Data, Data, Every Where, Nor Any Drop to Drink

I have been meaning to write an article about the impact of generative AI on the environment but it always felt a little too big. Too daunting. But recent events have sharpened my focus and given me a way into the topic that I hadn’t seen before.

The Devon Data Centre

I live in Devon, in the south west of England. It is a large county and a lot of it is countryside. We have two national parks, Dartmoor and Exmoor, and two coast lines. It’s a beautiful place.

It is not the first place you would think of if you were deciding where to build Britain’s largest data centre.

The “Devon Data Campus” is a “proposed AI data and energy storage development at Alverdiscott, North Devon”.¹ At 850 acres, it would be the largest such site in Britain (the next largest is a 435 acre proposal for a site in North Lincolnshire: the Elsham Tech Park development).² The proposal consists of a set of data centres (five large buildings rising to a height of 40m) and a separate battery energy storage system (BESS), both of which will occupy a greenfield site just outside the small town of Torrington, right in the middle of the countryside.

At first glance, it seems like an unlikely location for this type of development. And it has certainly been met with a lot of local opposition.³ However, Xlinks, the company behind the proposal, have a different view. The mild climate, easy access to the national grid, and the ability to place it in the countryside, away from houses, are the reasons why they have chosen this particular site.⁴

My immediate reaction on hearing this news was dismay. This doesn’t feel like the right place for this sort of development. But is that reaction the correct one? And as I pondered that question, I found a way to start talking about the environmental impacts of generative AI in a more focused way.

The wrong way to talk about generative AI and the environment

One of the loudest talking points about generative AI and its environmental impact is water use. The Washington Post’s article, Every email written using ChatGPT uses a whole bottle of water, sparked a rush to try to put a number on the amount of water generative AI is using.

The trouble is, it’s not an easy calculation to make. The number you will come up with will depend on what you include or leave out of your calculations. Do you include the water used in the construction of a data centre, or just the water used when it is up-and-running? Do you restrict yourself to only the water used to cool the servers, or do you also consider the water used when generating the electricity that powers the site? What about the water used in the manufacture of all of the specialised electronics needed for a data centre?

There is no simple answer. By choosing what to include or exclude, you can make the number larger or smaller. Which means both sides of the debate can come up with numbers to support their conclusions. And neither will be wrong, necessarily; they will simply be looking at different things.

I have written a longer explanation of why I don’t put much stock in these numbers at the foot of this article.

Instead of trying to come up with a specific number, I have an alternative suggestion.

A different way to talk about generative AI and the environment

Rather than focus on a contentious calculation in an attempt to put a number onto the problem, I would rather ask two questions:

Can you draw a straight line from increased use of generative AI to an increased demand for data centres?
What are the impacts on the local communities and the environment in the places where data centres are built?

If we can draw a line from the use of generative AI to an increased demand for data centres and we can show that there are significant impacts arising from those data centres, then I feel satisfied that you can say there is a link between generative AI and those environmental and societal impacts. And, to bring it back to my local issue, it will reinforce my belief that a Devon Data Campus is not the right option for Devon’s countryside.

Does generative AI drive a demand for more data centres?

The answer seems to be, unquestionably, yes.

Let’s ask a simple question: What is driving the demand for more data centres, particularly in the UK? According to a document put out by the UK Parliament, What are data centres and how sustainable are they?, the answer is “AI, general computation and digitalisation are expected to drive UK data centre growth.”⁵

From that same document, we find that the UK currently has around 450 large data centres. Not all data centres do the same thing. There are traditional data centres, “designed to support a wide range of computing needs, including web hosting, enterprise applications, databases and cloud services” and AI data centres, which are built “specifically for AI and machine learning (ML) applications… [and which are] optimized for processing large volumes of data, running deep learning models and supporting AI-driven tasks”.⁶ Of particular note are “hyperscale” data centres (the largest ones, like the proposed Devon Data Campus): “Large-scale, ‘distributed’ computing centres that support high volume data processing, computing, and storage services.”⁷

The resource requirements of a traditional data centre are lower than those of an AI data centre, partly because of the type of hardware they run on and partly because the nature of AI data processing means it comes in peaks and troughs. From that same document as before:

Depending on the components used, servers can be optimised for either AI or more general computing. AI servers have less predictable demands and higher energy usage of 50 to 120 kW per rack, while general computing servers have more predictable demand, and typically have an energy usage of 30 to 85 kW per rack. McKinsey estimates that hyperscale data centres are set to have increasing proportions of AI servers, from 40% of data centre server capacity in 2023 to 65% in 2030.

If we look at that quoted McKinsey document (with apologies for needing to bring them into this discussion!):

A big chunk of growing demand—about 70 percent at the midpoint of McKinsey’s range of possible scenarios—is for data centers equipped to host advanced-AI workloads. And the nature of those workloads is rapidly transforming where and how data centers are being designed and operated.⁸

In summary, the demand for data centres is being driven by increases in AI and general computation needs and, within that, the proportion of the servers given over to AI processing is expected to increase to where it will become the majority user of data centre capacity. I think that allows us to draw a fairly confident line from AI to the demand for more data centres.

(And for the nitpickers among you who may be asking, but can we draw a line from generative AI to this increased need for data centres? there are plenty of examples where the data centres have been built expressly to cater for new generative AI needs. For example, the Colossus data centre in Memphis was built specifically to train Grok, xAI’s racist⁹, abusive¹⁰ generative AI model. We won’t be covering the costs that community has had to suffer but please go and read about them. Their stories deserve your attention. And we can turn once again to – shudder – McKinsey: “Cloud service providers (CSPs) such as Amazon Web Services, Google Cloud, Microsoft Azure, and Baidu are the companies fueling most of today’s incremental demand for AI-ready data centers. That’s because of the capacity these hyperscalers require to run large foundational models developed in-house, such as Google’s Gemini, or to host models developed by AI companies, such as OpenAI’s ChatGPT.”)

Local impacts for local people

One of the problems with those numerical arguments I dismissed earlier in the article is that they try to average out the impacts of data centre use across the entire user base of that data centre. If a user sends an email written by ChatGPT then that is equivalent to using up one bottle of water, for example. Leaving aside whether that is true or not, the water being referred to in those sorts of calculations is not being extracted equally around the world. It is very unlikely that it will be taken from wherever the ChatGPT user is located.

The majority of the impacts from data centres are only felt in one location: the place where the data centres are built. Whether you use generative AI or not, if a data centre is built on your doorstep then you will be burdened with the costs of that data centre.

Energy use

Remember how the location for the Devon Data Campus was chosen because it offered a good connection to the power grid? Xlinks talk about needing “a robust connection to the National Grid” and access to “spare, highly reliable capacity at Alverdiscott substation — important for a facility that runs 24 hours a day.”¹¹ Clearly, it is important for a massive data centre to be able to access the energy it needs to power all of its servers. But does that have an impact on other users of the National Grid and what kind of energy are they using?

According to that above-mentioned UK Parliament document, “data centres use around 2% of the UK’s electricity.” That’s the figure for 2025, before any of these new hyperscale data centres come online. Over in Ireland, new figures released last month show that total energy consumption by data centres rose from 5% in 2015 to 23% in 2025.¹² Which is a pretty significant increase. For comparison, the total proportion given over to residential use was 28% in 2025.

A report from Friends of the Earth links this to increased energy costs for Irish consumers.¹³ So, even if you aren’t concerned about the environmental effects from an increasing demand for electricity, the fact that it will probably cost each of us more in energy bills should still give you pause. In America, an investigation by Bloomberg discovered that “electricity now costs as much as 267% more for a single month than it did five years ago in areas located near significant data center activity.”¹⁴

In theory, it shouldn’t be difficult to draw on renewable sources to power these data centres. But, in practice, data centres tend to be built faster than the infrastructure needed to connect them to renewable power so much of the new demand for capacity has been met by non-renewable sources.¹⁵ Opportunity Green, an NGO focused on climate action, point out that the current carbon budget process, the mechanism by which the UK manages emissions towards net zero, “makes no allowance for the appearance of a novel new demand for electricity in the form of data centres. Yet surging data centre installation and operation across the UK, with at least 100 installations planned, creates substantial new demands for electricity. Details from data centre planning proposals that have been submitted across the UK, if built and operated, could imply an additional electricity demand for the UK equivalent to almost the whole of Denmark.”¹⁶

They continue: “Data centre electricity demand is unusual in being both very high, and continual. Pressure is on data centre operators to go for quick solutions to this, frequently now meaning fossil fuel power sources.”

If you’re feeling uneasy about this then you are not alone. A survey quoted in the UK Parliament document found “75% of the British public support data centres powered by renewable energy, compared to 40% for nuclear and 20% for fossil fuels.” It seems as though the majority of people agree: if new data centres are going to be built then there needs to be sufficient renewable energy infrastructure in place to support them. And that doesn’t seem to be the case at the moment.

So, what about water?

I dismissed the arguments above, not because water use is not an important issue, but because framing it in terms of finding a magical number that proves your point is not – in my opinion – a useful way to talk about it.

Returning once again to the document from the UK Parliament:

Data centres consume water in three ways. From most to least direct they are:

Cooling systems for servers to maintain optimal operating temperatures and prevent overheating. Not all data centre cooling methods use water. Data centres often use drinking-quality water because it can be reliably sourced, and also prevents clogging of pipes, pumps, and heat exchangers with contaminants, but this is not universal.

The electricity supply, which may come from gas and nuclear power stations which evaporate water during operation

Semiconductor chip manufacturing, which requires large amounts of processed water to rinse residue from silicon wafers during various stages of the manufacturing process.

Point 3 is a particularly important one but it falls outside the scope of what we’re looking at here today. Instead, let’s concentrate on the effects on local water supplies when a data centre arrives in your town.

The good news is that existing UK data centres appear to have a fairly low impact. A TechUK survey found that 64% of UK data centres “use less than 10,000 m³ of water per year – less than a typical leisure centre.”¹⁷ Now, before we get too excited, this was a survey of only 73 data centre sites and that number relates only to the cooling systems used in the data centre. But those findings would suggest that, assuming new data centres are built with similar capabilities as existing ones, the demand on local water supplies will be no greater than if a new leisure centre was built in the area.

So why has WaterUK, the trade association for Britain’s water industry, written to the government warning that their “ambitions on data centres are at risk for no other reason than a lack of water.”?¹⁸ Clearly there is a difference of opinion between the data centre people and the water company people, and the picture painted by the water companies is a bleak one. In part, this is because the water companies, unlike TechUK, are considering all of the water use wrapped up in points 1 and 2 above. But if you are considering the impact of data centres on the nation’s water resources, then it is sensible to consider the entire picture. Because, despite a reputation for wet weather, the UK is facing more and more challenges when it comes to supplying water to the country. Y’know, because of climate change (coupled with a chronic lack of investment from the water companies over the last 30 years). From WaterUK:

There appears to be an assumption that the country will always have enough water for its economic needs. Nothing could be further from the truth. Right now, houses are being blocked and businesses are unable to expand for no other reason than a lack of water. The situation is so bad that, solely due to a lack of water, the Environment Agency objected to a cancer hospital being built in Cambridge. Water supplies are so low that it has forced a moratorium on business expansion in East Anglia and parts of Essex. A lack of water is the only thing stopping development in parts of Sussex, Suffolk and Lincolnshire. The problem is spreading.

This might just be me, but if I was faced with choosing between a cancer hospital and a data centre, I know which one I would want to prioritise.

Infrastructure, in for a pound

You may look at both of those issues – the lack of available water and the difficulty in connecting to renewable energy – and say that those are not issues restricted to data centres alone. And you would be right. In my opinion, both are symptoms of a lack of investment and forward planning in our national infrastructure that has been going on for decades.

That does not mean that adding data centres into the mix will not exacerbate an already difficult situation. That cancer hospital was approved in the end, over the objections of the Environment Agency¹⁹, but I think the EA were right to highlight the risk to the local water supply. Do we want to add a data centre on top of that? They are proposing one!²⁰ Won’t that only make the situation worse?

PFAS? Pfft! It’s the bugs you have to worry about

Concerns have also been raised about the effects on groundwater supplies, both during the construction of the data centres and from their ongoing running, particularly if they are pulling from aquifers for their water needs. There are horror stories of what it can be like to live next to one of these sites, with wells running dry and appliances getting clogged up,²¹ but it is difficult to get a handle on how isolated these cases are. It is possible that these are uncommon side effects, but that will be scant comfort if you are one of those affected.

Similarly, concerns are circulating about the use of harmful chemicals in the cooling process, additives that are introduced into the water “to prevent corrosion, scaling, and the growth of algae and bacteria. Common additives may include anti-corrosion/scaling chemicals, biocides, chlorine compounds, glycols, and so forth. Additionally, some nonwater liquid coolants may contain refrigerants and PFAS (also known as “forever chemicals”), requiring special forms of treatment.”²² An Environment Agency report and evidence submitted to a 2025 Environmental Audit Committee inquiry by TechUK, the technology sector’s trade body, “said that data centres contribute to PFAS pollution” and, while they are not intentionally released, “there is a risk that they might leak into the environment during operation and end-of-life disposal during data centre upgrades.”²³

And if it’s not PFAS, then it’s other unpleasant contaminants getting flushed into the local water system. A Meta data centre in Wyoming, USA, “flushed bacteria-contaminated water into public sewers during construction of a controversial new AI datacenter.”²⁴ It’s important to note that this did not end up in drinking water but the local authority did use it to water public parks, among other things, and they noted that spraying it in public areas led to it become airborne and, therefore, increased the risk to the public.

The buzz around data centres

Chemicals and contaminants may be ejected from data centres accidentally, but that’s not all they pump out into the local area. Data centres also emit constant low frequency hums as part of their normal operation and that constant noise can affect wildlife (as well as local residents). The UK Centre for Ecology & Hydrology summarises the issue:

The effects of man-made noise on nature are also a problem. Noise can be particularly problematic for marine organisms because it travels very well through water, but it also affects land-based species, disturbing animals’ behaviour when they are feeding, communicating or looking for mates.²⁵

Data centres run 24/7, of course, so this is an equal opportunity issue for both our daytime and nocturnal animals. A HackerNoon article, Will Data Centers Ruin Your Neighborhood?, in its reporting on communities affected by noise pollution from data centres, highlights that “the noise is even more noticeable in rural areas where massive, nondescript buildings replace spaces that used to be forests or farmland.”²⁶

Community pressure can have an impact, though. In a small community in Virginia, USA, a prolonged campaign against the noise pollution coming from a nearby Amazon Web Services data centre resulted in changes to the site that cut the noise levels by 10dB.²⁷ The problem hasn’t gone away completely, but the situation has been made less unbearable.

Something I hadn’t realised about the UK’s approach to noise management is that “assessment methodologies are based on an extensive baseline noise survey. This determines the noise levels at receptor locations and is affected by factors like nearby roads, railways, aircrafts, or industrial noise, and noise limits are then set relative to pre-existing conditions.”²⁸ This suggests that there will be stricter limits imposed on a data centre in the quiet Devon countryside, say, than on one built in a noisier urban area.

This all needs to be thought about up-front, though, because not only is it very expensive and not fully effective to address noise pollution after the fact, as AWS discovered in Virginia, but by the time remedial work is carried out (if it is!) then the damage has already been done.

An AI generator

One of the other sources of noise pollution from a data centre site can be the sound of diesel generators, which act as a backup in case of power cuts.

These are supposed to be used sparingly in the UK and are tightly regulated. According to the much-quoted UK Parliament document: “Unless permitted, data centres with generators exceeding 50 MW may only be used for a maximum of 50 hours per year outside of emergencies to comply with the medium combustion plant directive, and the specified generators regulation.” But there are a few issues that jump out to me here. Firstly, the phrase “unless permitted”. How many data centres are permitted to run their generators for more than 50 hours? Who grants this and who monitors it? Secondly, what about “emergencies”, because those are allowed for under the rules? Who dictates what constitutes an emergency? And thirdly, who is actually monitoring any of this? The Devon Data Campus will be out in the middle of the countryside. Who will be watching how often they fire up a generator?

You may think I am being needlessly pessimistic but I would encourage anyone feeling hopeful about either (a) an industry’s ability to do the right thing and only use their emergency powers in real emergencies, or (b) the regulators ability to adequately monitor things like this and ensure they are keeping to the regulations, to seek out the recent docu-drama Dirty Business. It’s a bleak look at an industry misusing its emergency powers to avoid doing its actual job, coupled with a failure or the regulator to do any meaningful oversight or enforcement. It’s not about data centres, but if you watch that and cannot see how similar failings could also play out here, then you are a far more optimistic person than I.

Because it’s not just noise pollution. In the US, on-site diesel generators, along with the fossil fuel power stations that fuel the data centres, have been linked with “air quality concerns and public health costs through the emission of criteria air pollutants. These pollutants include fine particulate matter (PM2.5), sulfur dioxide (SO2), and nitrogen dioxide (NO2). Exposure to criteria air pollutants, especially PM2.5, is causally linked to premature mortality, asthma, cardiovascular disease, and other health effects.”²⁹ The study I have just quoted from, continues:

Our estimates show that U.S. data centers could contribute to various health outcomes including approximately 600,000 asthma symptom cases and 1,300 deaths in 2028 under the highgrowth scenario, with total public health costs exceeding $20 billion… Although national-level impacts remain modest, data center health costs vary substantially across counties: the highest county-level per-household health cost is about seven times the national average and approximately 200 times the lowest county-level value, warranting closer attention.

And this shows, once again, why coming up with one overall number for an issue like this is not helpful. The study projects an increase in the number of cases of asthma and in additional deaths resulting from this air pollution. But when measured against the overall population of the US, “national-level impacts remain modest.” This gives naysayers an easy way of dismissing the issue. It’s a drop in the ocean, they will say, compared with… etc etc etc. But the effects are not felt nationally. They are concentrated in those areas in which the data centres and power plants are built.

That is why I believe it’s important to focus on the local impacts and not try to average out the effects across an entire country or an entire user base of any particular AI model.

The most local of issues

Which brings us to the most obvious impact of building a data centre: the data centre itself. Where once there were acres of Devon farmland, now there will be a data centre. Or five of them, to be more accurate. With car parks and roads and a 96 acre battery storage system. That’s a lot of habitat that has been taken away.

Before construction begins, forests, grasslands, wetlands or farmland may be cleared to make room for buildings, roads and electricity infrastructure. This destroys wildlife habitat and split ecosystems into smaller areas, making it harder for animals to find food, mates and safe places to live. Scientists call this habitat fragmentation, and it is considered one of the biggest threats to biodiversity worldwide.³⁰

The Environment Act 2021 requires all major building projects to support biodiversity gains of 10% over 30 years. In theory, this means that “a development will result in more or better quality natural habitat than there was before development.”³¹ I admire this as an aspiration but this does not mean that the area under development will have 10% better habitat at the end of the process. That is one possibility, but so is the option of improving habitat elsewhere or buying “biodiversity credits”.

What happens in the meantime? After all, there will be habitat loss. An example given in the above-quoted guide to the Environment Act 2021: “A developer cuts down some trees on the site. They must make up for this impact on the habitat, as well as delivering 10% BNG.” It is explicitly acknowledged that habitats will be removed. At the end of the day, those trees are still gone and what’s left behind is less than it once was.

Fragmented habitats are often lower quality. This is known as the ‘edge effect’. As a habitat is broken into smaller sections, the proportion of edge – where one habitat meets another – increases. While some species can thrive along habitat edges, others struggle to survive. For instance, species that have evolved to live in the interior of a wood are less suited to its edge, where conditions are quite different.³²

The Devon Data Centre will be located in the Transition Zone of the UNESCO North Devon Biosphere.³³ One of the aims of the biosphere is “to restore and reconnect North Devon’s treescapes—woodlands, hedgerows, and orchards—so they thrive for wildlife, climate resilience, and community wellbeing.” That feels at-odds with the creation of an 850 acre data centre campus that will inevitably end up causing habitat fragmentation.

You will remember that, earlier, we looked at how data centres are often built far faster than the infrastructure that would connect them to renewable power. That’s why a lot of them have turned to fossil fuels. In an article looking at data centres in the US, the National Wildlife Federation (NWF) states:

The challenge isn’t a lack of electricity generation. In many regions, clean energy is already available or under development. The true bottleneck is transmission: the long-distance infrastructure needed to deliver power from where it is produced to where new, geographically concentrated data center loads are sited.³⁴

You would therefore think that building new transmission infrastructure would be a good thing, as it will allow data centres to connect to cleaner sources of power. While that is true, it also comes with a downside. Again, from the NWF article:

Transmission development can cause direct habitat loss, fragmentation, and degradation, and disrupt migratory pathways. Research has shown that linear infrastructure can also indirectly impact animal behavior, underscoring the importance of early planning, corridor-based siting, and avoidance of sensitive landscapes. For example, greater sage grouse have been shown to avoid habitats near transmission lines, which can limit mobility and reduce access to safe relocation options.

Nationally Significant Infrastructure Projects

Last month, new rules came into effect that allowed data centre developments to be classed as Nationally Significant Infrastructure Projects.

In short, this means they are exempt from various steps in the planning and approval stage, with the intention being that data centres can be more efficiently approved.

Pre-application consultation requirements for Nationally Significant Infrastructure Projects will be replaced by “earlier technical support and meaningful advice from the Planning Inspectorate before applications are submitted” through a redesigned pre-application service for developers. The changes are delivered in the Planning and Infrastructure Act which also places limits on legal challenges to Nationally Significant Infrastructure Projects.³⁵

This does not remove all obligations on developers. For example, the 10% biodiversity gain still applies.³⁶ But in the eyes of one planning consultant, at least, it does provide an alternative route through the planning system, “perhaps where local level buy-in to the value of data centres to the UK economy is slow in gaining political traction.”³⁷ If I am interpreting that statement correctly, it is suggesting that the value of a data centre to the wider economy will end up trumping the concerns of those locals who are most affected by it.

The UK government views data centres as “essential services” and, if you look at their reasoning, I have some sympathy.

The vulnerability of this infrastructure was starkly illustrated in July 2022, when 2 separate data centres serving an NHS trust failed during a heatwave. The incident took down most clinical IT systems at three hospitals, as well as related community services. The disruption to patient care was massive and widespread.³⁸

If data centre growth was being driven by a need for more robust data services for the NHS, then I would be more understanding. However, although that forms one part of the problem the government is trying to solve, we have already shown that the larger driver of data centre demand is AI. Strengthening critical health, defence, and industry services are not the same thing as needing more data centre capacity so you can tell an AI to write you an email, let it decide what you should cook for dinner, or get it to make you a deepfake nude of your ex-girlfriend.

I can support some data centre development. I cannot support the large-scale explosion of the industry when it is being driven by a need to cater for people who have decided to outsource their creativity and critical thinking to a machine.

Mass adoption of generative AI is driving demand for more and more data centres and the costs of those data centre developments are felt most keenly in the local communities and environments in which they are built. The issues they create also ripple out through the country’s creaking energy grid and stressed water supply.

The profits they generate flow elsewhere, however, back into the hands of the AI companies.

It’s not NIMBYism to want all of the issues with energy connectivity, pressure on the water supply, loss of habitats, and risks of air and noise pollution to be properly considered before a data centre is given the green light. It is essential in this time of accelerating climate change that all of these things are joined-up, so that the new data centres are using renewable energy from day one and aren’t putting undue pressure on an area’s water resources. Noise and air pollution needs to be managed up-front, not with hard-won and less-than-completely-successful remedial work. And any impacts on local habitats need to be kept to a minimum. This calls for greater scrutiny, not a fasttrack process.

What can be done?

Sign the petition calling for an urgent moratorium on AI Data Centres and BESS outside AI Growth Zones: https://petition.parliament.uk/petitions/776949
Register with CPRE Devon’s campaign against the Devon Data Centre: https://devoncpre.org.uk/data-centre-and-battery-field/
Find out about plans for data centres in your area and ask whether those developments have properly considered all of the issues discussed in this article: I cannot find a single list for England and Wales but there is a handy map of proposals in Scotland

And, above all, stop using generative AI. Less demand means less need for new data centres. If they must be built then let it be because we want better infrastructure for the NHS, not because we can’t kick our ChatGPT addiction.

A longer rant about why water use calculations aren’t helpful for most people

As I said above, one of the loudest talking points about generative AI and its environmental impact is water use. In short, every data centre needs water. In many, it is used as a way to cool down all of the hardware; it’s not the only way, but it is a popular method for a variety of reasons. These sites also need a lot of electricity to power them and water is used in a lot of electricity generation.

Water use in data centres has become a major bone of contention.

Why? Because different people keep calculating the amount of water being used and then deploying those numbers to argue against or in favour of data centres. Which is confusing.

Let’s look at a few examples, starting with an article from Business Energy UK. Following on from the influential Washington Post article mentioned above, the Business Energy UK article runs several comparisons to come up with attention-grabbing summaries like this:

Globally, ChatGPT uses around 39.16 million gallons daily, the equivalent of everyone in Taiwan flushing their toilet at once. Over a year, that’s enough to fill Central Park Reservoir seven times.

On the other side of the argument, Andy Masley, a writer on AI topics, has written several very long articles about AI and water use. His headline view is: “The idea that… AI is using an inordinate amount of water that merits any kind of boycott or national attention as a unique serious environmental issue is innumerate.”³⁹ To support this view, he also runs a lot of calculations and comes up with some numbers. I do not have the expertise, nor the energy, to cross-check his numbers and, honestly, I don’t think they are likely to be wrong.

I just think his numbers, and those that Business Energy UK calculated, are functionally useless.

If you read his article, you will find lots of assumptions, lots of things included or left out, and all sorts of other decisions that were made when coming up with the final calculations. That’s very sensible. But it also means that they paint an incomplete picture. You can decide to include or exclude things as you see fit (and so that they work best with the argument you’re trying to make). Unless you work in this sector and need this type of number for a specific purpose (and are conforming to a set methodology that allows proper analysis to be made), then I don’t see much value in them.

If you can get the numbers to agree with your argument, whichever side of the debate you’re on, then the numbers stop being useful.

The second thrust to this type of argument is to compare generative AI’s water usage with another industry’s water usage and say “Look! This other thing is even worse. Ergo, AI isn’t the problem, this other thing is.” You can see this in Andy Masley’s article (and I’m not targeting him specifically, it’s just that his writing keeps getting recommended to me by people online when these types of discussion crop up so it seems sensible to use him as an example of this type of argument).

So the water all American data centers will consume onsite in 2030 is equivalent to: 8% of the water currently consumed by the U.S. golf industry… [or] the water usage of 260 square miles of irrigated corn farms, equivalent to 1% of America’s total irrigated corn.

Again, I have no reason to doubt the accuracy of his numbers. But this type of “gotcha” revelation feels like it is intended to shock people into thinking that data centres aren’t a problem because they are so negligible compared with these other industries. For me, it has the opposite effect. If you’re putting data centre use in the same sentence as two of the most infamous industries when it comes to wasteful water use then I am seeing the similarities more than the differences.

And it is not as though they are either/or situations. We aren’t choosing to run a data centre or grow corn. To build a data centre or a golf course. We’re doing all of those things at the same time. So what you’re really saying is that we still have the (already very well known) issues with water use from the corn and golf club sectors and we are adding another industry on top of that. So things are worse. More water is being used overall.

If your argument boils down to “Data centres can’t be bad because this other thing is worse” then I don’t buy it. I am perfectly capable of holding more than one idea in my head at the same time. I can acknowledge that this other thing you’re pointing at is bad and I can think that data centres are also bad. You’ve just given me two things to worry about rather than one.

Water use is a major recipient of this type of argument, but generative AI’s wider environmental impact also comes in for similar treatment. Here’s a Forbes article from December 2025: “AI Is Almost Green Compared To Netflix, Zoom, YouTube”

In this article, they are looking at overall CO2 emissions, rather than water use, but you can see the same arguments being deployed. “AI is almost green” they say in the headline, which is a bold statement when the article itself reminds us that “It’s not as if any digital use is necessarily green activity, of course. The tech sector as a whole uses a huge amount of energy which is a significant contributor to global pollution.” Again, to me, this isn’t an argument for generative AI, but an argument against tech in general. If you are telling me that using generative AI is less harmful than watching Netflix for an hour⁴⁰, then my conclusion is not that I should start using generative AI, it is that I should carry on not using it and I should cut down on how much stuff I watch on streaming.

If you jumped down here from the top of the article and want to know how I would recommend we think about the environmental impacts of generative AI, then you can resume your reading here.

You Might Also Like

Inside the Art Automaton

Bad Actors and Bad Acts

The World According to AI