rapidly AI instruments generate solutions, however they hardly ever know which information centre processed their requests, how a lot electrical energy was required or what the related water footprint may need been.
With out higher transparency, it turns into tough for regulators, researchers and communities to evaluate the true environmental price of the expertise.
What Can Be Accomplished to Cut back AI’s Water Footprint?
Decreasing AI’s water consumption mustn’t imply abandoning technological progress.
As a substitute, it requires a extra accountable method to how AI techniques are developed, powered and operated.
A number of steps might make a significant distinction:
1. Spend money on water-efficient information centres. Expertise corporations ought to prioritise cooling techniques that minimise freshwater consumption, notably in areas experiencing water stress.
2. Enhance the usage of recycled water. The place acceptable and environmentally secure, handled wastewater can scale back dependence on freshwater provides.
3. Enhance electrical energy effectivity. Extra environment friendly processors, information centres and electrical energy techniques can scale back the oblique water footprint related to AI.
4. Keep away from constructing water-intensive amenities in weak areas. Corporations and governments ought to assess native water availability earlier than approving main information centre developments.
5. Make water consumption figures public. Expertise corporations ought to disclose significant details about their direct and oblique water consumption, relatively than relying solely on broad company sustainability claims.
6. Develop extra environment friendly AI fashions. Smaller, specialised fashions can typically carry out specific duties with fewer computing sources than a lot bigger general-purpose techniques.
Governments even have an vital position to play by establishing environmental requirements and guaranteeing that communities are thought-about when massive expertise tasks are authorised.
So, Is Synthetic Intelligence Consuming All Our Clear Water?
No. Synthetic intelligence shouldn’t be consuming all of the world’s clear water, and there’s no credible proof that it’s about to go away humanity with out freshwater.
Nevertheless, dismissing the difficulty just because the headline sounds exaggerated can be a mistake.
AI is making a rising demand for computing infrastructure, and that infrastructure requires electrical energy, cooling and more and more refined {hardware}.
All three can contribute to freshwater consumption.
The best concern shouldn’t be essentially the whole quantity of water utilized by AI worldwide, however the stress that enormous amenities might place on specific communities and water-stressed areas.
Because the expertise continues to advance, its environmental footprint should turn into a part of the dialog about innovation.
The world can not afford to pursue technological progress whereas ignoring the pure sources required to maintain it.
Synthetic intelligence could also be getting smarter day by day, however the business should additionally turn into smarter about the way it makes use of the planet’s most important sources.
The query is not simply how highly effective AI can turn into. It’s whether or not we are able to make that progress with out placing further stress on the water sources that folks, communities and ecosystems depend upon.
