> Using the prompt, “How much water is likely to evaporate from data centers in California per year, assuming they are all using mostly evaporative cooling?” several free AI websites provided ranges of estimates, below. These AI also can provide ranges and sources for calculation assumptions.
Data centers with closed loop cooling systems are absolutely built all of the time. Total evaporative cooling has the advantage of being more power efficient (and therefor cheaper) - the only reason they bother with total evap is because the water is being offered plentifully and cheap.
People have no reality of how cheaply priced industrial water is in this country. My parents had a cherry orchard and their annual water bill was $100 an acre per year for as much as they wanted. Which is why the water consumption for data centers is only still a fraction of what we lose to evaporation from inefficient spray irrigation.
Yeah, there are alfalfa fields in central Arizona. Alfalfa basically turns water and sunlight into cellulose about as quickly as plants can.
Worse, the owners of those fields are often foreign companies. That means they use tremendous amounts of water in one of the driest regions on earth, in the middle of a multiple decade drought, and the wealth these farms generate disappears overseas.
Part of the issue is not systematically using a pricing structure that charges disproportionately more for usage above high thresholds.
The 101-level "solution" is to just raise the price to account for demand. The problem with that is that it treats all usage the same, whether it's a residence's first gallon or an alfalfa field's last gallon. But the former is something we need to protect.
It makes sense to price water, and electricity, in a fashion where the first X costs a certain amount, and the next X has a higher rate, and above some percentile of usage it has a much higher rate, and at some percentile of usage, customers should be very nearly paying for new required utility infrastructure themselves. That allows using pricing to solve supply problems, without penalizing normal levels of usage.
Some utilities already do this. But if there are actual issues with having enough supply for both datacenters/farms/smelters/etc and residential usage, then they're not doing this well enough, or don't have the pricing correct.
> People have no reality of how cheaply priced industrial water is in this country.
Driving between SF and LA, you see a gazillion signs from water leeches complaining that the government won't give them yet more nearly-free water. No, I don't want to go without fresh crops. But yes, I absolutely believe that, say, growing almonds in basically a desert should be a financially expensive operation, and if that makes the end result more expensive, then so be it. And if that means it's no longer viable to empty rivers for the sake of a tasty bag of snack nuts, I can learn to live with it.
This is a really awkward situation because while we'd really want the market to sort of auto-balance the costs between different suppliers it's also really hard to look at PE ratios right now and believe that the market is anywhere near sane. OpenAI could trivially monopolize the water supply in CA[1] with its current warchest and that would, for everyone, be terrible in some fundamentally obvious ways - so we've clearly got a pretty gigantic misalignment in the market which means we're reliant on the government specifically picking winners but ideally doing so in a sensible manner.
How many well cooked dinners is a prompt worth? Not nearly as many as the market currently says. If it were anything less vital we could probably just ride it out until the bubble bursts but if acceleration continues then in time water usage might actually rise to the levels that the most fear mongering folks are saying it's at.
1. Accidentally even - without even reaching into the realm of malicious intent.
Also just because something is cheap doesn't mean it's not depleting resources and making life worse for a community somewhere. People are constantly trying to build pipelines to the west to deplete the great lakes. There is a societal and ecological limit and these AI companies are not worth it.
"closed loop" or "total evaporative" cooling is a myth and a marketing term not a scientific one. Maybe learn about something before prompting AI and spilling out some bs.
Comparing water usage of AI to agriculture and cities is a little misleading. The cities' water usage is to keep people alive with basically mandatory things, like hygiene, and drinking. Agricultural water usage is required because we have to eat to live. Don't compare something optional to something mandatory.
Instead, compare AI water usage to that of optional things in a city, such as car washes and water parks. Or compare AI water usage to that of what it would take a human to do a comparable task (what does it take to keep a human alive for a few hours compared to running a 15 minute long task to write a report with AI?). While AI water usage might still not look that bad, it would be a more honest comparison.
Oh, so that's why we're growing alfalfa in the middle of deserts, flooding the fields with excess water so we can keep water rights, and then shipping the alfalfa to China. It's so we can eat!
If we're shipping the alfalfa to China, I assume that means it's supporting some Chinese person's food source, whether they are directly eating the alfalfa, or some animal is eating it that later becomes food.
If someone is flooding a field unproductively just to use up their quota of water, that is a bad thing that should be addressed. But even if you excluded that unproductive usage and compared AI water use to legitimate agriculture use, that would still be an unfair comparison. If you were to compare AI water use to the amount of water that people are wasting just for legal reasons, then I honestly think that would be a pretty apt comparison.
Don't be disingenuous. They already were dividing things out by type of usage, like talking about water park usage vs. the usage of an entire city for all purposes. They are already admitting that "water usage of a city" isn't only about quenching thirst and maintaining hygiene, it's not a stretch to assume that they also realize that they can be water wastage in agriculture as well. They can't split out every instance of wastage that could be eliminated, and it's ridiculous to expect them to.
My wife works with farmers professionally as part of a conservation district and just responded "THIS PERSON KNOWS FARMING" when showing her the discussion. I genuinely have no idea what you guys are talking about but she immediately got heated.
There was massive controversy about that so I don't know how good counterexample it's that. Unless the argument is "we already waste a lot why would you care about wasting more??" Which is not a great argument.
A pretty easy 'optional' comparison would be golf course watering. I saw a much more detailed write up on this that I can't find now, but a quick google shows 500 billion gallons a year for US golf courses and 180 billion gallons a year for all data centers, not just AI data centers.
> The cities' water usage is to keep people alive with basically mandatory things, like hygiene, and drinking
Almost half of city water usage is for residential landscape irrigation, mostly spraying lawns, which is not exactly mandatory or a basic necessity. Landscape irrigation uses about 3.5 million acre-feet / year, which is 1 to 2 order of magnitude higher than the estimated AI data center usage.
In the article it lists a data point that beer production in Arizona used more water than the data centers in Arizona. People may vehemently disagree, but we absolutely do not need beer. Would I trade beer for AI? That's an easy choice, AI every time. If you just keep track of the water to keep a person alive and the bare minimum water required for agriculture (which isn't particularly efficient in most cases), it would be a fraction of a fraction of what we use now.
Comparing data centers to the bare minimum isn't particularly interesting, the point being made by the article is that we aren't efficient with our water usage in general, AI is a rather small source of waste in the scheme of things.
This is a hilariously misleading "study" and I would bet "beer" wasn't chosen arbitrarily for comparison:
The important difference WRT beer is that the water used in the process likely in a larger part goes towards... the beer itself. This in turn is going into the person who drinks it. So, the water here is actually hydrating human beings.
This can be argued as one of the 2-3 absolutely necessary uses of water. Hydrating people.
So, spending less than the beer industry is not that great of an achievement.
However, a casual reader may see comparison to "beer" and think "oh yeah, beer, just a random thing out of a million, so yeah AI is totally ordinary".
Beer has been around for like a thousand years and we haven't decided to get rid of it. We're five years into this fever dream and everyone either literally hates AI or has been driven at least a little crazy by it. It's a pretty darn easy choice for me (and most people I imagine).
This is an extremely frustrating angle to take because what you're implying is that anytime anyone comes up with any system that takes water they should go in front of a panel of experts (seniors) who get to decide whether their water usage is for an "approvable" purpose. Now I don't like water going to Golf courses either but to me even the intermediate solution is to price water accurately.
Barring that, long term we're surrounded by 70% body of water with infinite energy beaming down on us, this feels like a solvable problem without having large swaths of the country fight over scraps.
> anytime anyone comes up with any system that takes water they should go in front of a panel of experts (seniors) who get to decide whether their water usage is for an "approvable" purpose.
This is absolutely how things work, the water for farming and industry is cheap by design (at least in the US) so that people will have relatively cheap food and consumer goods.
Now you can absolutely try to go change that to a strictly capitalist "One gallon of water is 1 cent, whatever the usage", but you'll have a hard time finding a political group in this country that stands behind such a principal. Even the most conservative groups typically back farming subsidies.
We absolutely do not need to waste as much water as we do on agriculture. Their is more efficient watering systems, crops that do not feed humans, and inefficient crops that aren't needed. Any one of those improvements would dwarf the water usage by AI.
Heck, a better solution yet would be to charge these AI/datacenter companies enough to cover the costs for watering efficiency systems to cover their usage and then some. It's a fraction of their costs, and way better than being anti-growth.
Yeah people aren't mad about datacenters because they are "anti growth"
They don't want to see their local resources depleted and, no, this isn't some fantasyland where corporations will do anything "for the greater good" that isn't in line with their pockets.
> Or compare AI water usage to that of what it would take a human to do a comparable task (what does it take to keep a human alive for a few hours compared to running a 15 minute long task to write a report with AI?).
A very strange comparison. It seems to imply that we "need fewer humans" because of AI. It also assumes AI is primarily used to replace useful human work, something I very much doubt.
> The cities' water usage is to keep people alive with basically mandatory things, like hygiene, and drinking. Agricultural water usage is required because we have to eat to live.
Drinking water is barely a rounding error in cities' water usage.
Agricultural water usage doesn't go to the necessities to feed people. It goes to whatever is most profitable, even if that means growing water intense crops and exporting the produce overseas.
But why stop there, and why exclude all food equally? Does somebody living a vegan lifestyle (which typically needs vastly less resources, including water, per calorie of food produced) get to wash their car in exchange for their trouble? What if I take a cold instead of a hot shower; do I then get to wash my bike every once in a while?
Basically any discussion of water allocation is stupid. We already have a way to allocate water (or, really, any scarce resource) - markets. Instead of arguing over whether or not a hamburger is worth a car wash's worth of water, bill the person using the water for that amount of water. Let the water user and the price discovery mechanism fight it out. If it is not worth it to them, then they can move to somewhere where water is cheaper.
We don't do this, at least not in the western half of the US. Instead, the biggest consumers of water have "water rights" - the right to use a certain amount of water every year, for free, simply for owning a particular piece of land. And these water rights were all staked out based on estimates of the Colorado River that were wildly optimistic, so there's a century-long waiting list of claims that will permanently supercede your own if you fail[0] to actually consume the water you are entitled to.
This is insane, and it leads to some pretty insane incentives. Because agriculture was here first, it has the strongest claims to water, and a pretty heavy incentive to waste as much water as they are legally allowed to. A lot of the discussions surrounding water usage assume that because agriculture is necessary for human survival, that the water it uses is also necessary. It's not - and the only way to get an industrial water user to actually care about their water usage is to actually bill them for it.
Once we have an actual market for water (not just water "rights"), then we can start talking about what usages are actually necessary - i.e. what uses should we explicitly subsidize through taxes rather than implicitly subsidize through a terribly designed system.
[0] In the interest of fairness, I ran this comment through Google's chatbot, which would like you to know that TECHNICALLY SPEAKING, it takes ten years of intentional disuse to lose a water claim, and that there is a market for water rights. My counterargument is that most farmers do not care about how much water they can not use, and that a market for water rights is not the same as a market for water, because farmers can still decide to just use the water for free. The pricing mechanism cannot work if there are a class of protected users who do not feel backpressure from the pricing mechanism.
A lot of agricultural water usage (more water than AI) is for growing corn to turn into ethanol so we can add it to gasoline. It's not a small amount either, 40% of all corn in the US is used for this purpose.
We use about two orders of magnitude more water (each!) on corn and alfalfa than on data centers as of 2023, and while we're ramping data centers up fast, it'll still be an order of magnitude at the 2030 data center estimates (which may heavily overestimate, now that there's so much opposition popping up).
What about golf courses which use up 476 Billion of water every year? Way more than data centers. People complain about Nestle using water in californa for bottled water but it doesn't compare to what single golf course uses in a year.
There's not really any NEED to grow almonds. Most agriculture in California is not required to sustain life in CA. However, without AI people wouldn't have jobs that could afford CA rents, so AI is required so people can live. Lets get rid of unnecessary uses like agriculture, unless farmers can justify that the usage is actually required to sustain life.
If you look at water distribution you'll find that its unevenly distributed so farmers should pay a water tax and distribute that water to the less water fortunate. CA has an extremely high water GINI with a few farmers consuming far more than their fair share.
This is even more misleading. You have to eat to live, but absolutely not all water usage for food is mandatory.
If you gave me a budget of how much water I could "use" water every year, and I was close to going over, I could easily pay for my annual AI use just by changing what I eat for lunch on a day or two. I could pay for years of AI use just by forgoing buying a new pair of jeans.
The water argument has always felt so intellectually dishonest to me because it's never approached from the perspective of "hey, we're using too much water, how can we conserve it?" If we approached it from that perspective, reducing AI usage would not even crack the the top 100 list of things we would do. But that's not the goal of the water argument, because it quite obviously actually has nothing to do with water.
This is the response to have in mind when confronted with AI-water arguments. It's not about HOW the water is used, it's that, if you're truly concerned about water usage, AI is a non-factor compared to basically everything else you do on a daily basis.
My understanding is that data centers (at least in LA) are using mostly grey/industrial water, not water you can consume or use for agriculture. It feels like we're measuring water as one entity when not all water is equally useful to a human.
one of the biggest health problems in US is obesity.
30 to 40% of the food produced in US goes to waste.
Just these two facts will tell you that while, yes, we do need food to live, but on another hand we have an abundance of food and if AI data centers use 0.05% of the water used for humans.
Yes and no. We shouldn't compare datacenter water usage to residential water usage. We should compare it to industrial water usage, as that is what it is. The question like "how does datacenter water cooling compares to concrete factory water cooling?" makes some sense from engineering perspective, as you are comparing oranges to oranges to a degree.
Residential water usage is way too different in way too many ways to be meaningfully compared to industrial usage. The scale is different, the waste water treatment is different, the infrastructure cost is different. The water quality standards are different...
Rice is not a luxury for most people. It’s a staple. It uses ca. 40% of all irrigation water globally. Also cotton is not a luxury, though it also uses quite a bit of irrigation water.
While a couple months back an article[1] discussed how Google was keeping the water requirements a secret from locals who wanted transparency, claiming it was proprietary knowledge.
So they sued and discovered it will use 2-8 million gallons of drinking water per day[2], seemingly near the limit of their capacity to handle, judging by comments from officials.
> 'That water supply that otherwise would not be required until 2060 or the 2060s, suddenly becomes something that we need to be worried about during the 2030s.’
> If it exceeds that demand, they’re going to have to start looking for a new water source.
So I'm not sure how this fits with the claims of the article from the OP. I suppose if anything it disproportionately affects certain places not as well equipped for it?
You can go millions of prompts before you use up as much water as it took to make a single beef burger.
You can go tens of thousands of prompts to match the C02 emissions.
There are many legitimate concerns around AI. Water use/CO2 emissions isn’t currently one of them. Going vegan will make up your AI water consumption/CO2
Emissions many thousands of times over.
Now measure the amount of electricity the same prompt will use in 6 years when both algorithmic efficiency and 3-4 generations of silicon lower that by 95% (or more).
Will your microwave become 95% more efficient over the next 6 years? No.
Also how many video prompts will the average person run in a given year? Almost certainly 0. I heavily use AI daily and have probably played with AI video less than 4 times, ever.
Yet certainly the average person will use 20,000-100,000 microwave minutes over their lifetime. I use my microwave for 2-3 minutes every day at lunch for example.
From first principles, the idea that electricity use = bad is wrong. If your electricity comes from burning coal or lignite, then obviously yes using that electricity has bad externalities.
But a french person running their microwave on Nuclear powered grids? This is good. Dirty energy sources is the problem. Not energy use itself.
Obviously, there's different options and variables and bla bla bla, but considering how consolidated and highly industrialized and standardized meat production is, this data is very likely close enough to true for the wast majority of beef burgers eaten by the people complaining about AI resource consumption: https://ourworldindata.org/environmental-impacts-of-food
> U.S. specific estimates put beef water use at 317, 441 and 808 gallons per pound of boneless beef when precipitation water is not accounted for in calculations.
So, let's just say around 400 gallons of water per pound of beef if you don't include rainfall use.
Let's look at Almonds which _could be_ produced where water is not scarce, maybe, but instead is grown in central valley CA based on quickly depleting ground water.
Let's be generous and say that the Almond farmers hit all of their future efficiency goals, so each almond only takes .5 gallons, and that the drop/gallon math is off by a factor of 2.
That means eating 1 almond is about ~4,000 google searches.
I have a few cows and rarely ever give them water. In the winter they get enough from snow and when it’s rainy we have a small pond that forms with a stream. They also prefer either of those to drinking well water from a cattle waterer. They are grass fed and rarely get fed stuff like corn.
For for thousands of pounds of beef, I’ve barely used any water at all. Don’t notice the extra consumption on my well at all, and I have a very low producing spring fed well (1 gallon per minute).
“Vegan” crops on the other hand line corn which are irrigated in many parts of the country use a great deal of water and often very inefficiently so.
Well, I've got a small server rack and roof top solar, therefore data centers don't actually use water.
In other words, bringing up some anecdotal, hyper specific (how many meat eaters just "have a few cows"?) information says absolutely nothing about the truth of the matter, but a lot about what you believe constitutes an argument.
> So much of our public discourse on water and other subjects is choked by chatter, untamed by reasoned evidence, data, and quantification. Today, with AI, we have little excuse for not attempting and using honest estimates to inform our discussions and tame our fears and hopes.
Are these things usually convincing? The general pattern is that people take a position on something and then find one paper with a DOI identifier that backs the position. The Elephant and The Rider and so on. Trying to provide someone with evidence of the falsehood of their claims rarely makes them reconsider and often makes them dig their heels in while they search for a new paper with a DOI identifier.
We're in an unprecedented time in the information age when people can rapidly achieve basic competency at many things using Wikipedia, Google, and LLMs critically. If information availability and search were the constraint, one would expect us to reach greater alignment with facts.
The reality in the way information is used, I believe, is the opposite from what we think of. We believe that if there is sufficient information, we can use it to form an accurate model of reality. In practice, I think the social psychologists are right. For the most part, we form the model of reality and then we seek information that supports it. So if you increase the total amount of information what you do is increase the ability for someone to select out that which supports their model.
That's not to say I don't appreciate these things. It's just that I don't think facts move public opinion very much.
The interesting thing that more information and better search provides is that it accelerates the divide between truth-seekers (who will attempt to find primarily falsifying information) and confirmation-seekers (who will attempt to find primarily supporting information). In general, one can imagine that the former will be more successful at modeling the world ex humanity at least. But if others believe something is true, often a direct approach at their facts is not the best approach to get the outcome.
> The reality in the way information is used, I believe, is the opposite from what we think of. We believe that if there is sufficient information, we can use it to form an accurate model of reality.
You should read Yuval Noah Harari's Nexus. He calls this "the naive view of information," which is ignorant of the existence of what he astutely identifies as "intersubjective" realities (see also Angela Cooper-White's entry on "intersubjectivism" in The Encyclopedia of Psychology and Religion):
[...] its deepest and most complex
usage is related to the postmodern
philosophical concept of
constructivism or, in social
psychology, social constructionism –
the notion that reality is co-
constructed by participants in a
relationship and in society.
This is the endgame of postmodernist and constructivist thinking that exalts narrative and story as the ground source of truth. In some ways what we are seeing is a return to religious and superstitious thinking where sufficient belief in a dogma or a pantheon is enough to reify those narratives into consensus reality.
Historically Jungian psychology and indeed religion (a form of proto-psychology, from which Jung inherits by way of the alchemical tradition; see Jung's Psychology and Alchemy) was humanity's collective storehouse of wisdom and techniques for managing intersubjective realities and group "information hygiene." Such techniques are now being lost to antiquity with the late 20th and 21st century focus on only objectively verifiable, quantitative measurements (as opposed to the private subjective, qualitative phenomena experienced as the inner ruminations, contemplations, and dream life of the individual).
White Rose: Do you ever think
that if you imagined or
believed in something, it
could come true... Simply by
will?
Angela: Yes. Actually, I did
believe that. But I'm slowly
having to admit that's just
not the real world... Even if
I want it to be.
White Rose: Well, I guess it
all depends on what your
definition of real is.
This is a bit of a dead horse, but the magnitude of how off the public is on this continues to amaze me. Pete Buttigieg did a Tulsa town hall a week or so ago where someone cited it taking "10,000 gallons of water just to generate one photo".[0]
Honestly, it's weird to me how fixated both sides are on water.
People against data centers overestimate water usage, but people who think we should build as many as we can, as fast as we can seem the think that "actually they don't use that much water" somehow negates the more real issues with them.
Water is pretty scarce in some of the places they want to build these things. I know people in West Texas that own ranches that have been approached by the datacenter people and it’s basically a desert, oil industry consumes a lot of their water, and the public water they get in the city smells toxic, the well water is flammable. So water use is concerning and I don’t think there’s any reliable or trustworthy source for them to use as a gauge for what to expect so they have to ask.
What are the real issues with them in your estimation? Anti-data-center people bring up water use as a reason why the government should legally prevent data centers from being built, and pro-data-center people bring up water use to argue against the anti-data-center position. I agree that the anti-data-center people are overestimating water usage, as well as the degree to which the amount of water data centers do use is a problem; and that they're doing so because they have some other objection to data centers that doesn't sound as convincing. It would be better to talk about those issues.
The other part of this problem is the idea that if you disagree with someone about the facts you're interpreted as disagreeing with them about the thing they're mad about: You disagree that AI somehow destroys fifty billion-trillion gallons of pure water every time someone asks Claude something, therefore you're fully in favor of Grok making nudes of underage girls.
Some people get an Angry. They love their Angry, and nobody will take it from them.
That's populism for ya, and it's sadly extremely effective.
Meanwhile, both China and India are giving free electricity, providing dollar-for-dollar capex subsidizes, and 25 year tax exemptions to build data centers [0][1].
Love how HN wants to strangle the infrastructure that underlies our entire industry and why HNers get paid. It really highlights how much of the opposition to AI comes from the "chattering classes" and other white collar types as is constantly seen in polling [2][3].
It's funny seeing people who are also part of my party but told coal miners and autoworkers to "learn to code", treated blue collar workers derisively, and ignored concerns by employees in manufacturing and skilled trades which led them to shift to the right now act the exact same way.
Edit: can't reply
> AI Datacenters are not how all or probably even most HNers get paid
Most data centers colo multiple types of compute, not just those dedicated to inference or model training. Additonally, strangling the economics of the infrastructure layer makes entire ecosystems move abroad.
You saw similar opposition to semiconductors fabs back in the early 2010s in the US, and the entire ecosystem virtually out within a decade until the CHIPS act was signed and executed on.
Same with nuclear power in Germany and GreenTech in much of the America.
Populism is effective because the proverbial rising tide is not lifting all boats. AI is currently driving yet another extreme wealth inequality inflection point. Founded just five years ago, Anthropic is going to be a trillion dollar private company maybe this year! This is a staggering outcome and will further divide the gap between the wealthy and everyone else.
So whether populist outrage is expressed through fears of job losses, higher energy prices or concerns over water usage, IMHO, wealth inequality is the cause.
AI Datacenters are not how all or probably even most HNers get paid...
> Most data centers colo multiple types of compute, not just those dedicated to inference or model training. Additonally, strangling the economics of the infrastructure layer makes entire ecosystems move abroad.
Sure but we are talking about whether the enormous investment into AI infrastructure is prudent or not. Also I reckon most people on here made a living just fine before everything moved to remote data centers, and many if not most HNers workloads could run on individual machines... But that's another conversation.
I suspect soon young learners of the future may tilt their heads in curiosity when finding that Obama was a "Democrat" in the same way they did in the past when finding that Lincoln was a "Republican".
you're arguing against things that have no material effect. "oh won't you think about adversarial discourse about the most well funded industry in recent history"
Compare it to alfalfa and you’ll be laughing your ass off at how much water alfalfa consumes.
~340 acres of alfalfa in California growing year round uses as much water as Google’s data center in The Dalles uses in one year.
That data center used 550M gallons for evaporative cooling in 2025, which is 1687 acre-feet of water.
One acre of alfalfa in California uses ~5 acre-feet of water per acre of alfalfa per year. There are around a million acres of alfalfa grown in California, or 5 million acre-feet of water per year on alfalfa. Which is used to feed cows.
Feed cows in places without the water and sun to grow this stuff locally. Which is tantamount to exporting water from the American West which will eventually be turned into a desert. We effectively can't be trusted to govern our natural resources more than 5 years out.
I’d recommend you read the following report: Proximity to Golf Courses and Risk of Parkinson Disease
Individuals living within water service areas with a golf course had nearly double the odds of PD compared with individuals in water service areas without golf courses…
Did you read the paper carefully? It's about pesticide use. (It's not especially plausible as epidemiological studies go, though I'm unsurprised if a better study finds a firmer correlation between pesticides and PD.)
I don’t really get the water concerns in datacenter cooling. Even if a lot of water was used for cooling with every prompt (which he argues against here, but, even if)… water “used up” by cooling just comes out a little hotter, right? Maybe evaporated. Then it’ll come back in the form of rain. This isn’t an industrial chemistry process that leaves some toxic waste in the water. Or an agricultural one that puts water in plants and then ships it off to some other region. It just becomes another path through the water cycle.
I actually don’t get how this can be a real thing that people are worried about. Is there some astroturfing behind this? Maybe an attempt to make environmentalists and AI skeptics look stupid?
The absolute strongest complaint is that DCs consume treated, potable water, which is less abundant / easily re-created than any old non-potable source. (Of course the easy solution here is DCs just ingest / treat their own non-potable source. Or utilities charge rates sufficient to price in the externality of drawing down more potable water. The economics still work for DCs if they need to treat their own water -- the fundamental problem is that utilities are underpricing their potable water, so DCs prefer it all else being equal.)
Why don’t data centers use gray water more often? Wouldn’t that be better for basically everyone?
My guess is it’s some combination of the infrastructure not existing, the distribution being bad, and the treatment costs not penciling out.
But that feels like the kind of thing municipal utilities could solve with pricing. Potable water should probably be priced differently for residential use than for big commercial/industrial users, in a way that pushes them toward non-potable sources wherever possible.
A fun Texas water fact I always bring up: the entire state’s monthly freshwater use is roughly a week of freshwater inflow into the Chesapeake Bay. Texas would be the 8th-largest GDP in the world if it were a country, and its whole monthly freshwater demand is basically a few months of water that the Chesapeake just dumps into the ocean. (Of course, estuaries make use of the water so it's not just wasted but it's illustrative imo)
Another fun comparison point is yearly Texas uses 0.08% the volume of the Great Lakes in freshwater but ~ 30-50% of the volume of all the lakes in Texas.
We've got a lot of water but it's not distributed evenly and we should probably build some sort of water pipeline eventually so water rich states can sell to water poor states.
Again, this is all just speculation by someone who knows not a damn thing about municipal water management.
Because they're taking water from already parched regions, often pumping it out of the ground. Even if the water did come back locally as rain (it doesn't), it still makes it impossible for people to live off the same aquifers and water sources sustainably.
If only we could do water-intensive activities in areas where water is abundant and then ship the resulting products to where they are needed...
In a far science fiction future, I could e.g. imagine connecting LLM inference data centers to a global data network instead of always having to drive up to them to ask my prompts.
It doesn't come out a little hotter, it gets evaporated in cooling towers. Same result as any other water usage. Cooling towers can't use seawater either. Most datacenters are in places where fresh water is abundant anyway, but some are not.
Anyway agricultural water usage is way worse in California.
The water isn’t gone but if it comes back as rain, it at least has to be cleaned again, since data centers probably don’t use raw rainwater for cooling.
It’s probably still not too bad but there’s at least some work done that’s „used up“ by letting tap water (or probably demineralized water used for cooling) evaporate.
The problem is that data centers use SO MUCH water... sure we humans let water evaporate, but this is a new source of water "waste" to the tune of nearing 2 billion gallons/year, just in Loudon County Virginia & connected water users [0].
When that water source is underground wells, this can take years (on the fast end) or decades (on the moderate end) to get back down. Look at California's water issue -- so many wells extracting water for farming has changed the land topography.
Also, when water 'comes back', it might come back in the ocean and not on land... reducing the available fresh water without desalination.
Data centers need the water to cool... but maybe there's room to find incentives for them to do so while making sure our water bills don't go up like our electric bills are because of the extra load they are putting on utilities.
The rain doesn’t happen directly above where it evaporates. And “slightly warmer” waste water can have major ecological impacts, destroying native life in the lakes and rivers where the wastewater is ejected. Plus, if the water is taken away from underground aquifers that may not be refilling fast enough, or if it’s taking water from downstream users, that’s something to be concerned with.
I have also wondered this and came to a similar conclusion about the politics.
This whole time I've been wondering how it's possible that people don't realize how common evaporative cooling is for much larger buildings that are far more numerous than these data centers, and especially in dry climates where drought is common.
tldr: chip immersion uses less water but is more expensive. Water evaporation is the opposite. Datacenters will use the cheapest they can get away with. Water is scarse; evaporated water is as unavailable as contaminated water. Read the information sources.
The explanation about chip immersion is wrong though. It's not a water-saving technique, it's for cooling dense racks more rapidly and maybe saving energy. That warm coolant still needs to be cooled, likely the same way, evaporative cooling tower.
Worse, the article says "Since the technology uses synthetic fluids, it requires significantly less water than other approaches." This is like saying that a new coolant-based car radiator doesn't use water while a very old one does, like yeah technically it uses it but doesn't spend it.
> tldr: chip immersion uses less water but is more expensive. Water evaporation is the opposite. Datacenters will use the cheapest they can get away wi
This suggests a simple fix: charge more to the datacenters (not people) for the water, to make the other option competitive.
No need to throw baby with the ... erm, bathwater.
By that argument water use is never a bad thing since all water comes back as rain. The problem is that data centers need to use clean water, which has to be treated. On a local scale, a large data center could starve a community of potable water, even if the state-wide water use is very small.
> 6. Using the prompt, “How much water is likely to evaporate from data centers in California per year, assuming they are all using mostly evaporative cooling?” several free AI websites provided ranges of estimates, below.
Is this what passes for a citation nowadays? I’m sympathetic to the message but this ain’t it.
The data coming from the University of Calgary about the data centres they're building in Alberta, Canada seems to indicate that they're using evaporative cooling, which is very expensive water wise.
The bigger concern though, is the power requirements. Which are set to double or triple the energy use of the entire Province (analogous to a State in the US).
Air pollution, GHG and water use are concerns, but these projects will not dramatically increase the load on the electric grid.
Natural gas is cheap and abundant in Alberta, and the province (actually the whole country, via transfer payments) benefits financially from resource revenues from extracting the gas. So, these projects are generally an easy sell to the public.
I keep hearing about natural gas and on-site power for these data centres. I'll believe it when I see it.
There are already have a couple in Calgary and they're hooked directly to the grid. The cost of electricity for the city shot up at the same time. Also, there have been a few brownouts caused by them not being ready to handle late night draws from those data centres.
That's at least what I'm seeing. Though, admittedly, it's from older project articles. Maybe something has changed in recent months?
I stopped reading at the crappy ChatGPT comic that shows "water usage" as some pipes pouring water and others pumping it in. How trustworthy is the text going to be after that?
I don't find it weird, it's about what I'd expect from AI sycophants. They don't seem to realize that it comes off as not even being able to defend their own ideas.
A lot of confusion around AI water usage might stem from whether it's an open-loop or a closed-loop cooling system.
e.g. an open-loop system which disposes of waste heat through evaporation is naturally going to draw a lot more water than a closed-loop system which recycles the water. Open-loop is likely cheaper to build, and importantly, it _does_ use up a lot of water that could otherwise be going to a municipality.
So, what's the actual breakdown between these two? I absolutely _could_ imagine many datacenter operators cheaping out and using open loop cooling, particularly if building next to a source of fresh water like a river.
A lot o the confusion around data centers is that these companies purposely hide this information from the public. We already know how damaging normal data centers are:
To simplify things, "closed loop" shouldn't even be part of the discussion. Usually they just mean a closed-loop cooling system somewhere inside, either to directly cool machines (typically ML) or to cool air for air-cooled machines (standard). That's separate from how you eject the heat from the coolant to the environment. That's either cooling towers (like swamp cooler, uses water), chillers (like A/C, no water but more power), or passive air cooling (like car radiator, no water but only practical if very cold outside).
So you could have a closed-loop water system cooling your machines or chips, but still be consuming water to cool the coolant. And they will advertise this as "closed loop." Better to ask if they have a cooling tower.
Only (mostly) water evaporate, salt and most contamination don't, so you get a brine that you must manage because otherwise it clog your heat exchangers and evaporation towers. Also, it must be returner to a river carefully to not kill all fish and life forms there.
As my friends in Agriculture like to point out, most of the water isn't used at all, it goes right on down the river to the ocean. Ag is second, but less than 50%.
It depends on the river. IIRC the water of the Colorado is stolen 100%, mostly for agriculture. A few years ago I think a small leftover was let, so a tiny part of the river can reach the sea.
One good way to save water is to use treated wastewater for cooling. xAI is building this kind of system in Memphis.[0] It'll connect to a nearby wastewater treatment plant and they'll need to build an additional treatment plant before the water can be used for cooling. It's a closed-loop system inside the data center, where they use clean water, and it connects to open-loop evaporative cooling towers with heat exchangers.
I’ve seen a tremendous amount of content about AI water usage, mostly from pro AI sources. The most common type is comparing AI to particularly water intensive agriculture.
The result is that now I think water usage should be taken into account when siting data centers. Great Lakes and eastern seaboard fine, maybe not as much in California or Arizona.
The bigger concern is more around the pollution of the gas turbines. Populations around the DC are going to see higher rates of Asthma, Respiratory diseases, Heart problems, and certain cancers.
If data center water use is such a concern, why not require that data centers invest in closed-loop cooling systems? By closed-loop, I'm talking about re-condensing evaporated water and allowing the water to cool. Cooling the water would be more expensive in hotter environments, but still achievable. These data centers seem to have wild amounts of money for investment, why not just mandate conservation requirements?
Condensing/cooling the water takes even more electricity though. So you're trading water savings for increased energy use. Maybe OK if it's all renewable, but in most areas it's not.
Ok how about golf courses? This AI is evil look at the water use, is obvious propaganda. It makes no sense to call out data center water use when something that’s purely an optional recreational use consumes 25 times the amount.
I often get side tracked into commenting on regular social media like Instagram and I'm somehow surprised over and over how poor critical thinking skills in the greater population. The zeitgeist of US politics is "this doesn't directly benefit me so this must be bad". According to the Instagram demographic, ALL industrial uses of water and electricity are bad because they "compete" with household use. The massive Agricultural industrial complex is actually OK because I like meat, almonds, etc. AI is bad because it doesn't make my job easier.
Even among the more "globally conscious", there's a severe misunderstanding of how much industry, factories, and overall "consumption" it takes to feed the Western - especially American - way of life. If running data centers can actually sustain the next 10-15 years of ~2% GDP growth, that's literally an economic miracle. An industry that takes in water & electricity yet produces no long term pollutants is literally the closest you can get to money growing on trees.
What other industry in history of the US's economic development has been this clean? I can't think of any. I'm surprised more data centers are not just built in Mexico or other countries that would support rather than oppose/block their development.
I find it deeply ironic that you accuse the public of lacking critical thinking about the externalities of agriculture but claim data centers produce no long term pollutants whatsoever. Demand for compute hardware has skyrocketed, and producing that hardware creates massive pollution from factories and mining. I shouldn’t have to explain how rare earth mining harms millions around the globe. To borrow your expression, you’d be more accurate in saying “this doesn’t directly harm me so it must be good”.
> claim data centers produce no long term pollutants whatsoever
While running. Incurring a pollution penalty once in fungible location (i.e where mines are approved and "hopefully" managed responsibly) is better than incurring pollution proportional to the output (e.g. plastic and chemical waste).
> shouldn’t have to explain how rare earth mining harms millions around the globe.
Is rare earth mining specifically for semiconductor manufacturing actually a significant driver? My intuition is that rare earth and most raw material mining would be driven much more by EV car motors and batteries.
Certainly you can say all energy use is indirectly responsible for the pollution of the oil, solar, wind, etc. I don't disagree at all! I'm say in-addition to the pollution of raw inputs like energy - contemporary industries have additional and unavoidable side products.
> are earth mining harms millions around the globe.
Those mines are going to operate day after day because it's unfortunately the best economic opportunity in those areas. Those areas deserve our support to improve their socioeconomic realities but opposition to data centers in rich countries does not suddenly provide better opportunities to those regions.
1. Tallying the total water consumption impact, embodied water (construction), operational water (cooling), indirect water (electricity generation), supply chain water, etc.
2. Mapping current water intensity onto AI growth forecasts through 2030+
And if you look at those things in combination, there are reasons to be alarmed.
> But AI will bring more important concerns, such as the end of human civilization
Who are these people who think AI will end civilization? Ya'll know it's just autocomplete and deepfakes, right? Maybe they need to read a book about the industrial revolution? It changed the world entirely, but it didn't end it.
> Their water use is mostly for cooling needs from the heat produced from their electricity use.
You should also include the water needed to produce the electricity, which is the biggest water user in the US:
> The three largest water-use categories were irrigation (118 Bgal/day), thermoelectric power (133 Bgal/day), and public supply (39 Bgal/day), cumulatively accounting for 90 percent of the national total.
If AI used as much water as the public "think"(lets say as much as the hysteria suggests the public thinks) then governments would have raised rates on them and they would have reduced usage...
I ran 8 internal audits against my agent stack end-to-end, to figure out if I was destroying the planet. Turns out it uses 12x less energy over a 10minute snapshot when compared to youtube, instagram, facebook ect.
Are you saying any industry that brings in net new jobs with above median wages is bad? Or just ones with few employees and high additional property tax revenue?
I'm concerned with the ones that create temporary jobs, few permanent ones, drive up water and electrical rates and then help deskill other industries.
If we could magically guarantee that our [starry-eyed|gullible|treacherous] political leaders didn't give back most of those property taxes before the DC even broke ground...
What a strange thing to say. This is peak NIMBYism and I urge you to reconsider. Loudoun country as an example generates ~1B USD annually [1] from taxes through data centres. That's equivalent to paying an annual salary of $40,000 to around 30k people. That's a LOT.
Do you really not consider taxes before repeating this tired argument?
Look over here! Not over there at grid infrastructure and generating capacity, or noise and pollution from on-site generators.
The scale of electricity use in data centers is much more likely to cause disruption and the shifting of costs onto residential customers to pay for a new infrastructure and generating capacity.
You can be against lying about water use and for being honest about additional electricity demand at the same time. You can't smear someone for rejecting falsehoods just because you have an unrelated complaint.
I've always found it quite sad and cringeworthy when people talk about AI's water usage. The first thought that comes to my head is whether its even worth trying to talk the person out of their delusion, or just accept that they are lost and can't be helped.
I think people are giving the AI-water-use claims too much credibility. The idea that AI datacenters are heavy water users is trivial to refute, and was trivial to refute when it was first introduced. It should be written about in the same tone as one writes about ridiculous conspiracy theories.
Whether it is or isn't happens to be beside the point. It's water being removed from the system en masse for a non-essential function, i.e. other than sustaining life, while driving up the cost of other utilities.
If we're trying to deny the usage "tier," I'd argue we're being intentionally obtuse at worst and foolish at best.
I appreciate the data driven approach. The article is spot on, it's really hard to distinguish all the discourse with the reality. Things most people grew up with in the 70s had years of propaganda convincing the public they were a net positive to society.
Sidebar, I'm very curious to see where AI goes. Definitely not on the hype train. More curious than anything. This article was a breath of fresh air.
What they don’t mention is that the water is being polluted by the datacenters. It’s not as simple as “water go into datacenter, water come out of datacenter”
Data centers can inadvertently pollute water through chemical runoff from evaporative cooling systems, including biocides, corrosion inhibitors, and heavy metals that accumulate at scale when facilities discharge up to 5 million gallons daily.
“His 68-year-old hardware with 50,000-year-old architecture is enjoying and struggling with the promise, threats, and turbulence of the AI revolution.”
This article conflates agricultural use, which is not treated and is drawn directly from groundwater, rainfall, and rivers, with urban use, which is treated and much more expensive. I find it baffling that the person who put their name on this article would fail to make this critical distinction, given their credentials.
> Jay Lund is an Emeritus Distinguished Professor of Civil and Environmental Engineering and Geography at the University of California – Davis. He is also a Vice Director of the Center for Watershed Sciences
And the main evidence he presents is a summary of a prompt he gave to LLM's? Be serious, please. This is challenging my suspension of disbelief a bit.
So tired of these articles. Yes, it’s possible for them to use very little water. But naive comparisons to non-potable agricultural or other irrigation use or comparisons that don’t take into account growth rates of specific uses or local bottlenecks are useless.
For perspective, 28 million gallons of water per day is roughly equivalent to what 93,000 households consume per day. There are ~130,000,000 households in the United States.
To be fair to people's objections - agriculture is significantly more important than AI model training when it comes to improving the average standard of living - and to be fair to model training a lot of the water usage in agriculture is used on extremely water inefficient crops.
Water usage is, in my opinion, a fair reason to object to AI datacenter placement and growth - but in the arena of public opinion it's more nuanced than some of the other arguments that could be made (noise and power usage being much more suitable ones) but it seems to have struck a cord.
There are absolutely terrible takes on each side of the water argument but this seems to be the one people are focused on so I guess it's up to folks in the know to try and give as much clarity on the topic as possible.
Marginal agricultural water use is alfalfa / nut farming in the desert and ethanol corn, not products consumers actually care about. Consumers aren't clamoring for E15 fuel over E10.
I’m actually surprised it’s so low. That’s about 7 seconds of the Mississippi River at its exit per day. Maybe a week or two of alfalfa farming per year, or even less?
You could imagine running way more water, but I guess these racks are extremely dense.
My conspiracy theory is the whole AI datacenter water consumption outrage is a psyop by state actors to worsen public sentiment around AI, so China and others can catch up. Obviously we should lessen the environmental impact of our technology, while considering it's relative impact vs benefit, especially compared to other technology, in this case in particular to other datacenter usage.
But it's comical to see the average person commenting online, outraged at new datacenters and their water usage (separating this from legitimate zoning issues), when all their posts are in fact being transmitted, stored, and served by relatively similar datacenters.
Is the average person allergic to asking follow-up questions?
> Using the prompt, “How much water is likely to evaporate from data centers in California per year, assuming they are all using mostly evaporative cooling?” several free AI websites provided ranges of estimates, below. These AI also can provide ranges and sources for calculation assumptions.
Data centers with closed loop cooling systems are absolutely built all of the time. Total evaporative cooling has the advantage of being more power efficient (and therefor cheaper) - the only reason they bother with total evap is because the water is being offered plentifully and cheap.
People have no reality of how cheaply priced industrial water is in this country. My parents had a cherry orchard and their annual water bill was $100 an acre per year for as much as they wanted. Which is why the water consumption for data centers is only still a fraction of what we lose to evaporation from inefficient spray irrigation.
Yeah, there are alfalfa fields in central Arizona. Alfalfa basically turns water and sunlight into cellulose about as quickly as plants can.
Worse, the owners of those fields are often foreign companies. That means they use tremendous amounts of water in one of the driest regions on earth, in the middle of a multiple decade drought, and the wealth these farms generate disappears overseas.
Part of the issue is not systematically using a pricing structure that charges disproportionately more for usage above high thresholds.
The 101-level "solution" is to just raise the price to account for demand. The problem with that is that it treats all usage the same, whether it's a residence's first gallon or an alfalfa field's last gallon. But the former is something we need to protect.
It makes sense to price water, and electricity, in a fashion where the first X costs a certain amount, and the next X has a higher rate, and above some percentile of usage it has a much higher rate, and at some percentile of usage, customers should be very nearly paying for new required utility infrastructure themselves. That allows using pricing to solve supply problems, without penalizing normal levels of usage.
Some utilities already do this. But if there are actual issues with having enough supply for both datacenters/farms/smelters/etc and residential usage, then they're not doing this well enough, or don't have the pricing correct.
2 replies →
[delayed]
How is it legal? Shouldn’t water be the most regulated (as in protected) substance of all?
6 replies →
[flagged]
> People have no reality of how cheaply priced industrial water is in this country.
Driving between SF and LA, you see a gazillion signs from water leeches complaining that the government won't give them yet more nearly-free water. No, I don't want to go without fresh crops. But yes, I absolutely believe that, say, growing almonds in basically a desert should be a financially expensive operation, and if that makes the end result more expensive, then so be it. And if that means it's no longer viable to empty rivers for the sake of a tasty bag of snack nuts, I can learn to live with it.
This is a really awkward situation because while we'd really want the market to sort of auto-balance the costs between different suppliers it's also really hard to look at PE ratios right now and believe that the market is anywhere near sane. OpenAI could trivially monopolize the water supply in CA[1] with its current warchest and that would, for everyone, be terrible in some fundamentally obvious ways - so we've clearly got a pretty gigantic misalignment in the market which means we're reliant on the government specifically picking winners but ideally doing so in a sensible manner.
How many well cooked dinners is a prompt worth? Not nearly as many as the market currently says. If it were anything less vital we could probably just ride it out until the bubble bursts but if acceleration continues then in time water usage might actually rise to the levels that the most fear mongering folks are saying it's at.
1. Accidentally even - without even reaching into the realm of malicious intent.
Also just because something is cheap doesn't mean it's not depleting resources and making life worse for a community somewhere. People are constantly trying to build pipelines to the west to deplete the great lakes. There is a societal and ecological limit and these AI companies are not worth it.
[dead]
"closed loop" or "total evaporative" cooling is a myth and a marketing term not a scientific one. Maybe learn about something before prompting AI and spilling out some bs.
https://finance.yahoo.com/news/illusion-waterless-ai-gneuton...
https://www.tiktok.com/@swholli/video/7627247643546193165?_r...
Comparing water usage of AI to agriculture and cities is a little misleading. The cities' water usage is to keep people alive with basically mandatory things, like hygiene, and drinking. Agricultural water usage is required because we have to eat to live. Don't compare something optional to something mandatory.
Instead, compare AI water usage to that of optional things in a city, such as car washes and water parks. Or compare AI water usage to that of what it would take a human to do a comparable task (what does it take to keep a human alive for a few hours compared to running a 15 minute long task to write a report with AI?). While AI water usage might still not look that bad, it would be a more honest comparison.
> have to eat to live
Oh, so that's why we're growing alfalfa in the middle of deserts, flooding the fields with excess water so we can keep water rights, and then shipping the alfalfa to China. It's so we can eat!
If we're shipping the alfalfa to China, I assume that means it's supporting some Chinese person's food source, whether they are directly eating the alfalfa, or some animal is eating it that later becomes food.
If someone is flooding a field unproductively just to use up their quota of water, that is a bad thing that should be addressed. But even if you excluded that unproductive usage and compared AI water use to legitimate agriculture use, that would still be an unfair comparison. If you were to compare AI water use to the amount of water that people are wasting just for legal reasons, then I honestly think that would be a pretty apt comparison.
5 replies →
Yes, and now please cut the non-essential philosophical discussion, the server hosting this site doesn't run on thought experiments alone either.
This comment could have been someone's hamburger!
3 replies →
Don't be disingenuous. They already were dividing things out by type of usage, like talking about water park usage vs. the usage of an entire city for all purposes. They are already admitting that "water usage of a city" isn't only about quenching thirst and maintaining hygiene, it's not a stretch to assume that they also realize that they can be water wastage in agriculture as well. They can't split out every instance of wastage that could be eliminated, and it's ridiculous to expect them to.
My wife works with farmers professionally as part of a conservation district and just responded "THIS PERSON KNOWS FARMING" when showing her the discussion. I genuinely have no idea what you guys are talking about but she immediately got heated.
Based in Colorado.
There was massive controversy about that so I don't know how good counterexample it's that. Unless the argument is "we already waste a lot why would you care about wasting more??" Which is not a great argument.
14 replies →
A pretty easy 'optional' comparison would be golf course watering. I saw a much more detailed write up on this that I can't find now, but a quick google shows 500 billion gallons a year for US golf courses and 180 billion gallons a year for all data centers, not just AI data centers.
> The cities' water usage is to keep people alive with basically mandatory things, like hygiene, and drinking
Almost half of city water usage is for residential landscape irrigation, mostly spraying lawns, which is not exactly mandatory or a basic necessity. Landscape irrigation uses about 3.5 million acre-feet / year, which is 1 to 2 order of magnitude higher than the estimated AI data center usage.
In the article it lists a data point that beer production in Arizona used more water than the data centers in Arizona. People may vehemently disagree, but we absolutely do not need beer. Would I trade beer for AI? That's an easy choice, AI every time. If you just keep track of the water to keep a person alive and the bare minimum water required for agriculture (which isn't particularly efficient in most cases), it would be a fraction of a fraction of what we use now.
Comparing data centers to the bare minimum isn't particularly interesting, the point being made by the article is that we aren't efficient with our water usage in general, AI is a rather small source of waste in the scheme of things.
This is a hilariously misleading "study" and I would bet "beer" wasn't chosen arbitrarily for comparison:
The important difference WRT beer is that the water used in the process likely in a larger part goes towards... the beer itself. This in turn is going into the person who drinks it. So, the water here is actually hydrating human beings.
This can be argued as one of the 2-3 absolutely necessary uses of water. Hydrating people.
So, spending less than the beer industry is not that great of an achievement.
However, a casual reader may see comparison to "beer" and think "oh yeah, beer, just a random thing out of a million, so yeah AI is totally ordinary".
Which is a very incorrect conclusion to reach.
Beer has been around for like a thousand years and we haven't decided to get rid of it. We're five years into this fever dream and everyone either literally hates AI or has been driven at least a little crazy by it. It's a pretty darn easy choice for me (and most people I imagine).
4 replies →
This is an extremely frustrating angle to take because what you're implying is that anytime anyone comes up with any system that takes water they should go in front of a panel of experts (seniors) who get to decide whether their water usage is for an "approvable" purpose. Now I don't like water going to Golf courses either but to me even the intermediate solution is to price water accurately.
Barring that, long term we're surrounded by 70% body of water with infinite energy beaming down on us, this feels like a solvable problem without having large swaths of the country fight over scraps.
> anytime anyone comes up with any system that takes water they should go in front of a panel of experts (seniors) who get to decide whether their water usage is for an "approvable" purpose.
This is absolutely how things work, the water for farming and industry is cheap by design (at least in the US) so that people will have relatively cheap food and consumer goods.
Now you can absolutely try to go change that to a strictly capitalist "One gallon of water is 1 cent, whatever the usage", but you'll have a hard time finding a political group in this country that stands behind such a principal. Even the most conservative groups typically back farming subsidies.
1 reply →
let's have the tool we created as a society called "the government" regulate it instead of waiting for "the market" to price things accurately.
Because let's be real golf courses will pay higher prices and poor people will suffer the burden if we wait for your idea to magically happen
1 reply →
We absolutely do not need to waste as much water as we do on agriculture. Their is more efficient watering systems, crops that do not feed humans, and inefficient crops that aren't needed. Any one of those improvements would dwarf the water usage by AI.
Heck, a better solution yet would be to charge these AI/datacenter companies enough to cover the costs for watering efficiency systems to cover their usage and then some. It's a fraction of their costs, and way better than being anti-growth.
Yeah people aren't mad about datacenters because they are "anti growth"
They don't want to see their local resources depleted and, no, this isn't some fantasyland where corporations will do anything "for the greater good" that isn't in line with their pockets.
1 reply →
> Or compare AI water usage to that of what it would take a human to do a comparable task (what does it take to keep a human alive for a few hours compared to running a 15 minute long task to write a report with AI?).
A very strange comparison. It seems to imply that we "need fewer humans" because of AI. It also assumes AI is primarily used to replace useful human work, something I very much doubt.
> The cities' water usage is to keep people alive with basically mandatory things, like hygiene, and drinking. Agricultural water usage is required because we have to eat to live.
Drinking water is barely a rounding error in cities' water usage.
Agricultural water usage doesn't go to the necessities to feed people. It goes to whatever is most profitable, even if that means growing water intense crops and exporting the produce overseas.
> we have to eat to live
You don't have to eat a burger.
Skip one McDonald's trip per year and you're going to offset all your prompting water waste (see other comments in the thread).
It seems strange to draw the line at car washes.
But why stop there, and why exclude all food equally? Does somebody living a vegan lifestyle (which typically needs vastly less resources, including water, per calorie of food produced) get to wash their car in exchange for their trouble? What if I take a cold instead of a hot shower; do I then get to wash my bike every once in a while?
Basically any discussion of water allocation is stupid. We already have a way to allocate water (or, really, any scarce resource) - markets. Instead of arguing over whether or not a hamburger is worth a car wash's worth of water, bill the person using the water for that amount of water. Let the water user and the price discovery mechanism fight it out. If it is not worth it to them, then they can move to somewhere where water is cheaper.
We don't do this, at least not in the western half of the US. Instead, the biggest consumers of water have "water rights" - the right to use a certain amount of water every year, for free, simply for owning a particular piece of land. And these water rights were all staked out based on estimates of the Colorado River that were wildly optimistic, so there's a century-long waiting list of claims that will permanently supercede your own if you fail[0] to actually consume the water you are entitled to.
This is insane, and it leads to some pretty insane incentives. Because agriculture was here first, it has the strongest claims to water, and a pretty heavy incentive to waste as much water as they are legally allowed to. A lot of the discussions surrounding water usage assume that because agriculture is necessary for human survival, that the water it uses is also necessary. It's not - and the only way to get an industrial water user to actually care about their water usage is to actually bill them for it.
Once we have an actual market for water (not just water "rights"), then we can start talking about what usages are actually necessary - i.e. what uses should we explicitly subsidize through taxes rather than implicitly subsidize through a terribly designed system.
[0] In the interest of fairness, I ran this comment through Google's chatbot, which would like you to know that TECHNICALLY SPEAKING, it takes ten years of intentional disuse to lose a water claim, and that there is a market for water rights. My counterargument is that most farmers do not care about how much water they can not use, and that a market for water rights is not the same as a market for water, because farmers can still decide to just use the water for free. The pricing mechanism cannot work if there are a class of protected users who do not feel backpressure from the pricing mechanism.
A lot of agricultural water usage (more water than AI) is for growing corn to turn into ethanol so we can add it to gasoline. It's not a small amount either, 40% of all corn in the US is used for this purpose.
We use about two orders of magnitude more water (each!) on corn and alfalfa than on data centers as of 2023, and while we're ramping data centers up fast, it'll still be an order of magnitude at the 2030 data center estimates (which may heavily overestimate, now that there's so much opposition popping up).
What about golf courses which use up 476 Billion of water every year? Way more than data centers. People complain about Nestle using water in californa for bottled water but it doesn't compare to what single golf course uses in a year.
yes I think these datacenters AND golf courses are a waste. crazy
There's not really any NEED to grow almonds. Most agriculture in California is not required to sustain life in CA. However, without AI people wouldn't have jobs that could afford CA rents, so AI is required so people can live. Lets get rid of unnecessary uses like agriculture, unless farmers can justify that the usage is actually required to sustain life.
If you look at water distribution you'll find that its unevenly distributed so farmers should pay a water tax and distribute that water to the less water fortunate. CA has an extremely high water GINI with a few farmers consuming far more than their fair share.
This is even more misleading. You have to eat to live, but absolutely not all water usage for food is mandatory.
If you gave me a budget of how much water I could "use" water every year, and I was close to going over, I could easily pay for my annual AI use just by changing what I eat for lunch on a day or two. I could pay for years of AI use just by forgoing buying a new pair of jeans.
The water argument has always felt so intellectually dishonest to me because it's never approached from the perspective of "hey, we're using too much water, how can we conserve it?" If we approached it from that perspective, reducing AI usage would not even crack the the top 100 list of things we would do. But that's not the goal of the water argument, because it quite obviously actually has nothing to do with water.
This is the response to have in mind when confronted with AI-water arguments. It's not about HOW the water is used, it's that, if you're truly concerned about water usage, AI is a non-factor compared to basically everything else you do on a daily basis.
My understanding is that data centers (at least in LA) are using mostly grey/industrial water, not water you can consume or use for agriculture. It feels like we're measuring water as one entity when not all water is equally useful to a human.
Yeah, but data centers allow for jobs which gives people money to buy food.
one of the biggest health problems in US is obesity. 30 to 40% of the food produced in US goes to waste.
Just these two facts will tell you that while, yes, we do need food to live, but on another hand we have an abundance of food and if AI data centers use 0.05% of the water used for humans.
It's a strawman.
Yes and no. We shouldn't compare datacenter water usage to residential water usage. We should compare it to industrial water usage, as that is what it is. The question like "how does datacenter water cooling compares to concrete factory water cooling?" makes some sense from engineering perspective, as you are comparing oranges to oranges to a degree.
Residential water usage is way too different in way too many ways to be meaningfully compared to industrial usage. The scale is different, the waste water treatment is different, the infrastructure cost is different. The water quality standards are different...
Loads of agricultural water usage in the western states is on totally optional stuff like beef and almonds
And corn for putting in gas tanks
Where's the beef?!
Agricultural water usage distribution prioritizes luxury consumption and drought areas are subsidized
Rice is not a luxury for most people. It’s a staple. It uses ca. 40% of all irrigation water globally. Also cotton is not a luxury, though it also uses quite a bit of irrigation water.
2 replies →
[flagged]
While a couple months back an article[1] discussed how Google was keeping the water requirements a secret from locals who wanted transparency, claiming it was proprietary knowledge.
So they sued and discovered it will use 2-8 million gallons of drinking water per day[2], seemingly near the limit of their capacity to handle, judging by comments from officials.
> 'That water supply that otherwise would not be required until 2060 or the 2060s, suddenly becomes something that we need to be worried about during the 2030s.’
> If it exceeds that demand, they’re going to have to start looking for a new water source.
So I'm not sure how this fits with the claims of the article from the OP. I suppose if anything it disproportionately affects certain places not as well equipped for it?
[1] https://pivot-to-ai.com/2026/03/06/how-much-water-do-the-dat...
[2] https://www.wsls.com/news/local/2026/02/26/google-data-cente...
You can go millions of prompts before you use up as much water as it took to make a single beef burger.
You can go tens of thousands of prompts to match the C02 emissions.
There are many legitimate concerns around AI. Water use/CO2 emissions isn’t currently one of them. Going vegan will make up your AI water consumption/CO2 Emissions many thousands of times over.
Depends on the prompt. Do a video prompt and one 30 second video will use as much electricity as running your microwave on high for 15 minutes.
I doubt it, but sure let's assume that's true.
Now measure the amount of electricity the same prompt will use in 6 years when both algorithmic efficiency and 3-4 generations of silicon lower that by 95% (or more).
Will your microwave become 95% more efficient over the next 6 years? No.
Also how many video prompts will the average person run in a given year? Almost certainly 0. I heavily use AI daily and have probably played with AI video less than 4 times, ever.
Yet certainly the average person will use 20,000-100,000 microwave minutes over their lifetime. I use my microwave for 2-3 minutes every day at lunch for example.
From first principles, the idea that electricity use = bad is wrong. If your electricity comes from burning coal or lignite, then obviously yes using that electricity has bad externalities.
But a french person running their microwave on Nuclear powered grids? This is good. Dirty energy sources is the problem. Not energy use itself.
1 reply →
Water I agree. C02 (which is really a tangential metric if energy consumption which will vary by energy mix) I'd want some citations.
Also agree there are other ways we should pursue in parallel regarding emissions.
Obviously, there's different options and variables and bla bla bla, but considering how consolidated and highly industrialized and standardized meat production is, this data is very likely close enough to true for the wast majority of beef burgers eaten by the people complaining about AI resource consumption: https://ourworldindata.org/environmental-impacts-of-food
3 replies →
[dead]
Source? Meat can be "produced" in a location where water is not as scarce. Rural areas. Datacenters "like" to grow in urban areas.
This source says that a 100 prompt spends half a liter of water https://www.eesi.org/articles/view/data-centers-and-water-co...
I remember this year google reported one google search spend a drop of water (or 5 drops, around that)
Your source cites https://www.washingtonpost.com/technology/2024/09/18/energy-... which in turn claims to be based on https://arxiv.org/abs/2304.03271 but uses 0.14 kWh as the energy consumption for a 100-token request to GPT-4, which is an order of magnitude larger than any figure in that paper. Based on a speed of 18 tokens/s https://openrouter.ai/openai/gpt-4/performance the implied power draw is ≈91 kW, about two thirds of a 72-GPU rack https://www.supermicro.com/datasheet/datasheet_SuperCluster_... I somewhat doubt that the model is large enough to require an entire rack's worth of GPU memory, but even if that were the case, a single request is going to get batched with hundreds or thousands of others at the same time, so the true energy consumption should be much smaller than that.
This beef industry organization cites 3 studies: https://www.beefresearch.org/resources/beef-sustainability/f...
> U.S. specific estimates put beef water use at 317, 441 and 808 gallons per pound of boneless beef when precipitation water is not accounted for in calculations.
So, let's just say around 400 gallons of water per pound of beef if you don't include rainfall use.
2 replies →
Let's look at Almonds which _could be_ produced where water is not scarce, maybe, but instead is grown in central valley CA based on quickly depleting ground water.
Each almond takes about a gallon of water to grow: https://www.almonds.org/why-almonds/growing-good/water-wise
A "drop" is not well defined, but some math says there's about 75,000 drops in a gallon: https://www.quora.com/How-many-drops-of-water-can-fit-into-o...
Let's be generous and say that the Almond farmers hit all of their future efficiency goals, so each almond only takes .5 gallons, and that the drop/gallon math is off by a factor of 2.
That means eating 1 almond is about ~4,000 google searches.
https://ourworldindata.org/environmental-impacts-of-food
2 replies →
I have a few cows and rarely ever give them water. In the winter they get enough from snow and when it’s rainy we have a small pond that forms with a stream. They also prefer either of those to drinking well water from a cattle waterer. They are grass fed and rarely get fed stuff like corn.
For for thousands of pounds of beef, I’ve barely used any water at all. Don’t notice the extra consumption on my well at all, and I have a very low producing spring fed well (1 gallon per minute).
“Vegan” crops on the other hand line corn which are irrigated in many parts of the country use a great deal of water and often very inefficiently so.
Well, I've got a small server rack and roof top solar, therefore data centers don't actually use water.
In other words, bringing up some anecdotal, hyper specific (how many meat eaters just "have a few cows"?) information says absolutely nothing about the truth of the matter, but a lot about what you believe constitutes an argument.
1 reply →
> So much of our public discourse on water and other subjects is choked by chatter, untamed by reasoned evidence, data, and quantification. Today, with AI, we have little excuse for not attempting and using honest estimates to inform our discussions and tame our fears and hopes.
Are these things usually convincing? The general pattern is that people take a position on something and then find one paper with a DOI identifier that backs the position. The Elephant and The Rider and so on. Trying to provide someone with evidence of the falsehood of their claims rarely makes them reconsider and often makes them dig their heels in while they search for a new paper with a DOI identifier.
We're in an unprecedented time in the information age when people can rapidly achieve basic competency at many things using Wikipedia, Google, and LLMs critically. If information availability and search were the constraint, one would expect us to reach greater alignment with facts.
The reality in the way information is used, I believe, is the opposite from what we think of. We believe that if there is sufficient information, we can use it to form an accurate model of reality. In practice, I think the social psychologists are right. For the most part, we form the model of reality and then we seek information that supports it. So if you increase the total amount of information what you do is increase the ability for someone to select out that which supports their model.
That's not to say I don't appreciate these things. It's just that I don't think facts move public opinion very much.
The interesting thing that more information and better search provides is that it accelerates the divide between truth-seekers (who will attempt to find primarily falsifying information) and confirmation-seekers (who will attempt to find primarily supporting information). In general, one can imagine that the former will be more successful at modeling the world ex humanity at least. But if others believe something is true, often a direct approach at their facts is not the best approach to get the outcome.
> The reality in the way information is used, I believe, is the opposite from what we think of. We believe that if there is sufficient information, we can use it to form an accurate model of reality.
You should read Yuval Noah Harari's Nexus. He calls this "the naive view of information," which is ignorant of the existence of what he astutely identifies as "intersubjective" realities (see also Angela Cooper-White's entry on "intersubjectivism" in The Encyclopedia of Psychology and Religion):
This is the endgame of postmodernist and constructivist thinking that exalts narrative and story as the ground source of truth. In some ways what we are seeing is a return to religious and superstitious thinking where sufficient belief in a dogma or a pantheon is enough to reify those narratives into consensus reality.
Historically Jungian psychology and indeed religion (a form of proto-psychology, from which Jung inherits by way of the alchemical tradition; see Jung's Psychology and Alchemy) was humanity's collective storehouse of wisdom and techniques for managing intersubjective realities and group "information hygiene." Such techniques are now being lost to antiquity with the late 20th and 21st century focus on only objectively verifiable, quantitative measurements (as opposed to the private subjective, qualitative phenomena experienced as the inner ruminations, contemplations, and dream life of the individual).
https://vimeo.com/387207936
This is a bit of a dead horse, but the magnitude of how off the public is on this continues to amaze me. Pete Buttigieg did a Tulsa town hall a week or so ago where someone cited it taking "10,000 gallons of water just to generate one photo".[0]
[0]: https://www.youtube.com/watch?v=MCc-ipWVShY&t=1h5m43s
Honestly, it's weird to me how fixated both sides are on water.
People against data centers overestimate water usage, but people who think we should build as many as we can, as fast as we can seem the think that "actually they don't use that much water" somehow negates the more real issues with them.
Water is pretty scarce in some of the places they want to build these things. I know people in West Texas that own ranches that have been approached by the datacenter people and it’s basically a desert, oil industry consumes a lot of their water, and the public water they get in the city smells toxic, the well water is flammable. So water use is concerning and I don’t think there’s any reliable or trustworthy source for them to use as a gauge for what to expect so they have to ask.
What are the real issues with them in your estimation? Anti-data-center people bring up water use as a reason why the government should legally prevent data centers from being built, and pro-data-center people bring up water use to argue against the anti-data-center position. I agree that the anti-data-center people are overestimating water usage, as well as the degree to which the amount of water data centers do use is a problem; and that they're doing so because they have some other objection to data centers that doesn't sound as convincing. It would be better to talk about those issues.
The other part of this problem is the idea that if you disagree with someone about the facts you're interpreted as disagreeing with them about the thing they're mad about: You disagree that AI somehow destroys fifty billion-trillion gallons of pure water every time someone asks Claude something, therefore you're fully in favor of Grok making nudes of underage girls.
Some people get an Angry. They love their Angry, and nobody will take it from them.
That's populism for ya, and it's sadly extremely effective.
Meanwhile, both China and India are giving free electricity, providing dollar-for-dollar capex subsidizes, and 25 year tax exemptions to build data centers [0][1].
Love how HN wants to strangle the infrastructure that underlies our entire industry and why HNers get paid. It really highlights how much of the opposition to AI comes from the "chattering classes" and other white collar types as is constantly seen in polling [2][3].
It's funny seeing people who are also part of my party but told coal miners and autoworkers to "learn to code", treated blue collar workers derisively, and ignored concerns by employees in manufacturing and skilled trades which led them to shift to the right now act the exact same way.
Edit: can't reply
> AI Datacenters are not how all or probably even most HNers get paid
Most data centers colo multiple types of compute, not just those dedicated to inference or model training. Additonally, strangling the economics of the infrastructure layer makes entire ecosystems move abroad.
You saw similar opposition to semiconductors fabs back in the early 2010s in the US, and the entire ecosystem virtually out within a decade until the CHIPS act was signed and executed on.
Same with nuclear power in Germany and GreenTech in much of the America.
[0] - https://www.reuters.com/world/asia-pacific/china-offers-tech...
[1] - https://www.reuters.com/world/india/india-gives-20-year-tax-...
[2] - https://www.cnbc.com/amp/2026/02/25/top-earners-are-more-afr...
[3] - https://www.pewresearch.org/social-trends/2023/07/26/which-u...
Populism is effective because the proverbial rising tide is not lifting all boats. AI is currently driving yet another extreme wealth inequality inflection point. Founded just five years ago, Anthropic is going to be a trillion dollar private company maybe this year! This is a staggering outcome and will further divide the gap between the wealthy and everyone else.
So whether populist outrage is expressed through fears of job losses, higher energy prices or concerns over water usage, IMHO, wealth inequality is the cause.
12 replies →
AI Datacenters are not how all or probably even most HNers get paid...
> Most data centers colo multiple types of compute, not just those dedicated to inference or model training. Additonally, strangling the economics of the infrastructure layer makes entire ecosystems move abroad.
Sure but we are talking about whether the enormous investment into AI infrastructure is prudent or not. Also I reckon most people on here made a living just fine before everything moved to remote data centers, and many if not most HNers workloads could run on individual machines... But that's another conversation.
I suspect soon young learners of the future may tilt their heads in curiosity when finding that Obama was a "Democrat" in the same way they did in the past when finding that Lincoln was a "Republican".
you're arguing against things that have no material effect. "oh won't you think about adversarial discourse about the most well funded industry in recent history"
Usually when people compare data center water usage to golf course water usage I feel a lot better about the whole thing.
Compare it to alfalfa and you’ll be laughing your ass off at how much water alfalfa consumes.
~340 acres of alfalfa in California growing year round uses as much water as Google’s data center in The Dalles uses in one year.
That data center used 550M gallons for evaporative cooling in 2025, which is 1687 acre-feet of water.
One acre of alfalfa in California uses ~5 acre-feet of water per acre of alfalfa per year. There are around a million acres of alfalfa grown in California, or 5 million acre-feet of water per year on alfalfa. Which is used to feed cows.
Feed cows in places without the water and sun to grow this stuff locally. Which is tantamount to exporting water from the American West which will eventually be turned into a desert. We effectively can't be trusted to govern our natural resources more than 5 years out.
California’s alfalfa is primarily for export.
I’d recommend you read the following report: Proximity to Golf Courses and Risk of Parkinson Disease
Individuals living within water service areas with a golf course had nearly double the odds of PD compared with individuals in water service areas without golf courses…
https://jamanetwork.com/journals/jamanetworkopen/fullarticle...
Did you read the paper carefully? It's about pesticide use. (It's not especially plausible as epidemiological studies go, though I'm unsurprised if a better study finds a firmer correlation between pesticides and PD.)
3 replies →
I don’t really get the water concerns in datacenter cooling. Even if a lot of water was used for cooling with every prompt (which he argues against here, but, even if)… water “used up” by cooling just comes out a little hotter, right? Maybe evaporated. Then it’ll come back in the form of rain. This isn’t an industrial chemistry process that leaves some toxic waste in the water. Or an agricultural one that puts water in plants and then ships it off to some other region. It just becomes another path through the water cycle.
I actually don’t get how this can be a real thing that people are worried about. Is there some astroturfing behind this? Maybe an attempt to make environmentalists and AI skeptics look stupid?
The absolute strongest complaint is that DCs consume treated, potable water, which is less abundant / easily re-created than any old non-potable source. (Of course the easy solution here is DCs just ingest / treat their own non-potable source. Or utilities charge rates sufficient to price in the externality of drawing down more potable water. The economics still work for DCs if they need to treat their own water -- the fundamental problem is that utilities are underpricing their potable water, so DCs prefer it all else being equal.)
Why don’t data centers use gray water more often? Wouldn’t that be better for basically everyone?
My guess is it’s some combination of the infrastructure not existing, the distribution being bad, and the treatment costs not penciling out.
But that feels like the kind of thing municipal utilities could solve with pricing. Potable water should probably be priced differently for residential use than for big commercial/industrial users, in a way that pushes them toward non-potable sources wherever possible.
A fun Texas water fact I always bring up: the entire state’s monthly freshwater use is roughly a week of freshwater inflow into the Chesapeake Bay. Texas would be the 8th-largest GDP in the world if it were a country, and its whole monthly freshwater demand is basically a few months of water that the Chesapeake just dumps into the ocean. (Of course, estuaries make use of the water so it's not just wasted but it's illustrative imo)
Another fun comparison point is yearly Texas uses 0.08% the volume of the Great Lakes in freshwater but ~ 30-50% of the volume of all the lakes in Texas.
We've got a lot of water but it's not distributed evenly and we should probably build some sort of water pipeline eventually so water rich states can sell to water poor states.
Again, this is all just speculation by someone who knows not a damn thing about municipal water management.
15 replies →
Because they're taking water from already parched regions, often pumping it out of the ground. Even if the water did come back locally as rain (it doesn't), it still makes it impossible for people to live off the same aquifers and water sources sustainably.
People are losing their minds in Wisconsin saying proposed data centers will drain lake michigan. I'm not kidding.
1 reply →
Just 30 mins from where I live data centers are having an impact on water used for farming.
https://www.theguardian.com/global-development/2024/sep/25/m...
https://www.bbc.com/news/articles/cx2ngz7ep1eo
If only we could do water-intensive activities in areas where water is abundant and then ship the resulting products to where they are needed...
In a far science fiction future, I could e.g. imagine connecting LLM inference data centers to a global data network instead of always having to drive up to them to ask my prompts.
It doesn't come out a little hotter, it gets evaporated in cooling towers. Same result as any other water usage. Cooling towers can't use seawater either. Most datacenters are in places where fresh water is abundant anyway, but some are not.
Anyway agricultural water usage is way worse in California.
Agriculture is used to grow _food_.
The water isn’t gone but if it comes back as rain, it at least has to be cleaned again, since data centers probably don’t use raw rainwater for cooling.
It’s probably still not too bad but there’s at least some work done that’s „used up“ by letting tap water (or probably demineralized water used for cooling) evaporate.
The problem is that data centers use SO MUCH water... sure we humans let water evaporate, but this is a new source of water "waste" to the tune of nearing 2 billion gallons/year, just in Loudon County Virginia & connected water users [0].
When that water source is underground wells, this can take years (on the fast end) or decades (on the moderate end) to get back down. Look at California's water issue -- so many wells extracting water for farming has changed the land topography.
Also, when water 'comes back', it might come back in the ocean and not on land... reducing the available fresh water without desalination.
Data centers need the water to cool... but maybe there's room to find incentives for them to do so while making sure our water bills don't go up like our electric bills are because of the extra load they are putting on utilities.
[0]: https://www.theregister.com/2024/08/19/virginia_datacenter_w...
The rain doesn’t happen directly above where it evaporates. And “slightly warmer” waste water can have major ecological impacts, destroying native life in the lakes and rivers where the wastewater is ejected. Plus, if the water is taken away from underground aquifers that may not be refilling fast enough, or if it’s taking water from downstream users, that’s something to be concerned with.
I have also wondered this and came to a similar conclusion about the politics.
This whole time I've been wondering how it's possible that people don't realize how common evaporative cooling is for much larger buildings that are far more numerous than these data centers, and especially in dry climates where drought is common.
> Or an agricultural one that puts water in plants and then ships it off to some other region
Just like an agriculture, data center puts water to cool chips and ships token to some other reason?
I honestly don't know if you are an AI atroturfing bot. No, I am not being sarcastic. Given this is the top comment and there is no reply, here you go
For a pre-chewed eli5 overview, check this: https://www.eesi.org/articles/view/data-centers-and-water-co...
A responsible human must always verify information. I DW as "secondary l" information source. For instance https://www.dw.com/en/why-does-ai-need-so-much-energy/video-...
tldr: chip immersion uses less water but is more expensive. Water evaporation is the opposite. Datacenters will use the cheapest they can get away with. Water is scarse; evaporated water is as unavailable as contaminated water. Read the information sources.
The explanation about chip immersion is wrong though. It's not a water-saving technique, it's for cooling dense racks more rapidly and maybe saving energy. That warm coolant still needs to be cooled, likely the same way, evaporative cooling tower.
Worse, the article says "Since the technology uses synthetic fluids, it requires significantly less water than other approaches." This is like saying that a new coolant-based car radiator doesn't use water while a very old one does, like yeah technically it uses it but doesn't spend it.
I’m not a bot, but maybe I was too quick to not inspect my gut response. I guess I’ll look into it more, maybe this can be a learning experience.
FWIW the comment is just at +2 at the moment, I think it is just at the top of the thread because it is recent and has discussion.
> tldr: chip immersion uses less water but is more expensive. Water evaporation is the opposite. Datacenters will use the cheapest they can get away wi
This suggests a simple fix: charge more to the datacenters (not people) for the water, to make the other option competitive.
No need to throw baby with the ... erm, bathwater.
By that argument water use is never a bad thing since all water comes back as rain. The problem is that data centers need to use clean water, which has to be treated. On a local scale, a large data center could starve a community of potable water, even if the state-wide water use is very small.
This image really helped me put it into perspective. https://x.com/AndyMasley/status/2032858292184117748
Is this what passes for a citation nowadays? I’m sympathetic to the message but this ain’t it.
The data coming from the University of Calgary about the data centres they're building in Alberta, Canada seems to indicate that they're using evaporative cooling, which is very expensive water wise.
The bigger concern though, is the power requirements. Which are set to double or triple the energy use of the entire Province (analogous to a State in the US).
https://ucalgary.ca/sustainability/mobilizing-alberta/climat...
There are data centre projects underway that use their own natural gas generators: https://www.cbc.ca/news/canada/calgary/olds-mihta-askiy-data...
Air pollution, GHG and water use are concerns, but these projects will not dramatically increase the load on the electric grid.
Natural gas is cheap and abundant in Alberta, and the province (actually the whole country, via transfer payments) benefits financially from resource revenues from extracting the gas. So, these projects are generally an easy sell to the public.
I keep hearing about natural gas and on-site power for these data centres. I'll believe it when I see it.
There are already have a couple in Calgary and they're hooked directly to the grid. The cost of electricity for the city shot up at the same time. Also, there have been a few brownouts caused by them not being ready to handle late night draws from those data centres.
That's at least what I'm seeing. Though, admittedly, it's from older project articles. Maybe something has changed in recent months?
https://www.cbc.ca/news/canada/calgary/ai-data-centre-albert...
Did anyone find it weird that the author uses AI itself to perform the calculations? Seems like a very poor quality piece
For a more informed deep dive into data center water usage (albeit many months old): https://www.construction-physics.com/p/i-was-wrong-about-dat...
I stopped reading at the crappy ChatGPT comic that shows "water usage" as some pipes pouring water and others pumping it in. How trustworthy is the text going to be after that?
I don't find it weird, it's about what I'd expect from AI sycophants. They don't seem to realize that it comes off as not even being able to defend their own ideas.
A lot of confusion around AI water usage might stem from whether it's an open-loop or a closed-loop cooling system.
e.g. an open-loop system which disposes of waste heat through evaporation is naturally going to draw a lot more water than a closed-loop system which recycles the water. Open-loop is likely cheaper to build, and importantly, it _does_ use up a lot of water that could otherwise be going to a municipality.
So, what's the actual breakdown between these two? I absolutely _could_ imagine many datacenter operators cheaping out and using open loop cooling, particularly if building next to a source of fresh water like a river.
A lot o the confusion around data centers is that these companies purposely hide this information from the public. We already know how damaging normal data centers are:
https://www.oregonlive.com/silicon-forest/2022/12/the-dalles...
Citizens had to sue their town to force them to give up water usage, something Google was adamant about hiding from the public.
When there is no accountability, trust plummets. There is no reason to trust anything from these corpos or their pro-corpo rags.
A lot of us work for and do business with companies that purposefully hide information from the public.
That doesn't seem to be an unusual state of affairs at all; it instead seems like a very normal way of doing things.
3 replies →
To simplify things, "closed loop" shouldn't even be part of the discussion. Usually they just mean a closed-loop cooling system somewhere inside, either to directly cool machines (typically ML) or to cool air for air-cooled machines (standard). That's separate from how you eject the heat from the coolant to the environment. That's either cooling towers (like swamp cooler, uses water), chillers (like A/C, no water but more power), or passive air cooling (like car radiator, no water but only practical if very cold outside).
So you could have a closed-loop water system cooling your machines or chips, but still be consuming water to cool the coolant. And they will advertise this as "closed loop." Better to ask if they have a cooling tower.
Open loop cooling can work fine if they use greywater. The water isn’t potable anymore, but it goes into the sky and becomes clean again.
It’s all just a lack of imagination.
Only (mostly) water evaporate, salt and most contamination don't, so you get a brine that you must manage because otherwise it clog your heat exchangers and evaporation towers. Also, it must be returner to a river carefully to not kill all fish and life forms there.
Most of the confusion just stems from anti-DC advocates lying about water usage, not any specific technical details.
As my friends in Agriculture like to point out, most of the water isn't used at all, it goes right on down the river to the ocean. Ag is second, but less than 50%.
It depends on the river. IIRC the water of the Colorado is stolen 100%, mostly for agriculture. A few years ago I think a small leftover was let, so a tiny part of the river can reach the sea.
One good way to save water is to use treated wastewater for cooling. xAI is building this kind of system in Memphis.[0] It'll connect to a nearby wastewater treatment plant and they'll need to build an additional treatment plant before the water can be used for cooling. It's a closed-loop system inside the data center, where they use clean water, and it connects to open-loop evaporative cooling towers with heat exchangers.
[0] https://www.datacenterdynamics.com/en/news/elon-musk-xai-mem...
This is an AI generated article, with AI generated images, claiming that AI isn't a resource problem.
I "like" how in their graphic agriculture and cities are both putting water into the lake, and only data centers are removing water from the lake.
The prompter should have redone this image a couple of times until they had all three actually draining the lake.
I’ve seen a tremendous amount of content about AI water usage, mostly from pro AI sources. The most common type is comparing AI to particularly water intensive agriculture.
The result is that now I think water usage should be taken into account when siting data centers. Great Lakes and eastern seaboard fine, maybe not as much in California or Arizona.
The bigger concern is more around the pollution of the gas turbines. Populations around the DC are going to see higher rates of Asthma, Respiratory diseases, Heart problems, and certain cancers.
If data center water use is such a concern, why not require that data centers invest in closed-loop cooling systems? By closed-loop, I'm talking about re-condensing evaporated water and allowing the water to cool. Cooling the water would be more expensive in hotter environments, but still achievable. These data centers seem to have wild amounts of money for investment, why not just mandate conservation requirements?
> These data centers seem to have wild amounts of money for investment, why not just mandate conservation requirements
This IS the complaint.
Data center water use is in fact not a valid concern.
Condensing/cooling the water takes even more electricity though. So you're trading water savings for increased energy use. Maybe OK if it's all renewable, but in most areas it's not.
imo this is a pricing problem more than a cooling-design problem. datacenters get cheap clean water while locals pay for the pipes and grid upgrades.
Regulating AI? America would never!
The tradeoff is power vs water. Water is currently cheaper.
A much more comprehensive article on this subject is here:
https://news.ycombinator.com/item?id=45946966
I was really into Andy's blog when it first came out, but he might be doing the same motivated reasoning he complains about everyone else doing.
https://blog.andymasley.com/p/data-centers-heat-exhaust-is-n...
> The one remaining question: why the clean step change?
In the middle of this piece, he runs into a critical flaw in his reasoning and just shrugs it off.
Ok how about golf courses? This AI is evil look at the water use, is obvious propaganda. It makes no sense to call out data center water use when something that’s purely an optional recreational use consumes 25 times the amount.
I often get side tracked into commenting on regular social media like Instagram and I'm somehow surprised over and over how poor critical thinking skills in the greater population. The zeitgeist of US politics is "this doesn't directly benefit me so this must be bad". According to the Instagram demographic, ALL industrial uses of water and electricity are bad because they "compete" with household use. The massive Agricultural industrial complex is actually OK because I like meat, almonds, etc. AI is bad because it doesn't make my job easier.
Even among the more "globally conscious", there's a severe misunderstanding of how much industry, factories, and overall "consumption" it takes to feed the Western - especially American - way of life. If running data centers can actually sustain the next 10-15 years of ~2% GDP growth, that's literally an economic miracle. An industry that takes in water & electricity yet produces no long term pollutants is literally the closest you can get to money growing on trees.
What other industry in history of the US's economic development has been this clean? I can't think of any. I'm surprised more data centers are not just built in Mexico or other countries that would support rather than oppose/block their development.
I find it deeply ironic that you accuse the public of lacking critical thinking about the externalities of agriculture but claim data centers produce no long term pollutants whatsoever. Demand for compute hardware has skyrocketed, and producing that hardware creates massive pollution from factories and mining. I shouldn’t have to explain how rare earth mining harms millions around the globe. To borrow your expression, you’d be more accurate in saying “this doesn’t directly harm me so it must be good”.
> claim data centers produce no long term pollutants whatsoever While running. Incurring a pollution penalty once in fungible location (i.e where mines are approved and "hopefully" managed responsibly) is better than incurring pollution proportional to the output (e.g. plastic and chemical waste).
> shouldn’t have to explain how rare earth mining harms millions around the globe.
Is rare earth mining specifically for semiconductor manufacturing actually a significant driver? My intuition is that rare earth and most raw material mining would be driven much more by EV car motors and batteries.
Certainly you can say all energy use is indirectly responsible for the pollution of the oil, solar, wind, etc. I don't disagree at all! I'm say in-addition to the pollution of raw inputs like energy - contemporary industries have additional and unavoidable side products.
> are earth mining harms millions around the globe.
Those mines are going to operate day after day because it's unfortunately the best economic opportunity in those areas. Those areas deserve our support to improve their socioeconomic realities but opposition to data centers in rich countries does not suddenly provide better opportunities to those regions.
From what I understand water usage critics are:
1. Tallying the total water consumption impact, embodied water (construction), operational water (cooling), indirect water (electricity generation), supply chain water, etc.
2. Mapping current water intensity onto AI growth forecasts through 2030+
And if you look at those things in combination, there are reasons to be alarmed.
> But AI will bring more important concerns, such as the end of human civilization
Who are these people who think AI will end civilization? Ya'll know it's just autocomplete and deepfakes, right? Maybe they need to read a book about the industrial revolution? It changed the world entirely, but it didn't end it.
> Their water use is mostly for cooling needs from the heat produced from their electricity use.
You should also include the water needed to produce the electricity, which is the biggest water user in the US:
> The three largest water-use categories were irrigation (118 Bgal/day), thermoelectric power (133 Bgal/day), and public supply (39 Bgal/day), cumulatively accounting for 90 percent of the national total.
https://www.usgs.gov/faqs/how-much-water-used-people-united-...
What about all the water used to generate electricity? You know human still boils water for electricity.
If AI used as much water as the public "think"(lets say as much as the hysteria suggests the public thinks) then governments would have raised rates on them and they would have reduced usage...
I ran 8 internal audits against my agent stack end-to-end, to figure out if I was destroying the planet. Turns out it uses 12x less energy over a 10minute snapshot when compared to youtube, instagram, facebook ect.
https://www.linkedin.com/posts/peterjamesmcgrath_i-ran-8-int...
Greater than $0 in cost of living increases for people living near these things is too much.
Are you saying any industry that brings in net new jobs with above median wages is bad? Or just ones with few employees and high additional property tax revenue?
When the new jobs number increases to four (or even three) digits people will take that more seriously.
1 reply →
I'm concerned with the ones that create temporary jobs, few permanent ones, drive up water and electrical rates and then help deskill other industries.
How many jobs are created? And how many jobs were also lost because of AI? A few jobs created vs thousands or more lost isn't a positive.
> that brings in net new jobs
Ah yes, those invaluable tens of jobs created by DCs....
4 replies →
If we could magically guarantee that our [starry-eyed|gullible|treacherous] political leaders didn't give back most of those property taxes before the DC even broke ground...
What a strange thing to say. This is peak NIMBYism and I urge you to reconsider. Loudoun country as an example generates ~1B USD annually [1] from taxes through data centres. That's equivalent to paying an annual salary of $40,000 to around 30k people. That's a LOT.
Do you really not consider taxes before repeating this tired argument?
[1] https://www.loudoun.gov/DocumentCenter/View/219184/General-F...
Look over here! Not over there at grid infrastructure and generating capacity, or noise and pollution from on-site generators.
The scale of electricity use in data centers is much more likely to cause disruption and the shifting of costs onto residential customers to pay for a new infrastructure and generating capacity.
wouldn't it be great if we hadn't actively sabotaged grid capacity and development at every turn
Wouldn't it be great if residential rate payers didn't end up holding the bag for botched nuclear plant construction and cost over runs.
You can be against lying about water use and for being honest about additional electricity demand at the same time. You can't smear someone for rejecting falsehoods just because you have an unrelated complaint.
Most people don't know AI uses water.
The Empire of AI book seriously did permanent damage on this talking point.
Asking chatbots for estimates of water usage and then taking their average is a great way to alienate your audience. It's embarrassing, as well.
What does the public actually think?
i believe it was like amount of water gold uses in the USA alone is 10x more than water used by AI globally
I've always found it quite sad and cringeworthy when people talk about AI's water usage. The first thought that comes to my head is whether its even worth trying to talk the person out of their delusion, or just accept that they are lost and can't be helped.
I think people are giving the AI-water-use claims too much credibility. The idea that AI datacenters are heavy water users is trivial to refute, and was trivial to refute when it was first introduced. It should be written about in the same tone as one writes about ridiculous conspiracy theories.
Whether it is or isn't happens to be beside the point. It's water being removed from the system en masse for a non-essential function, i.e. other than sustaining life, while driving up the cost of other utilities.
If we're trying to deny the usage "tier," I'd argue we're being intentionally obtuse at worst and foolish at best.
I appreciate the data driven approach. The article is spot on, it's really hard to distinguish all the discourse with the reality. Things most people grew up with in the 70s had years of propaganda convincing the public they were a net positive to society.
Sidebar, I'm very curious to see where AI goes. Definitely not on the hype train. More curious than anything. This article was a breath of fresh air.
c/o Jay Lund, Vice Director, Center for Watershed Engineering Distinguished Professor, Civil and Environmental Engineering
What they don’t mention is that the water is being polluted by the datacenters. It’s not as simple as “water go into datacenter, water come out of datacenter”
Data centers can inadvertently pollute water through chemical runoff from evaporative cooling systems, including biocides, corrosion inhibitors, and heavy metals that accumulate at scale when facilities discharge up to 5 million gallons daily.
https://www.datacenterknowledge.com/sustainability/4-strateg...
I really love how he ends his bio:
“His 68-year-old hardware with 50,000-year-old architecture is enjoying and struggling with the promise, threats, and turbulence of the AI revolution.”
This article conflates agricultural use, which is not treated and is drawn directly from groundwater, rainfall, and rivers, with urban use, which is treated and much more expensive. I find it baffling that the person who put their name on this article would fail to make this critical distinction, given their credentials.
> Jay Lund is an Emeritus Distinguished Professor of Civil and Environmental Engineering and Geography at the University of California – Davis. He is also a Vice Director of the Center for Watershed Sciences
And the main evidence he presents is a summary of a prompt he gave to LLM's? Be serious, please. This is challenging my suspension of disbelief a bit.
As a more complete title...
AI uses less water than the public thinks and more water than Anthropic or OpenAI report.
Both sides have dishonest reporting
Fantastic news!
Very insightful bullet points, ordered lists and grok tables! Articles like this are certainly a net benefit to society
This whole meme never made sense. Data centres are cooled with AC. Where the fuck is water supposed to be going?
So tired of these articles. Yes, it’s possible for them to use very little water. But naive comparisons to non-potable agricultural or other irrigation use or comparisons that don’t take into account growth rates of specific uses or local bottlenecks are useless.
[dead]
[dead]
[dead]
[flagged]
[dead]
Does it use more than zero? Then I hate it. Maybe we should try to calculate how much water online advertisements take.
The author uses a measurement I'm not familiar with so I used AI to translate it.
>Using the broader initial AI water use estimate of 32,000 acre-ft/year to 290,000 acre-ft/year
Note : 1 acre-foot is approximately equal to 325,851 gallons.
AI : That estimate converts to approximately 10.4 billion to 94.5 billion gallons per year.
Ya 10 billion gallons of water (low estimate) is totally nothing. Thx for this informative blog post.
28.6 million gallons per day.
Everything is relative, 28.6m gallons per day is nothing.
Golf courses use nearly 100x more water per day than datacenters, nearly 2b gallons per day. [1]
Residential lawn water usage is ~9b gallons per day. [0]
0 - https://19january2017snapshot.epa.gov/www3/watersense/docs/f...
1 - https://www.usga.org/content/dam/usga/pdf/Water%20Resource%2...
For perspective, 28 million gallons of water per day is roughly equivalent to what 93,000 households consume per day. There are ~130,000,000 households in the United States.
To be fair to people's objections - agriculture is significantly more important than AI model training when it comes to improving the average standard of living - and to be fair to model training a lot of the water usage in agriculture is used on extremely water inefficient crops.
Water usage is, in my opinion, a fair reason to object to AI datacenter placement and growth - but in the arena of public opinion it's more nuanced than some of the other arguments that could be made (noise and power usage being much more suitable ones) but it seems to have struck a cord.
There are absolutely terrible takes on each side of the water argument but this seems to be the one people are focused on so I guess it's up to folks in the know to try and give as much clarity on the topic as possible.
Marginal agricultural water use is alfalfa / nut farming in the desert and ethanol corn, not products consumers actually care about. Consumers aren't clamoring for E15 fuel over E10.
1 reply →
I’m actually surprised it’s so low. That’s about 7 seconds of the Mississippi River at its exit per day. Maybe a week or two of alfalfa farming per year, or even less?
You could imagine running way more water, but I guess these racks are extremely dense.
one environmental concern down, hundreds to go! keep up guys!
My conspiracy theory is the whole AI datacenter water consumption outrage is a psyop by state actors to worsen public sentiment around AI, so China and others can catch up. Obviously we should lessen the environmental impact of our technology, while considering it's relative impact vs benefit, especially compared to other technology, in this case in particular to other datacenter usage.
But it's comical to see the average person commenting online, outraged at new datacenters and their water usage (separating this from legitimate zoning issues), when all their posts are in fact being transmitted, stored, and served by relatively similar datacenters.
Is the average person allergic to asking follow-up questions?