Irrespective of how this number is calculated, whether is blue/green water etc here is some comparison data for context, since I personally think that hearing big numbers without any point of reference is pointless.
Annual withdrawals (2015)
- Thermoelectric power: ~48.5 trillion gal
- Irrigation: ~43.1 trillion gal
- Public supply: ~14.2 trillion gal
- Self-supplied industrial: ~5.4 trillion gal
- Aquaculture: ~2.76 trillion gal
- Mining: ~1.46 trillion gal
- Livestock: ~730 billion gal
- Total: ~117.5 trillion gal (322 Bgal/day)
Source: Estimated use of water in the United States in 2015: U.S. Geological Survey Circular - https://doi.org/10.3133/cir1441
This is a good illustration how cursed the datacenter topic has become when people go out of their minds on computer cooling while continuing to eat meat which lets be perfectly clear - is mostly for entertainment purposes. I had someone berate me for defending datacenter water use and their last post was literally a cheese burger.
Well the thing with the livestock is that you have to give a cow water.
Datacenters do not need water. It is just slightly cheaper for them than cooling with electricity.
Datacenters should be required by law to pay to build adequate power generation and infrastructure to pay for their consumption, and ideally in a way where it isn't a ghastly eyesore.
> Datacenters do not need water. It is just slightly cheaper for them than cooling with electricity.
As long as electricity at least partially produced using fossil fuel I would prefer data-centers to use as little electricity as possible. There are regions which are getting plenty of rainfall on average (which can be collected and stored) or were getting some water from rivers will have negligible ecological impact.
But currently DC operators have little incentives to choose DC location based on outdoor temperature or water availability and there are plenty of DCs in hot and arid regions.
Differences in air temperature are largely below the noise floor in terms of impact on data center cost. Slightly lower energy prices and other operational or capital costs are more than enough to offset significantly hotter climate. Arid climates improve evaporative cooling efficiency.
> Datacenters should be required by law to pay to build adequate power generation and infrastructure to pay for their consumption, and ideally in a way where it isn't a ghastly eyesore.
Fact is, almost nobody "needs" meat in their diet, and even if all meat eaters did actually need it rather than merely want it, there are meats which use less than the average quantity of water.
From the above stats, the mean American's loss of water due to unnecessary use of water for unnecessary meat raised in their territory, is about 43 times the mean American's loss of water due to the unnecessary water use of datacenters that are in their territory.
You don't need a law to force people to pay for resources they consume; basic economics already ensures that. If there are externalities not already accounted for by that (increased demand for water/electricity driving up costs for others, noise, "eyesores", etc.) you could pass laws to ensure those costs are paid by the responsible party but that's not a datacenter-specific issue.
Alas, we do, because in the case of water, basic economics are stymied by pre-existing water-rights laws that are batshit insane. As such, we need legislative action to (at minimum) clean up that mess.
How does that work in a market where you are bidding amongst other participants and capacity is constrained? Supply is limited and demand is growing, the market response is to charge more.
From the above stats, a 2.3% change to USA meat consumption, up or down, has the same impact as a 100% change to data centre water consumption, up or down.
If I read the USDA stats correctly, US red meat and poultry production went up about 5.6% from Jan-Jun 2025 to Jan-Jun 2026, i.e. the increased production of meat led to an increase in water use that was twice as much as the water use of all the data centres combined.
Give up your meat consumption because it's more about societal excess than anything else.
Meat is and mostly always has been(in euro societies) a luxury item meant mostly for the rich.
Capitalism turned meat into a MUST HAVE item for keeping up with the Joneses.
We consume absolute insane amounts of meat nowadays, and we waste about half of it, just so people can feel good cuz theyre eating a high class excess meal.
Lol most people throughout history didn't eat much meat. It's wild how modern western people refuse to see how we'll destroy the planet trying to feed 8 billion people with the most inefficient method of creating calories(beef). The inefficiency IS THE POINT.
So? Many people are not willing to give up their cloud services either?
And there are other kinds of meat that cost less water to produce. I don't think someone who isn't willing to give up beef for chicken is at a higher moral ground than someone who wants cheap tokens.
True. I was going to hate the environmental destruction being wrought for the purpose of funneling more of our wealth to a handful billionaires who will use the AI against us, but then I remembered how bad meat eaters are for the plight of the common man, and now data centers don’t seem so bad after all.
There would need to be some articulable upside to data centers, though. Even if we agree there should be less livestock, I don't see how that has any bearing on the data centers topic.
I assume HN runs on machines in a datacenter. The servers that serve my business also reside in a datacenter. I see lots of upsides to datacenters, and they've been around long before they were controversial.
It's to put water usage in perspective. I think the best example is golf courses.
If you think datacenters are a waste of water and you want to do a global campaign to stop building / shutdown golf courses first, then your values don't really make sense.
Golf is pure leisure, and uses 531 gallons compared to 17.4 gallons a year.
Datacenters on the other hand, power the site we're using right now to discuss the "problem". You wouldn't be able to have this conversation without them.
--
Why aren't people screaming constantly about golf courses and pressuring to shut them down?
Why the focus on datacenters if the bigger win is staring you right in the face (if you're that concerned about water usage)?
I would get rid of every single golf course in the world if I could. One of the worst uses of land out there. Imagine the horrible, gross volume of herbicides each course uses to ensure that only the "correct" grass is growing over all those acres.
The datacenter topic is cursed because people are afraid of ai. People seized on water usage because it is a more legible and tractable focus for their fear.
calling it "actual food" is clearly misleading when majority of meat is used purely for entertainment not for survival sustenance. By that logic compute is "actual medicine" because some of compute is used for medicine?
Yes I use LLMs at work because I'm "highly encouraged" to as a condition to keep my job. All I see is more paying customers finding more bugs and having a worse experience using our products.
You don't see the issues with your line of thinking? You get to absolve yourself of the responsibilities of your choice because "someone is making you" use it? How can you say you value almonds more than LLM slop when your individual choices don't reflect those values?
Good context, but aren't many of your examples simply putting that water back into the ground? Also, where the water is taken from is the key issue. If this was ocean water nobody would care, but if it's Reston, VA that's a local issue.
For perspective, the total annual water consumption of all US data centers is about 100 days' water production of the nearby water treatment plant to Reston, VA. Or put differently, all of the US data centers can be comfortably supplied solely with the water from the next phase of expansion of said treatment plant (an upgrade of 75MGD).
"17 billion gallons of water a year" sounds like a lot of water, but it's really not.
I'm not sure what led you to believe that the water is being put back into the ground "simply" or otherwise but maybe you shouldn't be acting like it's a non-issue when you clearly don't understand the problem.
DCs are pumping the water through evaporative cooling towers because the latent heat of vaporizaton means there's a huge amount of heat sucked up in the phase change from liquid to vapor. None of the water is being "put back into the ground" except maybe that used during flushing and other maintenance.
"Dry" cooling exists, where there is a closed coolant or refrigerant loop run through radiators, but those require more power, which increases cost and limits how much power can be used for compute (but they also get the say "it makes us use more power, bad for the environment! We're being Green by using evaporative cooling!")
Nice perspective. So even if all these data centers were built in the desert, it's entirely within the realm of possibility to transport 17B gallons of water, unlike the 43 trillion used for irrigation. Some back-of-the-napkin math:
- A truck holds 5000 gallons
- It would take 3.5 million trips to transport that much
- If each trip was 500 miles roundtrip, that's about 2 billion miles
- Trucks travel 140 billion miles each year in the US. That would add less than 2% to the total miles
- The trucking industry is about $1 trillion in revenue per year. So 2% of that is $20 billion total to transport 17 billion gallons. Or around $1 per gallon.
The problem is that these comparisons do not have equal utility. Trying to compare them to AI datacenter consumption is equivalent to justifying a gambling habit on a fixed income because groceries consume plenty of dollars as well.
> But that's tripled of water use across all sectors due to datacenter buildout. So no "comparison" to other uses is needed.
You're saying almond farm water use tripled due to data center buildout? What other meaning of "all sectors" could make sense of this claim? The report clearly says 17 billion while all other sectors are in the trillions.
You have to look at the local level, water isn’t distributed the same way around a country as large as the US. Some regions have no water problems while others have a lot of consumption and competition to get access to it
I haven't seen a single article doing that though (from either side) and I honestly think that nobody would even trust these companies at this point (Meta et al) to give an honest analysis anyways.
If you need to denigrate pure context as "whataboutism", then maybe your core argument is trash?
I mean, sure, lets not build datacenters that are 100% evaporation-cooled into each desert-oasis.
But if general water availability is your concern, than datacenters are so far down the priority list that any argument involving them is just laughable. And if you are against datacenters, then find a proper argument instead of making environmentalism look stupid.
One gallon of almond milk requires 1,600+ gallons of water (likely much more — given some estimates are each almond requires up to 3 gallons of water). US almonds are almost entirely grown in California and consume 1.5T to 2T gallons of the state’s water.
It’s pretty dumb to compare AI/data center usage to water usage when something that’s actually wasteful such as almond milk exists.
There’s better metrics to make one’s anti-AI point.
I feel like we need some citations for these figures. Our World In Data [1] disagrees. I've not seen any analysis that puts any plant milk at a higher water use figure than dairy.
If you want to play these stupid games how much water did you think was used to make the concrete for the data centers? Or how much water was used to cool the steel? How much water did all the workers drink while building the data center? How much water did they use to build all those GPUs?
Almonds provide a real utility for humans: food. LLM slop has done nothing but made the public miserable to help prop up the delusional fantasies of the tech elite (all of which hold deeply anti-human + anti-democratic views).
It's really weird to me that datacenter fans all have detailed knowledge of almond farming, but never seem to notice that the nutritional value of ChatGPT is not on par with your average box of almond milk.
Because nutritional value is not the only thing of value. We invented this clever way of uniformly valuing things called "money" that abstracts over all of those differences though.
Potable well water should NEVER be used for such purposes. Aquifers recharge at a rate far lower than industry wants to pull from them. At some point potable water for humans will run out.
If you're using water just to evaporate it for cooling then use surface water.
Using water for cooling significantly reduces cooling energy use. The decision of whether to use dry, evaporative, or hybrid cooling approaches should be driven by the specific water availability in the area where the cooling process is to be used.
Main regions do have enough water to support evaporative cooling. When water is available, using it is prudent because it saves so much electricity.
Using surface water is not a practical approach for most cases because the process would need to be built next to a river or lake - there is limited industrial real estate in these areas.
What does it mean to be consumed in this context? After used for cooling isn’t it clean enough for drink and other uses? Can’t it go back into other systems? Can’t it be used for agriculture?
"consumed" means it was used for evaporated cooling and let off into the atmosphere. It's obviously a resource that replenishes, but wells replenish at a limited rate [and can run dry] so it's not a negligible issue.
Feel free to google this yourself, or ironically ask an AI, it'll happily confirm it's a real issue.
That's a small portion of the consumption. According to the article, the vast majority of the water is for electricity generation, and the vast majority of that water is from hydroelectric power generation. Hydroelectric generation does not evaporate the water, it releases it downstream in a clean, usable form.
Can that vapor be captured and condensed again? The energy transfer of evaporation happens before it reaches the cooling tower exit. It's a conservation of energy thing for sure but condensation is doable.
This refers to water that is evaporated for cooling of both the datacenters themselves and the power plants that give them electricity. The remaining water after evaporation is so full of minerals that it has to be discharged as sewage.
This evaporation process is used to cool the closed loop of fluid that actually moves heat in and out of the data halls.
Is anybody working on using all this waste heat for desalination?
I think cold seawater can be used for the external loop, through a heat exchanger where it would meet the heated pure water from the internal cooling loops. This would cause evaporation of the salt water which could be captured.
Waste heat from data centers is significantly below the temperature that desalination systems typically use. While a desalination system could theoretically use data center waste heat, it would require far higher capital costs than other desalination systems.
An acre of alfalfa is typically irrigated at 5.5 AF per year. The standard center pivot field is 130 acres ~ 700 AF of irrigation, to grow alfalfa to ship overseas. Converting to gallons, 210 million gallons, for one field.
80 fields of alfalfa equals all the data center water usage.
USDA estimates that total irrigation uses 55 million AF of water per year.
Support for data centers jumps a net 31 points after voters are told modern facilities can reuse the same cooling water for up to 10 years, new poll finds. -- https://x.com/Polymarket/status/2092209623583404461
Fluid renewal occurs only every 7–15 years and is performed in a fully contained process that meets state and federal environmental standards. This design ensures zero water withdrawal, zero discharge, and continuous operation — fully aligned with TeraWulf’s commitment to sustainable digital infrastructure. -- https://www.lakemarinerdata.com/closed-loop-cooling
Can, but do they? Feels a bit like Musk's Colossus that "could" use nuclear power but has been running on "temporary" natural-gas turbines for years now.
Lots of places in the US have plentiful water. Don't build data centers where there's a lack of water! And it's a dead horse at this point, but we don't need to be growing almonds in CA. If you actually care about water use, focus on that.
Because people are more "against capitalism" than ever, free markets have become an underrated solution to many of our problems. E.g. 1. stop subsidizing wasteful, environmentally damaging agriculture and 2. legalize building housing. Markets aren't perfect but they could improve the status quo in certain areas.
The worst designs evap up to 80%. New designs use closed-loop system where very little evaps (~6%).
Many designs in-between with air cooling, hybrid, etc.
Still a non-issue, it's going back into the water cycle.
Any water issue would be specific to the municipality, it's not a national issue, nor is it an issue specific to datacenters, it's true for anything being built in the town (golf courses, power plants, etc.)
Article title is extremely misleading. These are figures from 2023 - almost three years old now, and since then US datacenter power consumption has doubled.
The article source for the 17B gallons is a Congressional Research Service report and its source is a Lawrence Berkeley National Laboratory report. Both are interesting reads. The 17B number is an estimate. Here are some of the inputs:
* reported numbers of server deliveries
* power draw per server
* average utilization based on various workloads, including a Llama 70B 8 node training run
* mix of data center cooling types
* do math to get to 66B liters, which rounds down to 17B galoons
It's weird how all these statistics could end if the owners of these data centers would just simply release the numbers for the public to learn from. Instead they are keeping these figures private and we have people trying to best-estimate the reality. Mostly for reasons we all understand: true figures would cause an even larger public backlash.
Fun fact, citizens of The Dalles, OR had to sue their own government to get them to release numbers on how much Google data centers were consuming (info Google stipulated should be kept "proprietary"):
This is a manufactured propaganda issue by dishonest and/or innumerate people.
However, can’t data centers just use closed-loop cooling and avoid this whole thing? With all the money going into AI, it’s hard to imagine it being prohibitively costly.
With regards to evaporative cooled DCs (either primarily cooled that way or assisted some of the time): Imagine an air conditioner that instead of requiring energy to drive the compressor to make the phase change work, all the phase change was provided by taking in liquid water and evaporating it off as vapor, taking the latent heat of vaporization (2257 kJ/kg or 970 BTU/lb) with it.
I don't even know who owns forgeeks.net. They're only a couple months old and is disseminated by a new HN account. Some of the "author" social media links appear AI generated.
Red flags galore. Wouldn't surprise me if it was bankrolled by bad actors and/or a website created/operated by AI agents.
People have always used a closed loop to move heat in and out of the data hall - but something has to keep that closed loop cool. That's often a cooling tower which uses evaporating water to reject heat into the air in a way that's very energy efficient.
Additionally, the power plants which supply electricity to the data centers have the same cooling problem. They also typically use cooling towers.
It's not dishonest at all. Datacenters and fossil fuel power plants are absolutely evaporating fresh water for cooling. The piece that's missing from that discussion is that it helps them use less electricity.
They use closed loop systems. That reduces the water consumption on-site, but you still need to move the energy out of the DC, which requires a lot of energy. That extra energy itself is likely produced by upstream power plants evaporating water. In practice you mostly move the consumption upstream.
The article links to a report from the library of congress, from which it gets it's numbers. The congressional report is quite informative, and and non-partisan.
Context: 1 acre-foot, a usual unit of measure for agricultural water, is a bit under 1/3 of a million gallons. 17B gallons is about 52k acre-feet.
Vs. just the state of Arizona's historic water allocation, from just the Colorado River system, is over 2M acre-feet. Which, if you're familiar, has created a whole industry of growing extremely water-hungry but rather low-value crops in the desert.
The latest Federal plan to cope with long-term drought in the Colorado River system will cut Arizona's allocation by about 750,000k acre-feet. Vs. that headline about 52k, across the whole US of A.
Yes, locally, data center water use can be a serious issue. And it's useful rhetoric if you are already an anti-data center partisan. But if you're thinking that cutting off water to DC's could fix anything more than some local & isolated problems - sorry, no.
If you are anti-datacenter, then the water argument might look appealing to you, but its NOT a good argument, and it makes not only look you like an innumerate idiot, it also negatively impacts adjacent environmental movements/concerns.
It’s more nuanced, the water consumption can be a real local problem. The evaporated water doesn’t necessarily come back to the original source, and if the water supply was already stressed you’re adding even more pressure to it. You have to look about the actual details, like all arguments about resource consumption
Yeah, but the anti-datacentre people won't care about that. That is an issue for local people and specific cases. Myrmidon isn't addressing the comment at people who have actual local issues with water issues; it was advice for the partisans.
You'd have to look locally at individual datacenter projects then and not be against all of them.
If it's really the case that a local municipality grants a new datacenter to be built that already has a stressed water supply, the people in the area should complain, and if nothing is done then we can be mad at just that ONE datacenter/company. Not the entire industry.
But I just don't see that happening. Noone is going without water because of datacenters. Noone is going without water other than people in California who get there's stolen from farms. That's a real issue, not a hypothetical.
I want to understand if we're really comparing apples to apples. What does 'consumption' mean in these contexts? How much of the water returns to the local water table vs evaporation vs becomes waste water?
At least ~30% of water is evaporated when you use it in agriculture (that is the whole point; plants typically don't go the the trouble of absorbing water just to pump it back into the ground).
Even if you use such an extremely conservative ratio (and assume that all water used in datacenters is 100% evaporated away for cooling) the argument stands.
The question is how much is returned to the original source. Given that the water is ~100% evaporated in the case of DCs it’s a genuine and reasonable question to ask
No, the water is extremely hot. It takes a massive amount of infrastructure before this water can be "released." Unless you are one of those massive polluters that think discharging extremely hot water into the ground is safe and not something that immediately kills the life it's dumped into.
Most of those partisans have no idea what other sources of water consumption are by magnitude. They seem to quote any number as if it is a lot. A billion gallons is a lot. 500000 gallons is not a lot.
Each datacenter uses as much water as 1,000 homes. You're welcome to put such resource constraints in your own local community if you don't care about the increased demand.
That is a number that gets quoted, but I just ran the latest numbers for my local city.
A couple of dozen planned data centers, with an estimated combined use of 160k gallons per day.
Apparently the average household uses 300 gallons a day, which seems insane to me. I guess a lot of people water their lawns. That gets us to a mere 23 households per data center. Even with 100 gallons per day per household or less, the local data centers are not at 1000 households per data center.
And where I live, water is not a concern. They are building a pipe out from Lake Michigan, which has been in the works now for 10 years. That primarily gets the tri-city area off of the aquifers which have been drained at an unsustainable rate for the last 100 years.
If I lived along the Colorado river system, that's a different discussion.
The Colorado water system is a disaster that long predates the invention of computers. The way water rights are implemented in the western US (doctrine of prior appropriation) meant that anyone who owned land through which the river flowed could take enough of it to plant lettuce in the desert.https://en.wikipedia.org/wiki/Yuma_Lettuce_Days
That's nuts. These practices took so much water that native populations could no longer sustain themselves from the river. Complaints about the relatively low use of water by data centers seems pretty weak in comparison, although I still wouldn't locate them in the desert; the desert is a better place to load up with solar and wind farms.
Golf courses use about 1.63 million acre feet, about 32X more, some of that probably reenters watersheds and aquifers, not necessarily great with fertilizer runoff. But they pose lower risk of human extinction, so have that going for them.
Irrespective of how this number is calculated, whether is blue/green water etc here is some comparison data for context, since I personally think that hearing big numbers without any point of reference is pointless.
Annual withdrawals (2015)
- Thermoelectric power: ~48.5 trillion gal
- Irrigation: ~43.1 trillion gal
- Public supply: ~14.2 trillion gal
- Self-supplied industrial: ~5.4 trillion gal
- Aquaculture: ~2.76 trillion gal
- Mining: ~1.46 trillion gal
- Livestock: ~730 billion gal
- Total: ~117.5 trillion gal (322 Bgal/day)
Source: Estimated use of water in the United States in 2015: U.S. Geological Survey Circular - https://doi.org/10.3133/cir1441
> Livestock: ~730 billion gal
This is a good illustration how cursed the datacenter topic has become when people go out of their minds on computer cooling while continuing to eat meat which lets be perfectly clear - is mostly for entertainment purposes. I had someone berate me for defending datacenter water use and their last post was literally a cheese burger.
Well the thing with the livestock is that you have to give a cow water.
Datacenters do not need water. It is just slightly cheaper for them than cooling with electricity.
Datacenters should be required by law to pay to build adequate power generation and infrastructure to pay for their consumption, and ideally in a way where it isn't a ghastly eyesore.
> Well the thing with the livestock is that you have to give a cow water.
you don't _need_ a cow though
Exactly. You also don't need to grow acres of grass in the desert to feed the cow, but we do that too.
Most folks have highly distorted views of how animal agriculture works. It's disgusting and insanely inefficient.
> Datacenters do not need water. It is just slightly cheaper for them than cooling with electricity.
As long as electricity at least partially produced using fossil fuel I would prefer data-centers to use as little electricity as possible. There are regions which are getting plenty of rainfall on average (which can be collected and stored) or were getting some water from rivers will have negligible ecological impact.
But currently DC operators have little incentives to choose DC location based on outdoor temperature or water availability and there are plenty of DCs in hot and arid regions.
> there are plenty of DCs in hot and arid regions
Differences in air temperature are largely below the noise floor in terms of impact on data center cost. Slightly lower energy prices and other operational or capital costs are more than enough to offset significantly hotter climate. Arid climates improve evaporative cooling efficiency.
While I agree fully with this part:
> Datacenters should be required by law to pay to build adequate power generation and infrastructure to pay for their consumption, and ideally in a way where it isn't a ghastly eyesore.
Fact is, almost nobody "needs" meat in their diet, and even if all meat eaters did actually need it rather than merely want it, there are meats which use less than the average quantity of water.
From the above stats, the mean American's loss of water due to unnecessary use of water for unnecessary meat raised in their territory, is about 43 times the mean American's loss of water due to the unnecessary water use of datacenters that are in their territory.
You don't need a law to force people to pay for resources they consume; basic economics already ensures that. If there are externalities not already accounted for by that (increased demand for water/electricity driving up costs for others, noise, "eyesores", etc.) you could pass laws to ensure those costs are paid by the responsible party but that's not a datacenter-specific issue.
Alas, we do, because in the case of water, basic economics are stymied by pre-existing water-rights laws that are batshit insane. As such, we need legislative action to (at minimum) clean up that mess.
Don't quote theoretical econ idealism at us when our actual electricity rates are going up 20% or more per year.
Increased demand absolutely DOES increase the cost both in theory and practice.
I... agree? Did you even read the comment you replied to?
How does that work in a market where you are bidding amongst other participants and capacity is constrained? Supply is limited and demand is growing, the market response is to charge more.
We do not need meat.
Datacenters have a readily available substitute to water if they just make less profit.
I'm not giving up my meat so that datacenters can use it for cooling, if that's what you're suggesting.
From the above stats, a 2.3% change to USA meat consumption, up or down, has the same impact as a 100% change to data centre water consumption, up or down.
If I read the USDA stats correctly, US red meat and poultry production went up about 5.6% from Jan-Jun 2025 to Jan-Jun 2026, i.e. the increased production of meat led to an increase in water use that was twice as much as the water use of all the data centres combined.
Give up your meat consumption because it's more about societal excess than anything else.
Meat is and mostly always has been(in euro societies) a luxury item meant mostly for the rich.
Capitalism turned meat into a MUST HAVE item for keeping up with the Joneses.
We consume absolute insane amounts of meat nowadays, and we waste about half of it, just so people can feel good cuz theyre eating a high class excess meal.
Lol most people throughout history didn't eat much meat. It's wild how modern western people refuse to see how we'll destroy the planet trying to feed 8 billion people with the most inefficient method of creating calories(beef). The inefficiency IS THE POINT.
So? Many people are not willing to give up their cloud services either?
And there are other kinds of meat that cost less water to produce. I don't think someone who isn't willing to give up beef for chicken is at a higher moral ground than someone who wants cheap tokens.
True. I was going to hate the environmental destruction being wrought for the purpose of funneling more of our wealth to a handful billionaires who will use the AI against us, but then I remembered how bad meat eaters are for the plight of the common man, and now data centers don’t seem so bad after all.
In fact many humans need meat because the proper plant protein simply cannot grow/exist in their area.
This is why we evolved to be omnivores. Basic biology laughs at your poorly-researched position.
There would need to be some articulable upside to data centers, though. Even if we agree there should be less livestock, I don't see how that has any bearing on the data centers topic.
I assume HN runs on machines in a datacenter. The servers that serve my business also reside in a datacenter. I see lots of upsides to datacenters, and they've been around long before they were controversial.
And you own clothing that was made with child labor. You monster. Why do you love child labor so much?
It's to put water usage in perspective. I think the best example is golf courses.
If you think datacenters are a waste of water and you want to do a global campaign to stop building / shutdown golf courses first, then your values don't really make sense.
Golf is pure leisure, and uses 531 gallons compared to 17.4 gallons a year.
Datacenters on the other hand, power the site we're using right now to discuss the "problem". You wouldn't be able to have this conversation without them.
--
Why aren't people screaming constantly about golf courses and pressuring to shut them down?
Why the focus on datacenters if the bigger win is staring you right in the face (if you're that concerned about water usage)?
It's not about the water.
I would get rid of every single golf course in the world if I could. One of the worst uses of land out there. Imagine the horrible, gross volume of herbicides each course uses to ensure that only the "correct" grass is growing over all those acres.
The datacenter topic is cursed because people are afraid of ai. People seized on water usage because it is a more legible and tractable focus for their fear.
yeah people should just accept starting to feed on AI slop for sustenance instead of actual food! How dare they have a preference
calling it "actual food" is clearly misleading when majority of meat is used purely for entertainment not for survival sustenance. By that logic compute is "actual medicine" because some of compute is used for medicine?
> majority of meat is used purely for entertainment
Source of that inane statement?
Mr. Gotcha, is that you?
https://thenib.com/mister-gotcha/
US golf courses hit 531 billion gallons in 2024 (1.63 million acre-feet) https://www.gcsaa.org/who-we-are/media/news-release/2025-new...
Almond farming - ~1.5 trillion gal (1.1 gal per almond)
Holy shit, so much precious water used just for almond?
Only 0.5 trillion gallons is precious water, the rest is just regular water
I value almonds more than LLM slop.
I think most of the people who act like this are secretly using ChatGPT like a Republican on Grindr.
Yes I use LLMs at work because I'm "highly encouraged" to as a condition to keep my job. All I see is more paying customers finding more bugs and having a worse experience using our products.
You don't see the issues with your line of thinking? You get to absolve yourself of the responsibilities of your choice because "someone is making you" use it? How can you say you value almonds more than LLM slop when your individual choices don't reflect those values?
They live in a country where you are not allowed to change jobs.
I hate almonds and agree with you
I hate almonds and love LLMs.
Good context, but aren't many of your examples simply putting that water back into the ground? Also, where the water is taken from is the key issue. If this was ocean water nobody would care, but if it's Reston, VA that's a local issue.
For perspective, the total annual water consumption of all US data centers is about 100 days' water production of the nearby water treatment plant to Reston, VA. Or put differently, all of the US data centers can be comfortably supplied solely with the water from the next phase of expansion of said treatment plant (an upgrade of 75MGD).
"17 billion gallons of water a year" sounds like a lot of water, but it's really not.
I'm not sure what led you to believe that the water is being put back into the ground "simply" or otherwise but maybe you shouldn't be acting like it's a non-issue when you clearly don't understand the problem.
DCs are pumping the water through evaporative cooling towers because the latent heat of vaporizaton means there's a huge amount of heat sucked up in the phase change from liquid to vapor. None of the water is being "put back into the ground" except maybe that used during flushing and other maintenance.
"Dry" cooling exists, where there is a closed coolant or refrigerant loop run through radiators, but those require more power, which increases cost and limits how much power can be used for compute (but they also get the say "it makes us use more power, bad for the environment! We're being Green by using evaporative cooling!")
The water that evaporates during cooling does rain down somewhere else at some point.
It isn’t like the water magically disappears. It just gets turned into moisture in the air.
Though that can be very far away depriving the local environment or other industry of the water. Extra bad if that water is pumped up ground water.
No I'm saying that water used in mining likely goes back into the ground.
Then maybe you should have said that the first time.
Nice perspective. So even if all these data centers were built in the desert, it's entirely within the realm of possibility to transport 17B gallons of water, unlike the 43 trillion used for irrigation. Some back-of-the-napkin math:
- A truck holds 5000 gallons
- It would take 3.5 million trips to transport that much
- If each trip was 500 miles roundtrip, that's about 2 billion miles
- Trucks travel 140 billion miles each year in the US. That would add less than 2% to the total miles
- The trucking industry is about $1 trillion in revenue per year. So 2% of that is $20 billion total to transport 17 billion gallons. Or around $1 per gallon.
The problem is that these comparisons do not have equal utility. Trying to compare them to AI datacenter consumption is equivalent to justifying a gambling habit on a fixed income because groceries consume plenty of dollars as well.
The report says that water consumption tripled due to AI datacenters, in 2023! Most of it was on electricity generation to support the datacenters.
But that's tripled of water use across all sectors due to datacenter buildout. So no "comparison" to other uses is needed.
EDIT: Ah, I did misread it. Sorry!
No, it doesn't.
It says that datacenters now use 3× the amount of water they used to use--it's specifically limited to that very, very narrow sector.
> But that's tripled of water use across all sectors due to datacenter buildout. So no "comparison" to other uses is needed.
You're saying almond farm water use tripled due to data center buildout? What other meaning of "all sectors" could make sense of this claim? The report clearly says 17 billion while all other sectors are in the trillions.
You have to look at the local level, water isn’t distributed the same way around a country as large as the US. Some regions have no water problems while others have a lot of consumption and competition to get access to it
Yeah that's fair.
I haven't seen a single article doing that though (from either side) and I honestly think that nobody would even trust these companies at this point (Meta et al) to give an honest analysis anyways.
Yeah in a lot of areas water is the opposite of scarce, practically free, and your bill is 98% infrastructure fees.
I don't hate data centers any less. That's 17B gallons wasted on something I don't care about and actively want less of.
HN Thread with a whataboutism comment as top voted? Must be the datacenter water thingy again. It's always a rhyme of the same comment.
If you need to denigrate pure context as "whataboutism", then maybe your core argument is trash?
I mean, sure, lets not build datacenters that are 100% evaporation-cooled into each desert-oasis.
But if general water availability is your concern, than datacenters are so far down the priority list that any argument involving them is just laughable. And if you are against datacenters, then find a proper argument instead of making environmentalism look stupid.
Datacenter water usage exploded in the last year... here look how much water gets used by all people flushing the toilet. It's whataboutism.
One gallon of almond milk requires 1,600+ gallons of water (likely much more — given some estimates are each almond requires up to 3 gallons of water). US almonds are almost entirely grown in California and consume 1.5T to 2T gallons of the state’s water.
It’s pretty dumb to compare AI/data center usage to water usage when something that’s actually wasteful such as almond milk exists.
There’s better metrics to make one’s anti-AI point.
As a reference, one gallon of cow milk requires about 250 gallons of water.
I feel like we need some citations for these figures. Our World In Data [1] disagrees. I've not seen any analysis that puts any plant milk at a higher water use figure than dairy.
[1] https://ourworldindata.org/environmental-impact-milks
If you want to play these stupid games how much water did you think was used to make the concrete for the data centers? Or how much water was used to cool the steel? How much water did all the workers drink while building the data center? How much water did they use to build all those GPUs?
Almonds provide a real utility for humans: food. LLM slop has done nothing but made the public miserable to help prop up the delusional fantasies of the tech elite (all of which hold deeply anti-human + anti-democratic views).
>Almonds provide a real utility for humans: food.
Let's do not act as if there wasnt other food, a lot of it actually.
Meanwhile LLMs aren't that easy to replace, they're biggest human achievement up to date.
Are they? What has gotten better because of LLMs?
"Meanwhile LLMs aren't that easy to replace"
A BBS chatbot from the 80s (LISA) was almost as good as modern LLMs and that was just based on key words.
So, uh, yea, chatbots are very, VERY easy to replace.
It's really weird to me that datacenter fans all have detailed knowledge of almond farming, but never seem to notice that the nutritional value of ChatGPT is not on par with your average box of almond milk.
Too late.
Because nutritional value is not the only thing of value. We invented this clever way of uniformly valuing things called "money" that abstracts over all of those differences though.
Potable well water should NEVER be used for such purposes. Aquifers recharge at a rate far lower than industry wants to pull from them. At some point potable water for humans will run out. If you're using water just to evaporate it for cooling then use surface water.
Using water for cooling significantly reduces cooling energy use. The decision of whether to use dry, evaporative, or hybrid cooling approaches should be driven by the specific water availability in the area where the cooling process is to be used.
Main regions do have enough water to support evaporative cooling. When water is available, using it is prudent because it saves so much electricity.
Using surface water is not a practical approach for most cases because the process would need to be built next to a river or lake - there is limited industrial real estate in these areas.
What does it mean to be consumed in this context? After used for cooling isn’t it clean enough for drink and other uses? Can’t it go back into other systems? Can’t it be used for agriculture?
"consumed" means it was used for evaporated cooling and let off into the atmosphere. It's obviously a resource that replenishes, but wells replenish at a limited rate [and can run dry] so it's not a negligible issue.
Feel free to google this yourself, or ironically ask an AI, it'll happily confirm it's a real issue.
That's a small portion of the consumption. According to the article, the vast majority of the water is for electricity generation, and the vast majority of that water is from hydroelectric power generation. Hydroelectric generation does not evaporate the water, it releases it downstream in a clean, usable form.
The title seems to be accidentally confusing, the 17 billion is ALL evaporative.
If we want the other 80% (which I don't personally think we care about) 211 billion more.
Can that vapor be captured and condensed again? The energy transfer of evaporation happens before it reaches the cooling tower exit. It's a conservation of energy thing for sure but condensation is doable.
There is also the issue that evaporated water doesn’t necessarily go back to the original water source
What if it’s in an area where there is abundant surface water and 40” of precipitation per year? Not every data center pumps groundwater.
This refers to water that is evaporated for cooling of both the datacenters themselves and the power plants that give them electricity. The remaining water after evaporation is so full of minerals that it has to be discharged as sewage.
This evaporation process is used to cool the closed loop of fluid that actually moves heat in and out of the data halls.
Exactly. Is that figure about the water 'flowing' through the datacenter? I want to believe they would not just dump it after use.
So this isn't really a problem? That's like a small city.
Is anybody working on using all this waste heat for desalination?
I think cold seawater can be used for the external loop, through a heat exchanger where it would meet the heated pure water from the internal cooling loops. This would cause evaporation of the salt water which could be captured.
https://en.wikipedia.org/wiki/Desalination#Multi-stage_flash...
Waste heat from data centers is significantly below the temperature that desalination systems typically use. While a desalination system could theoretically use data center waste heat, it would require far higher capital costs than other desalination systems.
That would require building data centers next to where salt water is or energy used for pumping will eat any savings from using waste heat.
An acre of alfalfa is typically irrigated at 5.5 AF per year. The standard center pivot field is 130 acres ~ 700 AF of irrigation, to grow alfalfa to ship overseas. Converting to gallons, 210 million gallons, for one field.
80 fields of alfalfa equals all the data center water usage.
USDA estimates that total irrigation uses 55 million AF of water per year.
I’ll not dance around it. It seems like most people don’t like the inputs or outputs to this massive generational endeavour.
For a closed loop cooling project:
> modern facilities can reuse
Can, but do they? Feels a bit like Musk's Colossus that "could" use nuclear power but has been running on "temporary" natural-gas turbines for years now.
Nope, because dry cooling uses much more power. Latent heat of evaporation is "free" work.
Would support still jump if voters were told that cooling towers which don't use water need to use significantly more electricity instead?
Lots of places in the US have plentiful water. Don't build data centers where there's a lack of water! And it's a dead horse at this point, but we don't need to be growing almonds in CA. If you actually care about water use, focus on that.
Because people are more "against capitalism" than ever, free markets have become an underrated solution to many of our problems. E.g. 1. stop subsidizing wasteful, environmentally damaging agriculture and 2. legalize building housing. Markets aren't perfect but they could improve the status quo in certain areas.
Now do the amount of water required for subsidized corn ethanol. 30 million acres, trillions of gallons of water, and almost no redeeming value.
What happens with the water in the DC after it has been consumed?
Well it is primarily evaporated into the air. It rains down somewhere, often over the ocean.
It could be infinite, but yes, in the end - vaporating
a little evaps, goes back into the water cycle.
non-issue, chinese propaganda to get us to slow down.
unfortunately some people parrot it, but most are smart enough to know it's bs.
you'd think they would want to shutdown golf courses first, it's pure leisure and much more water, but it's not about that :)
A little evaps? It's like 80%
The worst designs evap up to 80%. New designs use closed-loop system where very little evaps (~6%).
Many designs in-between with air cooling, hybrid, etc.
Still a non-issue, it's going back into the water cycle.
Any water issue would be specific to the municipality, it's not a national issue, nor is it an issue specific to datacenters, it's true for anything being built in the town (golf courses, power plants, etc.)
Article title is extremely misleading. These are figures from 2023 - almost three years old now, and since then US datacenter power consumption has doubled.
The article source for the 17B gallons is a Congressional Research Service report and its source is a Lawrence Berkeley National Laboratory report. Both are interesting reads. The 17B number is an estimate. Here are some of the inputs:
0. https://www.congress.gov/crs-product/R49057#fn101. https://eta.lbl.gov/publications/2024-lbnl-data-center-energ...
It's weird how all these statistics could end if the owners of these data centers would just simply release the numbers for the public to learn from. Instead they are keeping these figures private and we have people trying to best-estimate the reality. Mostly for reasons we all understand: true figures would cause an even larger public backlash.
Fun fact, citizens of The Dalles, OR had to sue their own government to get them to release numbers on how much Google data centers were consuming (info Google stipulated should be kept "proprietary"):
https://www.oregonlive.com/silicon-forest/2022/12/the-dalles...
This is a manufactured propaganda issue by dishonest and/or innumerate people.
However, can’t data centers just use closed-loop cooling and avoid this whole thing? With all the money going into AI, it’s hard to imagine it being prohibitively costly.
With regards to evaporative cooled DCs (either primarily cooled that way or assisted some of the time): Imagine an air conditioner that instead of requiring energy to drive the compressor to make the phase change work, all the phase change was provided by taking in liquid water and evaporating it off as vapor, taking the latent heat of vaporization (2257 kJ/kg or 970 BTU/lb) with it.
I don't even know who owns forgeeks.net. They're only a couple months old and is disseminated by a new HN account. Some of the "author" social media links appear AI generated.
Red flags galore. Wouldn't surprise me if it was bankrolled by bad actors and/or a website created/operated by AI agents.
People have always used a closed loop to move heat in and out of the data hall - but something has to keep that closed loop cool. That's often a cooling tower which uses evaporating water to reject heat into the air in a way that's very energy efficient.
Additionally, the power plants which supply electricity to the data centers have the same cooling problem. They also typically use cooling towers.
It's not dishonest at all. Datacenters and fossil fuel power plants are absolutely evaporating fresh water for cooling. The piece that's missing from that discussion is that it helps them use less electricity.
They use closed loop systems. That reduces the water consumption on-site, but you still need to move the energy out of the DC, which requires a lot of energy. That extra energy itself is likely produced by upstream power plants evaporating water. In practice you mostly move the consumption upstream.
The article links to a report from the library of congress, from which it gets it's numbers. The congressional report is quite informative, and and non-partisan.
https://www.congress.gov/crs-product/R49057
Context: 1 acre-foot, a usual unit of measure for agricultural water, is a bit under 1/3 of a million gallons. 17B gallons is about 52k acre-feet.
Vs. just the state of Arizona's historic water allocation, from just the Colorado River system, is over 2M acre-feet. Which, if you're familiar, has created a whole industry of growing extremely water-hungry but rather low-value crops in the desert.
The latest Federal plan to cope with long-term drought in the Colorado River system will cut Arizona's allocation by about 750,000k acre-feet. Vs. that headline about 52k, across the whole US of A.
Yes, locally, data center water use can be a serious issue. And it's useful rhetoric if you are already an anti-data center partisan. But if you're thinking that cutting off water to DC's could fix anything more than some local & isolated problems - sorry, no.
Completely agree. Specifically:
If you are anti-datacenter, then the water argument might look appealing to you, but its NOT a good argument, and it makes not only look you like an innumerate idiot, it also negatively impacts adjacent environmental movements/concerns.
So don't push it.
It’s more nuanced, the water consumption can be a real local problem. The evaporated water doesn’t necessarily come back to the original source, and if the water supply was already stressed you’re adding even more pressure to it. You have to look about the actual details, like all arguments about resource consumption
Yeah, but the anti-datacentre people won't care about that. That is an issue for local people and specific cases. Myrmidon isn't addressing the comment at people who have actual local issues with water issues; it was advice for the partisans.
You'd have to look locally at individual datacenter projects then and not be against all of them.
If it's really the case that a local municipality grants a new datacenter to be built that already has a stressed water supply, the people in the area should complain, and if nothing is done then we can be mad at just that ONE datacenter/company. Not the entire industry.
But I just don't see that happening. Noone is going without water because of datacenters. Noone is going without water other than people in California who get there's stolen from farms. That's a real issue, not a hypothetical.
I want to understand if we're really comparing apples to apples. What does 'consumption' mean in these contexts? How much of the water returns to the local water table vs evaporation vs becomes waste water?
At least ~30% of water is evaporated when you use it in agriculture (that is the whole point; plants typically don't go the the trouble of absorbing water just to pump it back into the ground).
Even if you use such an extremely conservative ratio (and assume that all water used in datacenters is 100% evaporated away for cooling) the argument stands.
Well in agriculture the water "consumed" is returned with a whole bunch of chemical runoff.
In datacenters the water "consumed" is returned with a whole bunch of...water.
The question is how much is returned to the original source. Given that the water is ~100% evaporated in the case of DCs it’s a genuine and reasonable question to ask
No, the water is extremely hot. It takes a massive amount of infrastructure before this water can be "released." Unless you are one of those massive polluters that think discharging extremely hot water into the ground is safe and not something that immediately kills the life it's dumped into.
heat, too.
Most of those partisans have no idea what other sources of water consumption are by magnitude. They seem to quote any number as if it is a lot. A billion gallons is a lot. 500000 gallons is not a lot.
Each datacenter uses as much water as 1,000 homes. You're welcome to put such resource constraints in your own local community if you don't care about the increased demand.
That is a number that gets quoted, but I just ran the latest numbers for my local city.
A couple of dozen planned data centers, with an estimated combined use of 160k gallons per day.
Apparently the average household uses 300 gallons a day, which seems insane to me. I guess a lot of people water their lawns. That gets us to a mere 23 households per data center. Even with 100 gallons per day per household or less, the local data centers are not at 1000 households per data center.
And where I live, water is not a concern. They are building a pipe out from Lake Michigan, which has been in the works now for 10 years. That primarily gets the tri-city area off of the aquifers which have been drained at an unsustainable rate for the last 100 years.
If I lived along the Colorado river system, that's a different discussion.
The Colorado water system is a disaster that long predates the invention of computers. The way water rights are implemented in the western US (doctrine of prior appropriation) meant that anyone who owned land through which the river flowed could take enough of it to plant lettuce in the desert. https://en.wikipedia.org/wiki/Yuma_Lettuce_Days
That's nuts. These practices took so much water that native populations could no longer sustain themselves from the river. Complaints about the relatively low use of water by data centers seems pretty weak in comparison, although I still wouldn't locate them in the desert; the desert is a better place to load up with solar and wind farms.
https://coloradoriverscience.org/Water_law_and_policy if you're interested.
On the other hand, many homes could mothballing a misplaced farm provide water for?
As far as I know, there's not a single place in the world where someone went without water to cook, drink, or bathe because a data center consumed it.
Completely non responsive
Golf courses use about 1.63 million acre feet, about 32X more, some of that probably reenters watersheds and aquifers, not necessarily great with fertilizer runoff. But they pose lower risk of human extinction, so have that going for them.