I'm an environmental engineer in Southern California and reverse osmosis (RO) systems are one of the things I work on.
It's true as far as it goes that ocean desalination has grown as a water supply option and that energy abundance is beneficial for this. But from the standpoint of water utilities, the relevant question is how ocean desalination compares *to other potential sources* in terms of cost per unit volume of water they can deliver. After all, you generally operate a single interconnected distribution system- so if a new high-volume user (like a data center) forces you to bring online some much more expensive water source, everyone in your service area has to pay for it. And if that means your users are paying for some much more expensive source than the surrounding farmers, they probably won't like that!
Historically, SoCal has been highly dependent on water that has to come a long way through aqueducts from the Colorado River, the State Project, etc. The costs of these imported sources have risen over time, reflecting scarcity of the available flows and conveyance costs, such that a few alternative supplies are now generally economical. As a result, our region *really is* a world leader in potable reuse to replenish surface/groundwater supplies as well as brackish groundwater treatment. Both alternatives include RO and can now be done at costs comparable to imported river water, which is ultimately what drives adoption.
But treating the ocean via RO is still much more expensive because the ocean is orders of magnitude saltier than wastewater or brackish groundwater. So there is a fundamentally higher osmotic pressure and energy intensity as a first-order problem, and the corrosivity of seawater forces you to use expensive materials which also drives up cost. These (more than regulatory concerns) are why utilities right now who are making decisions about alternative sources are going much more in the direction of reuse and brackish groundwater. SoCal has a couple ocean desal facilities, but we probably won't see much more widespread treatment of the ocean unless/until imported water costs get a *lot* higher than they are right now.
It was one of the first things I noticed in his article, that it jumped to ocean desal rather than the much more economical (and often larger scale achievable) brackish (ground, runoff whatever).
In these discussions it's often noted as a ceiling on the cost of potable water -- there's an infinite amount of it relative to our ability to extract. Other things are often better but our worst case is pretty good! I've definitely run into people many times who insist we're about to run out of "cheap" water.
It all depends on what “cheap” means! Water is incredibly cheap, such that it can still be cheap compared to other commodities but hundreds of times more expensive than it is now, and thus too expensive for current uses.
How much of a cost differential is there caused by altitude and pumping? Desalinated sea water obviously starts at sea level and has to be pumped up to the altitude that the user is at. Reservoir water is already at altitude and pipes/aqueducts that carry it are gravity-powered.
Is that significant against the energy costs of RO, or is it trivial by comparison?
I ask this because my uncle was a water supply engineer who worked on some of the early desalination systems in Abu Dhabi and he commented that the pumping was a bigger technical problem than the desalination itself. I don't think that costs were a big factor in those designs based on some of the other things he said.
Sadly, I'm not in a position to ask him myself any more.
The answer to this question varies a lot by geography. If you are (say) the City of Washington, DC you are near the ocean but you also have ample river water available very nearby and conveyance costs for that are very low.
The issue in SoCal is (a) rivers with lots of water in them like the Colorado or Sacramento-San Joaquin system are far away and (b) there are mountains in between you have to pump over. So the conveyance facilities are really something - Metropolitan Water District has 5 pumping plants to get water all the way from Lake Havasu to the SoCal regional system, and they are IIRC the largest such facilities in the world.
"so if a new high-volume user (like a data center) forces you to bring online some much more expensive water source, everyone in your service area has to pay for it."
This is not true. You can just bill the new user a higher rate that covers the extra expense and leave everyone else's rate alone.
Maybe that's not what they CHOOSE to do, but it's definitely what they CAN do.
We have to do something similar at the company I work for. Certain types of new business make us bring in extra more expensive raw materials. We make those new customers pay for those extra costs instead of averaging it out amongst our entire customer base.
Part of the fundamental challenge is that it's usually not economical to build multiple pipe networks - generally speaking everyone has a meter and service line connected to the same distribution system using the same water. Utilities can have different tiers of rates but it's not necessarily straightforward to single out an individual user.
Another challenge is that the volumes being discussed are pretty large compared to the overall capacity of a medium-sized utility. Usually even pretty large commercial or industrial facilities are a drop in the proverbial bucket compared to the utility-wide demand ... but a facility whose needs are on the order of millions of gallons per day (a unit we use for treatment plant size) might require a city to expand its total water supply capacity. So it could well be the case that a utility faces a decision along the lines of "we need a significant new source for this individual user."
None of these are unsolvable problems but they are also non-trivial.
I didn't say anything about building extra pipe networks. This is just an accounting problem.
We have capacity for 100 gallons at $1 a gallon. And we are selling 100 gallons at $1.10. A new customer comes along that needs $10 gallons. To get those new gallons we need to build a desal plant that costs $1.25 a gallon.
You build the desal plant, and charge the new customer $1.35 a gallon, and leave the pricing for everyone else alone. You still ship all the water through the same pipes.
And you can always just create another pricing tier.
Agreed that any time you are taking on a significant new customer you need to do some extra analysis, especially if that requires large Capex spend.
This is basically the kind of calculus a utility would do ... except that the desalination plant would probably pencil out to be much more expensive to both build and operate than other alternatives. So as noted in the OP, a utility around here probably chooses potable reuse or (if an aquifer is handy) brackish groundwater instead. Whenever I've worked on this kind of alternatives analysis those are the winners - I've worked on plenty of reuse and groundwater facilities at this point but the only ocean desalination I've done was a demo project.
The other thing I'd imagine my clients would balk at would be the total demand - one of the linked articles talked about a 9 MGD flow rate which is a medium sized municipal water plant *all by itself* - so the utility would want to be sure it gets paid for the significant capital expense and ongoing O&M and that it can count on a rate structure with this user that can make it work.
Yeah, I was just using a desal plant as an easy example. The point remains the same. If you need new supply to meet new demand, then it's certainly possible to make the new demand pay a higher rate.
Commercial/industrial. I'll note that I'm a civil engineer, but haven't done much in California, so it could be different there. The utility could also maybe do some wheeling and dealing and get the data center to pay for some stuff up front, but typically the rates wouldn't be set differently than other industrial users.
I'm confused about this because development projects here in NH often pay "impact fees" to cover any infrastructure upgrades that the new project triggers, especially things like new traffic lights, road widening, or water/sewer upgrades. I think it's based on state law. Some towns that want to attract certain types of development will wave fees or be more lenient about them, but I think overall impact fees are pretty common.
It is a thing, but it's the difference between up front costs and on-going rates. IE, they typically need to pay for infrastructure needed for them to connect to the system, but they'd pay the same rates going forward.
Uh, but the raw material here - water - is the same for everyone. The data center just needs a relatively large amount of it. But there's no a priori reason to charge a data center using X million gallons more than some houses, factories or farms using X million gallons.
We pretty routinely charge differential prices for large consumers. Often the prices are lower for massive consumers, sometimes higher. And it's true that water is the same for everyone, but the usage isn't: household drinking water is much more essential to supporting life in some area than data-center usage.
Sure. But in general, the market is a suitable mechanism for allocating goods by their value. If water is scarce and I value it more than a data center does, you'd expect the market clearing price to be at a level that's OK for me but prohibitive for the data center, leading to either the data center investing in technology to lower water usage, or the data center locating somewhere with less water scarcity. A lot of California's water woes are the result of blocking the market from doing its thing and instead letting certain sectors (alfalfa!) use vastly more water for cheap than they should.
The entities with the greatest ability to adopt technological alternatives are, ironically, the ones who have the greatest ability to pay for water (i.e., data centers.) There is certainly a world where all resources are market-priced, and quality of living is much lower for many human beings who live in an area.
If existing supply isn't enough to meet the new demand, and you have to purchase more expensive supply to meet it (say a desal plant) then it makes sense for the new demand to pay higher prices.
Note the higher tier pricing doesn't need to be limited to just 1 new project. You could do the same for new housing development, new farm water quota's etc
It's so broken that we can't even think about prices.
Someone already linked Scott Alexander, but alfalfa, just that crop alone, uses more than 6% of California's water. This is used to produce $860 million (2015 numbers) worth of crop. That's with an "m".
One thing I always wonder about these alfalfa statistics is how appropriate it is to separate alfalfa, rather than price it in with whatever crop it’s being rotated with. Do people actually grow alfalfa as the main use of a field? Or is it all crop rotation to keep the field suitable for other crops, so that it’s appropriate to account its water use to those other crops?
where it's grown, irrigated alfalfa is usually the primary crop. the farmer may rotate to another crop for a few years to alleviate disease pressures, but alfalfa is the goal.
Good to know! I have never seen this explicitly discussed in any of the discussion of alfalfa and irrigation. I knew that alfalfa was a nitrogen fixer, and sometimes used as part of a rotation for that purpose, but if alfalfa is actually the primary crop in these places, then I don't have to worry about this accounting.
residential/commercial users usually get water from a local govt or water utility that provides treated water from a tap. users are paying for the water, but more they are paying for treatment and the delivery.
don't know much about industrial users, but it's probably mote similar to ag users.
ag users get water from rain (no charge), natural waterways (their costs to pump and distribute), groundwater (same) or manmade waterways, which are variable, but usually there is a fee charged that is supposed to cover amortized costs of the infrastructure construction plus pumping costs.
afaik, a town or water utility that's also getting water from a canal/pipeline will pay the same fees as an ag user of that canal would.
Tiered usage rates are very much a thing. The challenge here is the scale of the new user - generally even a pretty big commercial or industrial facility is a small share of the total water use utility-wide. But the linked article talks about 9 MGD for a data center which is just an eye-popping number for a commercial user and on the scale of a medium-sized municipal facility all on its own.
So if one of my clients were dealing with this in their service area, they wouldn't just straightforwardly apply their existing tiered rate structure because this user is "off the charts" if you will. They would need to make sure they can provide sufficient flow to meet this demand on top of their existing demands; and if they end up needing to add new capacity they'd need to ensure they have a rate structure that can pay for the capex/opex committment.
Cool stuff- when I lived in SoCal I worked near the Carlsbad Desalination Plant. Always wanted to tour it but never did, should've asked the company if they did tours or something. Maybe you could answer a question that I haven't had success with via Google. You mention that seawater is much saltier than other sources of saltwater. Usually extraction costs rise exponentially the more of something you want to extract. How does that affect the cost structure of desalination? On one hand you mention needing different materials, but it seems to me like the extra salt would be easier and cheaper to remove because it's at a higher initial concentration, limiting additional costs. Or would this be flipped on its head because it causes more osmotic pressure against the desired flow across the filter?
The answer is that energy per unit volume treated by RO depends on the osmotic pressure of the water to be treated, which in turn is proportional to salinity. This osmotic pressure is what must be overcome by high-pressure pumps feeding the RO system to drive water through the membranes.
To put some numbers to this: a brackish water RO system might treat groundwater or wastewater in the range of 1000-3000 mg/L total salts. Its feed pressure will be in the 100-200 psi range to overcome the moderate osmotic pressure of brackish water, and its energy use will be in the range of 1-2 kWh/1000 gal treated.
By contrast, seawater has total salts on the order of 30,000 mg/L. This much higher salinity leads to much higher osmotic pressure, much higher RO operating pressures (>500 psi), and much higher energy use (~10 kWh per 1000 gal). That's the fundamental issue
I know nothing about this, but I see you haven't gotten an answer. My guess (which I really bet is true) is that, if your goal is to produce salt (and don't care how salty the effluent water is), using ocean water would be easier/more cost-effective/requiring of less energy than using brackish water. But if your goal is to produce fresh water with very little salt in it (which is the goal of desalination), the situation is the exact opposite.
I think you can look at it as seawater having more salt and less water - so easier to extract salt and harder to extract water.
This is true, but I read it just to serves as proof-of-concept that there's no meaningful upper bound on supply once you're willing to pay that price.
There are a *lot* of things we can, should, and maybe even will do before resorting to large scale ocean water desalination in the US. Though, I have little hope of having a coherent water *or* energy policy as long as we remain so unwilling to accurately price things based on their costs.
Southern California has a surprisingly coherent policy of adopting potable reuse and other alternative supplies to make our water supply more sustainable. It helps that we have a regional consortium of utilities in the form of the Metropolitan Water District to organize this effort.
Bottom line is correct - we are already building out quite a lot of reuse and brackish desalination but we're still a long way from the ocean being widely adopted. Of course it could be done if the alternative is no water, but people don't like really high bills!
Putting aside the normative question of irrigating a lot of farmland in the arid Southwest, I can say that it will continue to happen as far as the eye can see so that scarcity is a fact of life.
And you make a good point about what it would take for a user to supply a few million gallons per day on their own. They would need to build essentially a municipal scale treatment system AND (if it's an RO system) a pretty large power plant to run it. Not impossible (we did something like this for a semiconductor facility with to treat realt complex wastewater) but it's a massive undertaking.
"The couple replaced most of their appliances in 2019, and then again in 2021 and 2024. Residue now gathers at the bottom of their backyard pool"
I can't read the whole article so maybe I'm missing something, but If your well is prone to drawing in sediment, the first thing you do is install a $30 whole house sediment filter. If you do this, your water pressure will drop when the filter gets clogged with too much sediment. Then you just swap out the replaceable filter. You can't ruin appliances because sediment can't get past the filter. If they had to replace appliances multiple times, they're likely not getting good advice or there's something else going on beyond sediment.
Also, sediment in well water is natural and I'd say it's the norm, not the exception. My house has a sediment filter that I change twice a year (5 minutes, $10). The filter canister ends up with an inch or two of brown/black sludge at the bottom that I need to rinse out with each change. This is normal. My water tests safe and tastes way better than most any muni water. The addition of the data center might have changed the level of the water table and that could theoretically change the exact composition of the water that nearby wells draw in, but so could a change in water usage by the homeowners.
I think there was a case like this in NH after a spring water company came into a town and built a bunch of extraction wells. The solution was that that company agreed to pay to have some nearby residential wells re-drilled to get them to a depth with decent water flows again.
The other thing I'll say about this as a long-time well water user, is that the swimming pool is a bit of a red flag. I'm not sure if the article covers it, but many people with wells need to hire a water company for filling the pool to avoid depleting the well or increasing sediment levels in the water.
Most residential wells are not infinite sources of water under the best of circumstances. I'm not an expert, but usually well drillers only drill deep enough to get a flow rate and water quality that's adequate for typical household daily use. The cost of an 800 foot deep well is nearly twice as much as a 400 foot well, so the goal is to do the minimum drilling that gets a workable result.
I don't have a sprinkler system at my house, but my neighbor installed one, and when they did, that they needed to spend over $10k to drill a second well just for the sprinkler system. Apparently, the installers found that their existing well wouldn't be able handle the additional draw of the sprinkler system.
Matt’s suggestion that we should regulate to prevent this sort of problem seems naive. If every potential $25k harm triggers preemptive regulation, the wheels of commerce would grind to a halt.
But context matters. This is a $750 million project and from the looks of the photo there are probably something like 10 wells within a 1/2 mile radius of the plant.
Working out some sort of deal that lowers the burden of proof for people that have well problems within a short period of the water draw from the data center (basically assume cause->effect without huge burden) would not grind the wheels of commerce to a halt and would likely be a good PR move, as long as they could structure the settlement to avoid an escalation of liability in the future.
It seems that the best way to do this is to allocate each property in the area a certain amount of groundwater they can pump per month or year or whatever, and let them sell these rights to one another. If the data center wants to pump more water than the land they're on is allocated, they can buy water rights from a nearby landowner or buy water from another source.
It does seem like something worth regulating. This is one of many issues where a negative externality hasn't generally been regulated because the usual user creates and experiences a de minimis amount of externality. When a new user suddenly creates a meaningful amount of the externality, or when new users are extremely susceptible to the externality, everyone suddenly has to play catch-up. This is worse than a situation where the problem is regulated up front.
A good example might be noise. In many places, noise is barely regulated; low density and common activities don't generate much noise or suffer much from noise. But if an extremely noisy user wants to move into an area (like an airport), noise can suddenly be an issue. Similarly, airport noise might not be a problem when the airport is in the middle of pasture land, but when the pasture land is developed into housing the airport noise is now a problem.
yeah, it wasn't clear in the article, but it sounded very much like the data center in question had put in a deeper well and was drawing a lot of water, harming nearby residents with shallow wells.
i'm surprised this wasn't a known issue that the data center would have had to agree to remedy as a condition of permits, tbh.
Having googled this case some more, it's less straightforward than it first seemed, because the data center started construction in 2018, but didn't start operation until 2020 or even later. The couple's first round of appliance replacements was in 2019, before operations even started. At first land clearing and site prep was blamed, not water extraction or data center operations.
This story is duplicated all over the place and seems to have become the poster-child story about why data centers are bad. The reporting all around is frankly just plain crap imo. Not one of the articles I've read includes any actual journalism or reporting out of the story. There's very little from any experts on how wells normally work (eg sediment is very normal with well water), or what's been attempted for remediation. There's also no mention of what water tests have been done or what water experts have said about it.
Dodged the segue into the hydra of water rights legislation! I'd love to read that full-length post someday though. Can't properly understand Californian history without that throughline. But I will note that even in perennially-experiencing-drought CA, home of hippies and granola, the average person's attitude toward water is still that it is...effectively too cheap to meter. Which makes it all the more disingenuous when people whinge about water usage whenever it becomes rhetorically convenient. An actually empirical "full lifecycle analysis" turns so many such environmental intuitions on their heads. Believe The Science(tm)!
Other fun rebuttals to lab-grown-pearl-clutching about AI water use: do you fret about water every time you google something? No? That's like 1/5th of a ChatGPT. Enjoy "devil nuts", also known as almonds? See this bag full of hundreds of them? Yeah, the water used to grow a single almond is enough for like 50-150 ChatGPTs. One could go on. Water is an essential building block of all life, all modern amenities, and we're victims of our own success that 99% of that cost takes place out of sight and far away now. People care so little that they'll continue to buy bottled water rather than drink Hetch Hetchy tap, some of the best government-provided in the nation, if not the world...(usually can't be bothered to recycle it afterwards either, not that that does anything, but you gotta commit to the bit)
Here in Los Angeles, water costs me two cents per gallon, so about $2/day. Of course, that's for Tier 1 water usage, the lowest, or most efficient, rate. If you're a super water hog and push your usage way up into Tier 4, that will increase your water cost on the margin by 36%. Or a little under three cents per gallon.
Isn’t the heavy energy use (and, I guess, water) of LLMs in the training phase? I figure once you have your weights, it’s a pretty straightforward process
My reading level on this is no longer cutting-edge (if it ever was), but I believe inference can still be reasonably expensive for frontier models, especially with chain-of-thought and multimodal reasoning. There are a lot of techniques to reduce compute and memory demands for inference but AFAIK they all come with some loss of performance (often much less than you'd expect, although not quite as much less as we hoped when people were writing papers last year about magical 1.58-bit quantization for every model).
That SoCal water has traveled all the way from the Colorado Rockies and picked up a lot of off flavors along the way. San Francisco dammed up a valley in Yosemite way back when governments were allowed to do such things, and the water tastes excellent.
You are getting somewhat better at understanding western water issues and yes, pricing is part of the problem but it’s a lot more complicated than that.
But one thing of several to understand is that here in the west we have a water rights system that is fundamentally different from the eastern US. It’s a seniority-based prior appropriation system that, in many cases (particularly the Colorado river basin) allocates more water than usually exists. And it’s usually the case that agricultural users have the senior rights compared to urban users.
And because it is a rights-based system, it’s not just about the prices. The problem for many urban areas in the west is not directly about pricing water to the end user, it’s about acquiring rights to the water in the first place.
Yes, obviously. So the state of CA could buy out the farmer's rights (after a few court cases, one assumes) and reallocate them to suburbanites, or whatever. But that's a more difficult process than just increasing prices to market and doesn't help vis a vis Mexico or Arizona.
I'd be totally in favor of a federal bailout so California could tell the farmers to f off. I consider this to be one of the biggest governing problems nationally. The largest state in the country has an insane, archaic, and cronyist water system and nobody's really even trying to fix it.
The state can get rights to water and then give them to municipalities, but that doesn’t mean it’s easy.
By way of analogy, it is super simple in theory for the state to acquire land for high speed rail and then build it, but in practice it is extremely difficult.
There is also the issue of interstate compacts and other history and agreements such as the Colorado River compact, which is a water sharing agreement among 7 states, one that overestimated the amount of water would be available, among other practical problems.
Imagine the Anton Chigurh "If the rule you followed brought you to this, of what use was the rule?" meme here, but instead of Anton Chigurh, it's Nobel prize-winning economist James M. Buchanan.
Who is the "you" is the critical question when it comes to the Colorado River. The rights in question here are multi-state and were negotiated in a compact signed in 1922. Changing that is the subject of ongoing negotiations and it's a really tough fight because there are lots of interests competing for diminishing water. Even if California wanted to raise the cost (or lower the share) of water for in-state agriculture, I'm not sure it would have a legal basis for doing that and the farmers could probably successfully sue based on the 1922 compact.
So it's a difficult multi-state issue that needs effective mediation and guidance from the federal government. Luckily we now have an excellent administration in place that will make that a breeze.
Do you have a cite? As far as I can tell, there's nothing stopping CA from doing what they want internally with the water they're allocated on an inter-state basis.
I haven't done a deep dive but apparently it's part of the senior rights of farmers in the Imperial Valley, e.g.: "Imperial’s farmers are protected by longstanding and increasingly controversial senior water rights. Even the largest cities in Southern California and Arizona have water rights junior to Imperial’s, which gives its farmers legal priority in times of scarcity."
That article is confusingly written, but the gist seems to be that this is about prices, not some kind of inalienable rights. The state apparently can and does buy the farmers out.
There doesn't seem to be any reason the state could not continue doing this and further reduce agricultural draw, save some water, sell some of the water saved to urban users, and resolve the state's water problems with no net fiscal cost.
Prices reflect exchange, not legal enforcement. It's because they have senior rights that the only way to reduce their share of California's water is to offer them something, in this case a buyout. In other words, they leverage their senior rights to get money.
It is about water rights. In Western water law, water rights are like property rights and get basically all the same protections. Yes, California could buy them, but the cost would be astronomical. Eminent domain is possible but would generate a political firestorm and years of litigation, plust the aforementioned astronomical costs.
There is an active market in water rights sales and transfers in California, although it is not a "free" market because almost everything has to be approved and permitted by California Water Resources Board. So there is undoubtedly some Abundance-relevant friction in the system that could be reduced.
One of my pipe dreams is that the federal government would decide to destroy the Glen Canyon Dam and drain Lake Powell into Lake Mead. This would drastically decrease the amount of evaporation and increase the total amount of water in the Colorado River. But alas, I don't think this will ever happen.
Beat me to it. I'm regularly saying that Matt needs to spend some time learning about how much differently water rights operate out here and how big of a deal it is.
I don’t think it’s worth learning. The tldr is that western water rights are extremely inefficient. If you want to optimize within the system, sure, learn the rules, but the point is the rules should be changed.
It's easy to say the rules should change on a forum like this, but it will be extremely difficult to do in practice. Which is why it would be good for a smart, pragmatic person like Matt thinks to consider a way to practically work through the rules.
I'm not sure it's more inefficient than the scheme in the east, where water rights are tied to land rights. It's a different set of tradeoffs. The advantage is that there is so much water in the east that water isn't scarce, so leasing rights isn't a big deal.
I don't think this is true. The idea behind Coasean bargaining is that the initial allocation of rights doesn't matter, as long as the rights are alienable, you can still reach the efficient final allocation via market mechanisms.
As far as I can tell, it is in fact possible to buy and sell water rights. So I don't see why it should matter how the rights work.
Yes, they can sell their water rights. But if you are a farmer or use those water rights for your agriculture (as is usually the case), then selling the water rights makes the land you work much less valuable at a minimum. It also often means you can no longer engage in agriculture. Here on the front range of Colorado, we've seen old ranches near urban areas sell their water rights, but also sell their land for (sub)urban development. But that isn't possible in most places.
The actual problem with water use in datacenters, which the NYT article doesn’t discuss but is a real issue specifically impacting that specific datacenter and community, is that they typically do use more water than a similarly sized warehouse or manufacturing facility and the watershed doesn’t adequately plan or augment for their impact. This can take some time to play out because the massive sites like Meta Stanton Springs generally take years to scale up to a significant utilization percentage of their space and power availability, which makes it even harder to accommodate if the water impact is unexpected. Newton County seemed to be caught blindsided about two years ago by how much water was being returned for treatment even though the facility first went live in 2018. At the same time, the industry is quickly adopting closed loop cooling systems, including ones that don’t use water at all (with questionable environmental impact!), and I suspect that in practice that water use won’t be the thing that folks complain about with their local datacenters in a few years.
I don’t understand why the data centers need a continuous stream of treated water.
Presumably, the water is not mixing with anything at the data centers, so why is it preferable to pump treated water in (and pay for the water and sewage fees) vs having water in a closed loop system that pumps it into the ground to cool it?
This definitely seems like a water pricing issue. I am skeptical the water is priced correctly if using that amount of treated water is more economical than utilizing the closed loop system.
The short version is that it's more expensive except for the largest and densest sites, and those sizes and densities didn't exist until the last couple of years, so in practice no one was going to deploy them. As I noted, that's changing and I expect that in the next 2-3 years this will be the norm, not that the exception.
We do use closed loops, but usually the cooling water is cooled by heat exchange rather than by pumping the water into the ground. The question is, is it cheaper to use power to cool the ground water in a closed loop, or cheaper to just flush the water once it gets too warm and replace it? Often the latter is cheaper, but it will depend on power costs. As you say, its a water pricing issue.
I'm sure they'd already be doing this if the economics made sense, but it would be cool if they mollified the neighbors by offering free or cheap hot water for heating in the winter.
Not much if it's a water based system, any impact would be related to construction/installation which is minimal relative to the larger facility. My comment was in reference to some of the newer systems which don't use water at all, generally leveraging glycol instead. You don't want that stuff leaching into soil.
Every time I’ve seen rationally pricing water come up it drives a certain kind of left person insane.
I can’t recall the title of it but there’s an anti nestle bottled water documentary that makes it seem like the whole goal of pricing water is the screw poor people out of drinking and bathing water and force them to buy it from big conglomerates and agribusiness who benefit from super cheap water gets at most a by the way mention.
I mean $8 for a bottle of water might set me off too, but the replies are full of people pointing out that Wrigley has free bottle filling stations. Which still get some people mad because someone might die if they don't get water delivered to their seat and poured down their throat for them, because they were taken at gunpoint to a Chicago Cubs game.
Man Disney World is looking judgmentally at Cubs games for having 8 dollar water. Incidentally also free water is available all over the place and yet people choose the worst possible bottled water over it.
The issue is that any way we can stop AI, ingenuous or not, is worth doing. If water works as a way to get them to stop building a data center in your area, use water. If electricity works, use electricity. These places *don’t hire locals* and take more than they contribute to every local economy they interact with, outside of buying off local politicos.
I’m generally against this kind of behavior when it comes to building useful things like housing, rail, even necessary LULUs, but when it comes to data centers, the “we shall fight on the beaches, we shall fight on the landing grounds, we shall fight in the fields and in the streets, we shall fight in the hills; we shall never surrender” approach is completely appropriate. It doesn’t matter how, stop the data centers.
Why are data centers any worse than any number of other businesses that can operate in an area? Is this based on fear of AI superintelligence killing humans or something?
I’m not particularly afraid of Skynet or murderbots swarming America, though maybe I should be, but that’s not something I have any control over. AI only benefits a few people, and they’re usually bad people like Elon Musk, Peter Thiel, and Sam Altman. And it harms basically everyone else. Through job losses, increased surveillance by our rogue government, data harvesting, enshittification, fucking up the education system, etc. We see arguments against AI here routinely, I agree with basically all of them. I’m for cutting the internet backbone most of the time.
At least when a new polluting industry builds a factory in town, it creates local jobs and probably makes a useful product. But a data center at most employs what, 20 people, usually from out of state? While driving up the energy prices for local users and possibly polluting or exhausting the local water? The juice isn’t worth the squeeze. Which is why Elon comes to places like Memphis to build his data centers because it’s too economically depressed and politically torn to fight it. And it puts our aquifer - the one real asset Memphis has - at risk (our water is the best in the US). If there’s a 0.000000000001% chance of a catastrophic risk or a $0.01 increase of a single citizen’s water or electric bill, it’s not worth it to allow a data center, because the costs outweigh the benefits even at those costs! They’re bads, not goods
They have almost no staff. Things aren't bought locally, they ship it all in. They just consume the electricity and water.
That's the argument as I'd make it, not sure I buy it, but an Amazon distribution center hires people and even with lots of robots at worst locals still get their packages faster.
Honestly, this sounds a lot like the usual NIMBYism. Data centers have diffuse benefits and concentrated costs. If a utility district is worried about water or electricity consumption, the right solution is to just raise the prices.
You can stop data centers being built in your area, but you can't stop them being built. The value is going to accrue *somewhere* and to someone. I wouldn't worry too much about Thiel, etc... China is gonna wreck them because they backed a degrowth fascism that is inimical to innovation and progress (which isn't to say that Xi is a nice man).
We could give an exemption for the first X gallons of water per person or something. The inability of lefties to think through scarcity issues is just baffling.
I don't know who came up with the phrase green on the outside, red on the inside but it is a problem. They don't want to solve the problem unless it means overturning capitalism.
My general impression is that things flow as follows: (1) libertarian types laugh at the left-winger for predicting that pricing will lead to capitalist abuse and monopoly rent-seeking, (2) the abuse and rent-seeking quickly follow, (3) libertarians don't ever discuss it again.
If you have a skeptical view of tech-capitalism in 2025, you're not necessarily correct. But you're also not going to be wrong more than you're right.
Are you familiar with water pricing in the United States? There is already plenty of rent seeking. Open this market up to newer entrants and markets, by all means, I’m sure it will be as rationally managed as Bitcoin and the nation’s housing stock has been.
it's funny because the progressive twitter discussion on water pricing always seems to assume that some entity shoyld be forcing wealthy farmers to pay someone for their water when actual water markets involve someone buying water from farmers with water rights.
That might be true in some places, but its not true in most of California. Few farmers have water rights beyond ground water. Most of the water in the Central Valley is coming from the state via one of the large aqueducts.
that's still a water right, not a market. how much a water user gets is governed by right and the irrigation district charges a fee to cover the cost of infrastructure construction, maintenance and energy use.
the water market is when the rights holder sells their allocation to whoever offers the most money.
Its not a water right in the legal sense. If you have a water right, you can take water from that source without paying anyone. In California everyone has groundwater rights to their property, anyone with a lake or river on their property has limited riparian water rights, but usually when we talk about water rights those are appropriative rights, where people harnessed and diverted water not on their property for beneficial use, often for mining but also for farming, thus gaining the next most senior right to that water source. Water from the CVP and SWP isn’t any of those things - you pay a state agency to take water from their aqueducts. Farmers pay the same way (although a lower amount) than municipalities and industry. Its not really a market - its not like there’s any competition and the price-setting is bureaucratic, but it doesn’t follow the model you describe where municipalities are paying appropriative rights holders. That does happen, but mostly in older mining areas where municipalities use water sources originally developed for hydraulic mining.
It seems like it would be relatively straightforward building a coalition around farmers if the lefties were just stock left neoliberals who wanted to make sure everyone could get enough water you know by pricing it reasonably.
“ The bad news is that the Trump administration is moving to allow more mercury, arsenic, and benzene into the air and water”
I don’t know what safe levels of these compounds and elements are, but could it be that we’ve already mandated that and are now regulating just to regulate? I ask because I have had experience with just that.
My career was in the oilfields for the most part. We took what came out of the well and separated it into its constituent parts, generally oil, natural gas and water. The water was oily, and in the old days they just dumped that overboard, which was bad. The government started regulating this. It started with “just don’t create a sheen”. Then they started mandating so many parts per million and kept lowering it past the point of absurdity. All it was doing was costing lots of money; it wasn’t saving a fish or helping a human in any way, unless he sold water treatment equipment.
There is an inertia to regulating, in which the regulation becomes the object, instead of the the safe air, water, or what have you. And we are way over-regulated.
I think this is an overly simplistic view (understandably) from someone who works in a regulated entity about how regulators actually go about making regulatory decisions.
I encourage people to actually check out regulations.gov and actually look at some rules. Someone mentioned mercury. OK, let's look at a mercury rule. This one is on chlor-alkali plants (chlor-alkali plants produce other chemicals using electrolysis):
After clicking on the link, click on the "docket documents" to review some of the supporting information to justify the rule. You can also review EPA's comments in response to comments made during the rulemaking phase from industry, other agencies (OMB), etc. Maybe you think this rulemaking is still arbitrary. Fine! But I think it's an uphill battle if you think EPA just invented some arbitrary standard unrelated to cost or health benefits.
Slow Boring commenters are generally pretty good, but they've also internalized "environmental regulations bad" (by which they actually mean NEPA is bad because it's the only environmental law people are broadly familiar with in the comments). It's mildly annoying.
I don’t know if these are good or bad, but the government regulators made bad regulations on PPM of oil in produced water. As technology came out that made it possible to reduce the PPM the regulations tightened and for no good reason. No one (fish, fowl nor human) was getting benefit from this.
If you had a bit more specific info, I could probably find it for you. Above, you mention "don't just create a sheen," which makes me almost certain you're referring to something called the Spill Prevention, Control and Countermeasure Rule. This has supplementary rulemaking, including something called the Discharge of Oil Regulation (aka the "Sheen Rule"). The Sheen Rule is not a regulatory standard; it's a reporting standard.
The Sheen Rule requires requires the person in charge of a facility or vessel responsible for discharging oil that may be "harmful to the public health or welfare" to report the spill to the federal government. This doesn't mean that no sheen is safe, it just means there is no responsibility for reporting. The "just don't create a sheen" standard your employer used was to avoid this reporting.
I've read that the clean air act generally relies on cost benefit anslysis to set pollutant limits, and nuclear regulation generally relies on "as safe as reasonably possible". Where does the clean water act fall on that spectrum?
Great column, and thanks for the link to the very interesting Ritchie piece, but this seems a misstatement: "according to Hannah Ritchie’s calculations, we could desalinate the average American’s annual drinking water needs for $154.45." That number was for total domestic use. The number for actual *drinking* water is just $2.30 per person per year -- shown in the chart you included and stated in her piece.
“If the average person in the United States uses 310 litres of water per day for domestic uses, it would cost them $0.42 per day. Or $154 per year.
In the UK, it would be similar: $0.43 per day, and $159 per year. British electricity is more expensive, but we also use less water so the costs balance out.
To get the WHO’s “minimum” domestic supply of 50 litres per day would cost around $0.11 per day and $38 per year.10
Here’s the surprising figure. Producing enough drinking water for someone — assuming 3 litres per day — costs just $2.30 for the entire year. That’s less than the cost of a single bottle of water in many countries.”
I understood Matt to be saying it would be enough water for a household’s direct uses (drinking water, laundry, shower, lawn, etc) but not indirect water uses (hamburger from store).
I thought it was clear (maybe incorrectly) that 'domestic' was what is used by all American households per capita for everything: drinking, bathing, washing the lawn, sprinkling the car, cooking.
It's not "total American usage divided by population" which is also important, but people really really fret about their *house* not having enough water.
1135 liters per household per day (per David) lines up with about 200 liters per person per day (per Matt's graph) if households have 5.6 people. That's probably big by 2x, but the numbers are so small that 2x doesn't matter too much.
I don't know for sure, but the EPA fact sheet linked in this sentence by Ritchie "The average American household uses around 1135 litres of water per day," includes discussion of outdoor use, including lawn irrigation. (https://www.epa.gov/watersense/statistics-and-facts)
Well I'm certainly better informed for reading this article. The two (very well presented) graphs were brutally conclusive.
Here in Australia I suspect that water scarcity is more of an issue than the US (especially over here in West Oz). I heard one expert say on a quality radio documentary that perhaps data centres could invest in ultra treated waste water (indeed sewerage) and wouldn't have the yuck factor that prevents it's residential use. He seemed to think it would be economically viable but I don't know how many data centres he'd built.
If there's a well informed reader of this who could tell me about the economics of waste water purification versus desalination I'd be very interested in his, her (or their!) comments.
There is a strain of left wing folk beliefs that spread like wildfire online. One is housing is expensive because "private equity" is buying up all the houses. It's certainly not community resistance to change. Then you look at the numbers and all corporations including private equity own 500,000 SFHs out of 85,000,000.
Similarly "ai is using all the water!!!" then you look at the numbers - it's 0.2% or whatever.
Two things can simultaneously be true: (1) private equity can own a tiny percentage of SFHs and (2) *in a given market* private equity or other large capital pools intending rentals can be sucking up most of the available inventory and driving up the price of available houses.
But when you hear commentary online do you think most people are referencing a specific location where they know for a fact that happened? Or is it a myth they cling to so they don't have to confront their NIMBYism.
I mean, I’ve read at least one article about it being the case in various markets, and it read as plausible to me. Yeah, I don’t think it’s the case everywhere. Though a lot of people, wrongly in my mind, have an anti-landlord bias, so anyone buying-to-rent or building-to-rent is an issue to people.
Also, that’s the nature of the game with the internet. If town X has this issue the whole world will find out because there’s a NYT article denouncing Blackrock or whatever. Same thing as when one school in San Francisco does a stupid reverse racism or tries to make everyone trans or whatever and the whole world finds out because, internet. The internet makes the discourse toxic
Beware of Big Wine. I haven't checked on this recently, but in Napa Valley at least a while ago, vineyards were scooping up all the available water and some cities were putting restrictions on how often people could flush their toilets (please don't ask me how they enforced this).
Yes, too bad for them, and others in the wine producing areas of California, that they're not connected to the State Water Project and have to rely pretty much entirely on local sources of water.
There are closed loop water and non-water cooling solutions and design now that this is emerging is starting to tackle the somewhat lazy/naive earlier designs of Cool Everything
I was merely pointing out a large body of nonpotable water. There's a lot of nonpotable water around. It accounts for more than 90 percent of all water.
I don’t think you could do this cost effectively. Waste heat is difficult to manage, I don’t think data centers heat up enough to phase change water etc
Right. The residual heat removal from running a data center at design temperature doesn't produce the heat required for phase change. There is a large up-front water investment in data centers (depending on the design) but in a closed-loop system, there isn't much in the way of water usage - that water is constantly absorbing heat and then rejecting that heat, but under normal operating conditions, there isn't much in the way of lost water. As mentioned, the real cost comes from energy usage.
Why does it seem like all the major American AI companies are trying to brute-force their way to the most energy-intensive models and none are trying to develop ones that can run on less energy like DeepSeek did so that for example people in Africa where they have very limited energy generation can run it on their phones? Tech companies 10 years ago at least tried to virtue signal that they were going to make things accessible to the poor and close the “digital divide” so this seems to be another example of a vibe-shift towards a more selfish culture.
It’s also concerning how this increase in data centers is going to increase data storage and thus surveillance capacity, especially as people start relying on cloud AI for more and more personal stuff. In addition to working on cheaper models that can run locally, we should have a GDPR-like law in the US that limits what data can be retained and guarantees people the right to delete their data.
Even so there’s a difference between a world where good AI is cheap enough to run on relatively small clouds that are widely dispersed across many countries and companies versus one where they can only be run on large clouds that are all controlled by a handful of US tech companies.
My understanding is AI can run on less than a million bucks in hardware — totally accessible to medium sized businesses and dedicated startups—. and the real constraint is getting the weights and IP.
The main labs are focused on developing agents (on the way to AGI), which is an energy arms race. Along the way, small, efficient models have come out of these efforts. Google makes Gemma; Anthropic makes Haiku; OpenAI makes a series of mini models; Meta has small models as well. Then you have labs focused on this, like AliBaba, Kimi, and Mistral. Together, all of them have made AI free to anyone with an internet connection, and the quality of the free option has been getting better continuously.
You mention DeepSeek, but that release actually did lead to drastic cost reduction efforts at the big labs. o3, the best model by my lights, has gotten very cheap compared to six months ago. That's a direct result of increased energy efficiency in response to competitive pressures.
Alibaba and Kimi are Chinese and Mistral is European. US companies don’t seem interested in doing this and if they do it incidentally it’s due to foreign competition. One reason US attempts to prevent diffusion of AI would actually be extremely bad for humanity if they succeed.
I'm not sure what 'interested in doing this' should look like if it's not models like Gemini 2.5 Flash or o4-mini. Granted, what you're talking about isn't the main object of these labs--but it wouldn't be the main object of those other non-US labs if they had the time and compute to compete in the main race. Competition gets the outcomes you want from the American labs, so I don't understand the concerns in this context.
Now if you're making a point about the US government's role, I think that's a different discussion and they could muddy what's been a very impressive global diffusion process for this technology.
Phones on 3,4 or 5g bandwith are no where near capable of AI platforming (and neither do such people hae the energy access)
This is like on the order of "why don't we shoot our garbage into the sun" - it's "not even wrong."
AI platforming centers are getting more energy efficient as costs on energy explode and so initially lazy designs are rapidly getting evolved. However industrial scale physical infra innovation doesn't occur at internet commentary speed.
The AI services are currently massively losing money. They are planning on scaling up based on the current consumption patterns and growth holding. But to make that all a going concern, they need it to be widely used and profitable. We are at the point "then a miracle occurs" stage. As that miracle continues to not happen, much of the planned expansion is going to slowly evaporate. My need for agents and generative AI is close to zero and will be zero if I have to use a cloud provider.
My bet is on Google for what you're talking about. They pulled it off with TPUs and Google Translate - they invested a huge amount of resources upfront to make the dollars per query as low as possible, so that they could release it for free. So efficiency is a little bit in their corporate DNA.
Nobody has focused heavily on monetizing "good enough" models, because they believe spending effort on this will be obsoleted by wildly-improved near-future models. So far they haven't exactly been wrong about this. So everyone is racing to find out where the ceiling is, rather than optimizing for anything else. At some point (I presume?) there will either be a ceiling or we'll get the singularity, and this will change.
There are lots of cheaper models you can run locally. But by virtue of being smaller (which is what makes them cheaper) they aren't as good as the big models. But both of them get better with time -- the small models of today are better than the big models from 3 years ago.
It’s not ignoring ECON101. Societies operate by human consensus. Our wiring is such that there is no way 100% of our relations with fellow humans will be operated under market principles. History has shown that 0% is also the wrong answer. So various societies have experimented with various allocations. US has an unusually high share of human interactions governed under market rules. Europe has less (helpful framing for both a “rude” or just not subservient waiter as well as a guy who helps you put your car bumper back vs asking for your insurance info).
But even in the US it is awfully hard to get people to put life’s basic necessities into the category of market exchanges. So we might have to find a construct that produces a more optimal outcome without ECON101 re education camps.
It’s true, market prices for water would stop you from allocating enormous amounts of water to growing alfalfa in the desert.
But even if you don’t like markets, I’m curious why your society can’t reach a non-market consensus that you should not allocate enormous amounts of water to growing alfalfa in the desert, as you are draining your aquifers dry and emptying your rivers? Most of you are not alfalfa farmers.
It's all about antiquated water rights in the west and no politicians have the balls to do something about it. Instead they put the squeeze on residents to curtail water usage which is lame because (in CA anyway) all of the residential/industrial/parks/golf course only accounts for 20 percent of water allocation. Cattle, directly and indirectly, account for about 50 percent(!) and don't add much to the state's economy. Almonds are popular to demonize but they are only 10 percent.
One solution may be to stop the unnecessary corn ethanol production and grow cattle feed in those areas which have much more water.
I'm an environmental engineer in Southern California and reverse osmosis (RO) systems are one of the things I work on.
It's true as far as it goes that ocean desalination has grown as a water supply option and that energy abundance is beneficial for this. But from the standpoint of water utilities, the relevant question is how ocean desalination compares *to other potential sources* in terms of cost per unit volume of water they can deliver. After all, you generally operate a single interconnected distribution system- so if a new high-volume user (like a data center) forces you to bring online some much more expensive water source, everyone in your service area has to pay for it. And if that means your users are paying for some much more expensive source than the surrounding farmers, they probably won't like that!
Historically, SoCal has been highly dependent on water that has to come a long way through aqueducts from the Colorado River, the State Project, etc. The costs of these imported sources have risen over time, reflecting scarcity of the available flows and conveyance costs, such that a few alternative supplies are now generally economical. As a result, our region *really is* a world leader in potable reuse to replenish surface/groundwater supplies as well as brackish groundwater treatment. Both alternatives include RO and can now be done at costs comparable to imported river water, which is ultimately what drives adoption.
But treating the ocean via RO is still much more expensive because the ocean is orders of magnitude saltier than wastewater or brackish groundwater. So there is a fundamentally higher osmotic pressure and energy intensity as a first-order problem, and the corrosivity of seawater forces you to use expensive materials which also drives up cost. These (more than regulatory concerns) are why utilities right now who are making decisions about alternative sources are going much more in the direction of reuse and brackish groundwater. SoCal has a couple ocean desal facilities, but we probably won't see much more widespread treatment of the ocean unless/until imported water costs get a *lot* higher than they are right now.
It was one of the first things I noticed in his article, that it jumped to ocean desal rather than the much more economical (and often larger scale achievable) brackish (ground, runoff whatever).
In these discussions it's often noted as a ceiling on the cost of potable water -- there's an infinite amount of it relative to our ability to extract. Other things are often better but our worst case is pretty good! I've definitely run into people many times who insist we're about to run out of "cheap" water.
It all depends on what “cheap” means! Water is incredibly cheap, such that it can still be cheap compared to other commodities but hundreds of times more expensive than it is now, and thus too expensive for current uses.
How much of a cost differential is there caused by altitude and pumping? Desalinated sea water obviously starts at sea level and has to be pumped up to the altitude that the user is at. Reservoir water is already at altitude and pipes/aqueducts that carry it are gravity-powered.
Is that significant against the energy costs of RO, or is it trivial by comparison?
Great question, I heard water pumping is one of the biggest energy users in the state of CA.
I ask this because my uncle was a water supply engineer who worked on some of the early desalination systems in Abu Dhabi and he commented that the pumping was a bigger technical problem than the desalination itself. I don't think that costs were a big factor in those designs based on some of the other things he said.
Sadly, I'm not in a position to ask him myself any more.
The answer to this question varies a lot by geography. If you are (say) the City of Washington, DC you are near the ocean but you also have ample river water available very nearby and conveyance costs for that are very low.
The issue in SoCal is (a) rivers with lots of water in them like the Colorado or Sacramento-San Joaquin system are far away and (b) there are mountains in between you have to pump over. So the conveyance facilities are really something - Metropolitan Water District has 5 pumping plants to get water all the way from Lake Havasu to the SoCal regional system, and they are IIRC the largest such facilities in the world.
"so if a new high-volume user (like a data center) forces you to bring online some much more expensive water source, everyone in your service area has to pay for it."
This is not true. You can just bill the new user a higher rate that covers the extra expense and leave everyone else's rate alone.
Maybe that's not what they CHOOSE to do, but it's definitely what they CAN do.
We have to do something similar at the company I work for. Certain types of new business make us bring in extra more expensive raw materials. We make those new customers pay for those extra costs instead of averaging it out amongst our entire customer base.
Part of the fundamental challenge is that it's usually not economical to build multiple pipe networks - generally speaking everyone has a meter and service line connected to the same distribution system using the same water. Utilities can have different tiers of rates but it's not necessarily straightforward to single out an individual user.
Another challenge is that the volumes being discussed are pretty large compared to the overall capacity of a medium-sized utility. Usually even pretty large commercial or industrial facilities are a drop in the proverbial bucket compared to the utility-wide demand ... but a facility whose needs are on the order of millions of gallons per day (a unit we use for treatment plant size) might require a city to expand its total water supply capacity. So it could well be the case that a utility faces a decision along the lines of "we need a significant new source for this individual user."
None of these are unsolvable problems but they are also non-trivial.
I didn't say anything about building extra pipe networks. This is just an accounting problem.
We have capacity for 100 gallons at $1 a gallon. And we are selling 100 gallons at $1.10. A new customer comes along that needs $10 gallons. To get those new gallons we need to build a desal plant that costs $1.25 a gallon.
You build the desal plant, and charge the new customer $1.35 a gallon, and leave the pricing for everyone else alone. You still ship all the water through the same pipes.
And you can always just create another pricing tier.
Agreed that any time you are taking on a significant new customer you need to do some extra analysis, especially if that requires large Capex spend.
This is basically the kind of calculus a utility would do ... except that the desalination plant would probably pencil out to be much more expensive to both build and operate than other alternatives. So as noted in the OP, a utility around here probably chooses potable reuse or (if an aquifer is handy) brackish groundwater instead. Whenever I've worked on this kind of alternatives analysis those are the winners - I've worked on plenty of reuse and groundwater facilities at this point but the only ocean desalination I've done was a demo project.
The other thing I'd imagine my clients would balk at would be the total demand - one of the linked articles talked about a 9 MGD flow rate which is a medium sized municipal water plant *all by itself* - so the utility would want to be sure it gets paid for the significant capital expense and ongoing O&M and that it can count on a rate structure with this user that can make it work.
Yeah, I was just using a desal plant as an easy example. The point remains the same. If you need new supply to meet new demand, then it's certainly possible to make the new demand pay a higher rate.
Of course you don't have to, but that's a choice.
On the private side sure, but not necessarily on the public side. They can't typically charge one customer more than others in the same class.
"They can't typically charge one customer more than others in the same class."
This is a man made problem though not a real one. It's certainly possible.
"same class" What class do data centers "fall into."
Commercial/industrial. I'll note that I'm a civil engineer, but haven't done much in California, so it could be different there. The utility could also maybe do some wheeling and dealing and get the data center to pay for some stuff up front, but typically the rates wouldn't be set differently than other industrial users.
I'm confused about this because development projects here in NH often pay "impact fees" to cover any infrastructure upgrades that the new project triggers, especially things like new traffic lights, road widening, or water/sewer upgrades. I think it's based on state law. Some towns that want to attract certain types of development will wave fees or be more lenient about them, but I think overall impact fees are pretty common.
Is this not a thing where you are?
It is a thing, but it's the difference between up front costs and on-going rates. IE, they typically need to pay for infrastructure needed for them to connect to the system, but they'd pay the same rates going forward.
Uh, but the raw material here - water - is the same for everyone. The data center just needs a relatively large amount of it. But there's no a priori reason to charge a data center using X million gallons more than some houses, factories or farms using X million gallons.
Unless the political alternative is not to have the data center in the jurisdiction at all.
We pretty routinely charge differential prices for large consumers. Often the prices are lower for massive consumers, sometimes higher. And it's true that water is the same for everyone, but the usage isn't: household drinking water is much more essential to supporting life in some area than data-center usage.
Sure. But in general, the market is a suitable mechanism for allocating goods by their value. If water is scarce and I value it more than a data center does, you'd expect the market clearing price to be at a level that's OK for me but prohibitive for the data center, leading to either the data center investing in technology to lower water usage, or the data center locating somewhere with less water scarcity. A lot of California's water woes are the result of blocking the market from doing its thing and instead letting certain sectors (alfalfa!) use vastly more water for cheap than they should.
The entities with the greatest ability to adopt technological alternatives are, ironically, the ones who have the greatest ability to pay for water (i.e., data centers.) There is certainly a world where all resources are market-priced, and quality of living is much lower for many human beings who live in an area.
If existing supply isn't enough to meet the new demand, and you have to purchase more expensive supply to meet it (say a desal plant) then it makes sense for the new demand to pay higher prices.
Note the higher tier pricing doesn't need to be limited to just 1 new project. You could do the same for new housing development, new farm water quota's etc
Do residential, commercial, industrial and argicultural users all pay the same rate for water? Are there no water usage tiers?
It seems as long as it is physically possible, a municpality can add new water users without dramatically increasing costs to existing users.
It's so broken that we can't even think about prices.
Someone already linked Scott Alexander, but alfalfa, just that crop alone, uses more than 6% of California's water. This is used to produce $860 million (2015 numbers) worth of crop. That's with an "m".
One thing I always wonder about these alfalfa statistics is how appropriate it is to separate alfalfa, rather than price it in with whatever crop it’s being rotated with. Do people actually grow alfalfa as the main use of a field? Or is it all crop rotation to keep the field suitable for other crops, so that it’s appropriate to account its water use to those other crops?
where it's grown, irrigated alfalfa is usually the primary crop. the farmer may rotate to another crop for a few years to alleviate disease pressures, but alfalfa is the goal.
Good to know! I have never seen this explicitly discussed in any of the discussion of alfalfa and irrigation. I knew that alfalfa was a nitrogen fixer, and sometimes used as part of a rotation for that purpose, but if alfalfa is actually the primary crop in these places, then I don't have to worry about this accounting.
Alfalfa is a perennial not an annual crop. So you don't want to be rotating it out all the time.
... is crop rotation still a thing? I thought it was something we did before we had abundant fertilizers that made nitrogenization a solved problem.
Diseases and pests like not having to move to infest the same crops year after year.
They also are rapidly growing resistant to pesticides, fungicides and herbicides.
The war against pests is far from over.
Also I wonder if there are other micronutrients that can replenish with intercropping.
residential/commercial users usually get water from a local govt or water utility that provides treated water from a tap. users are paying for the water, but more they are paying for treatment and the delivery.
don't know much about industrial users, but it's probably mote similar to ag users.
ag users get water from rain (no charge), natural waterways (their costs to pump and distribute), groundwater (same) or manmade waterways, which are variable, but usually there is a fee charged that is supposed to cover amortized costs of the infrastructure construction plus pumping costs.
afaik, a town or water utility that's also getting water from a canal/pipeline will pay the same fees as an ag user of that canal would.
Tiered usage rates are very much a thing. The challenge here is the scale of the new user - generally even a pretty big commercial or industrial facility is a small share of the total water use utility-wide. But the linked article talks about 9 MGD for a data center which is just an eye-popping number for a commercial user and on the scale of a medium-sized municipal facility all on its own.
So if one of my clients were dealing with this in their service area, they wouldn't just straightforwardly apply their existing tiered rate structure because this user is "off the charts" if you will. They would need to make sure they can provide sufficient flow to meet this demand on top of their existing demands; and if they end up needing to add new capacity they'd need to ensure they have a rate structure that can pay for the capex/opex committment.
Cool stuff- when I lived in SoCal I worked near the Carlsbad Desalination Plant. Always wanted to tour it but never did, should've asked the company if they did tours or something. Maybe you could answer a question that I haven't had success with via Google. You mention that seawater is much saltier than other sources of saltwater. Usually extraction costs rise exponentially the more of something you want to extract. How does that affect the cost structure of desalination? On one hand you mention needing different materials, but it seems to me like the extra salt would be easier and cheaper to remove because it's at a higher initial concentration, limiting additional costs. Or would this be flipped on its head because it causes more osmotic pressure against the desired flow across the filter?
The answer is that energy per unit volume treated by RO depends on the osmotic pressure of the water to be treated, which in turn is proportional to salinity. This osmotic pressure is what must be overcome by high-pressure pumps feeding the RO system to drive water through the membranes.
To put some numbers to this: a brackish water RO system might treat groundwater or wastewater in the range of 1000-3000 mg/L total salts. Its feed pressure will be in the 100-200 psi range to overcome the moderate osmotic pressure of brackish water, and its energy use will be in the range of 1-2 kWh/1000 gal treated.
By contrast, seawater has total salts on the order of 30,000 mg/L. This much higher salinity leads to much higher osmotic pressure, much higher RO operating pressures (>500 psi), and much higher energy use (~10 kWh per 1000 gal). That's the fundamental issue
Thanks for taking the time to respond!
I know nothing about this, but I see you haven't gotten an answer. My guess (which I really bet is true) is that, if your goal is to produce salt (and don't care how salty the effluent water is), using ocean water would be easier/more cost-effective/requiring of less energy than using brackish water. But if your goal is to produce fresh water with very little salt in it (which is the goal of desalination), the situation is the exact opposite.
I think you can look at it as seawater having more salt and less water - so easier to extract salt and harder to extract water.
This is true, but I read it just to serves as proof-of-concept that there's no meaningful upper bound on supply once you're willing to pay that price.
There are a *lot* of things we can, should, and maybe even will do before resorting to large scale ocean water desalination in the US. Though, I have little hope of having a coherent water *or* energy policy as long as we remain so unwilling to accurately price things based on their costs.
Southern California has a surprisingly coherent policy of adopting potable reuse and other alternative supplies to make our water supply more sustainable. It helps that we have a regional consortium of utilities in the form of the Metropolitan Water District to organize this effort.
Bottom line is correct - we are already building out quite a lot of reuse and brackish desalination but we're still a long way from the ocean being widely adopted. Of course it could be done if the alternative is no water, but people don't like really high bills!
"reflecting scarcity of the available flows"
and use of water for relatively low value stuff like crop irrigation.
And why can't the new facility bring it's own water desalinization supply using it's own SMR?
Because production SMRs don't exist in the West, and might not exist until the 2030s if ever.
Putting aside the normative question of irrigating a lot of farmland in the arid Southwest, I can say that it will continue to happen as far as the eye can see so that scarcity is a fact of life.
And you make a good point about what it would take for a user to supply a few million gallons per day on their own. They would need to build essentially a municipal scale treatment system AND (if it's an RO system) a pretty large power plant to run it. Not impossible (we did something like this for a semiconductor facility with to treat realt complex wastewater) but it's a massive undertaking.
"The couple replaced most of their appliances in 2019, and then again in 2021 and 2024. Residue now gathers at the bottom of their backyard pool"
I can't read the whole article so maybe I'm missing something, but If your well is prone to drawing in sediment, the first thing you do is install a $30 whole house sediment filter. If you do this, your water pressure will drop when the filter gets clogged with too much sediment. Then you just swap out the replaceable filter. You can't ruin appliances because sediment can't get past the filter. If they had to replace appliances multiple times, they're likely not getting good advice or there's something else going on beyond sediment.
Also, sediment in well water is natural and I'd say it's the norm, not the exception. My house has a sediment filter that I change twice a year (5 minutes, $10). The filter canister ends up with an inch or two of brown/black sludge at the bottom that I need to rinse out with each change. This is normal. My water tests safe and tastes way better than most any muni water. The addition of the data center might have changed the level of the water table and that could theoretically change the exact composition of the water that nearby wells draw in, but so could a change in water usage by the homeowners.
I think there was a case like this in NH after a spring water company came into a town and built a bunch of extraction wells. The solution was that that company agreed to pay to have some nearby residential wells re-drilled to get them to a depth with decent water flows again.
The other thing I'll say about this as a long-time well water user, is that the swimming pool is a bit of a red flag. I'm not sure if the article covers it, but many people with wells need to hire a water company for filling the pool to avoid depleting the well or increasing sediment levels in the water.
Most residential wells are not infinite sources of water under the best of circumstances. I'm not an expert, but usually well drillers only drill deep enough to get a flow rate and water quality that's adequate for typical household daily use. The cost of an 800 foot deep well is nearly twice as much as a 400 foot well, so the goal is to do the minimum drilling that gets a workable result.
I don't have a sprinkler system at my house, but my neighbor installed one, and when they did, that they needed to spend over $10k to drill a second well just for the sprinkler system. Apparently, the installers found that their existing well wouldn't be able handle the additional draw of the sprinkler system.
Matt’s suggestion that we should regulate to prevent this sort of problem seems naive. If every potential $25k harm triggers preemptive regulation, the wheels of commerce would grind to a halt.
But context matters. This is a $750 million project and from the looks of the photo there are probably something like 10 wells within a 1/2 mile radius of the plant.
Working out some sort of deal that lowers the burden of proof for people that have well problems within a short period of the water draw from the data center (basically assume cause->effect without huge burden) would not grind the wheels of commerce to a halt and would likely be a good PR move, as long as they could structure the settlement to avoid an escalation of liability in the future.
It seems that the best way to do this is to allocate each property in the area a certain amount of groundwater they can pump per month or year or whatever, and let them sell these rights to one another. If the data center wants to pump more water than the land they're on is allocated, they can buy water rights from a nearby landowner or buy water from another source.
Just regulate data centers out of existence and you solve the problem AND commerce becomes more healthy
Depends on the cost of the regulation. And of course the regulation is a lt better than just banning the new activity completely.
It does seem like something worth regulating. This is one of many issues where a negative externality hasn't generally been regulated because the usual user creates and experiences a de minimis amount of externality. When a new user suddenly creates a meaningful amount of the externality, or when new users are extremely susceptible to the externality, everyone suddenly has to play catch-up. This is worse than a situation where the problem is regulated up front.
A good example might be noise. In many places, noise is barely regulated; low density and common activities don't generate much noise or suffer much from noise. But if an extremely noisy user wants to move into an area (like an airport), noise can suddenly be an issue. Similarly, airport noise might not be a problem when the airport is in the middle of pasture land, but when the pasture land is developed into housing the airport noise is now a problem.
yeah, it wasn't clear in the article, but it sounded very much like the data center in question had put in a deeper well and was drawing a lot of water, harming nearby residents with shallow wells.
i'm surprised this wasn't a known issue that the data center would have had to agree to remedy as a condition of permits, tbh.
Having googled this case some more, it's less straightforward than it first seemed, because the data center started construction in 2018, but didn't start operation until 2020 or even later. The couple's first round of appliance replacements was in 2019, before operations even started. At first land clearing and site prep was blamed, not water extraction or data center operations.
This story is duplicated all over the place and seems to have become the poster-child story about why data centers are bad. The reporting all around is frankly just plain crap imo. Not one of the articles I've read includes any actual journalism or reporting out of the story. There's very little from any experts on how wells normally work (eg sediment is very normal with well water), or what's been attempted for remediation. There's also no mention of what water tests have been done or what water experts have said about it.
That sounds downright Coasian
Needs more bovine.
Dodged the segue into the hydra of water rights legislation! I'd love to read that full-length post someday though. Can't properly understand Californian history without that throughline. But I will note that even in perennially-experiencing-drought CA, home of hippies and granola, the average person's attitude toward water is still that it is...effectively too cheap to meter. Which makes it all the more disingenuous when people whinge about water usage whenever it becomes rhetorically convenient. An actually empirical "full lifecycle analysis" turns so many such environmental intuitions on their heads. Believe The Science(tm)!
Other fun rebuttals to lab-grown-pearl-clutching about AI water use: do you fret about water every time you google something? No? That's like 1/5th of a ChatGPT. Enjoy "devil nuts", also known as almonds? See this bag full of hundreds of them? Yeah, the water used to grow a single almond is enough for like 50-150 ChatGPTs. One could go on. Water is an essential building block of all life, all modern amenities, and we're victims of our own success that 99% of that cost takes place out of sight and far away now. People care so little that they'll continue to buy bottled water rather than drink Hetch Hetchy tap, some of the best government-provided in the nation, if not the world...(usually can't be bothered to recycle it afterwards either, not that that does anything, but you gotta commit to the bit)
Here in Los Angeles, water costs me two cents per gallon, so about $2/day. Of course, that's for Tier 1 water usage, the lowest, or most efficient, rate. If you're a super water hog and push your usage way up into Tier 4, that will increase your water cost on the margin by 36%. Or a little under three cents per gallon.
To put it in terms of crude oil, there are 43 gallons in a barrel so that’s over $1 a barrel!
Did you include the cost of the deposit on the barrel?
Isn’t the heavy energy use (and, I guess, water) of LLMs in the training phase? I figure once you have your weights, it’s a pretty straightforward process
My reading level on this is no longer cutting-edge (if it ever was), but I believe inference can still be reasonably expensive for frontier models, especially with chain-of-thought and multimodal reasoning. There are a lot of techniques to reduce compute and memory demands for inference but AFAIK they all come with some loss of performance (often much less than you'd expect, although not quite as much less as we hoped when people were writing papers last year about magical 1.58-bit quantization for every model).
Not really sure about Hetch Hetchy (that’s Northern CA, right?), but the water in SoCal is pretty bad. Not Texas or Florida bad, but pretty bad.
That SoCal water has traveled all the way from the Colorado Rockies and picked up a lot of off flavors along the way. San Francisco dammed up a valley in Yosemite way back when governments were allowed to do such things, and the water tastes excellent.
Water in SoCal is hard as a rock! I've got mineral deposits all over my teakettle. :(
You are getting somewhat better at understanding western water issues and yes, pricing is part of the problem but it’s a lot more complicated than that.
But one thing of several to understand is that here in the west we have a water rights system that is fundamentally different from the eastern US. It’s a seniority-based prior appropriation system that, in many cases (particularly the Colorado river basin) allocates more water than usually exists. And it’s usually the case that agricultural users have the senior rights compared to urban users.
And because it is a rights-based system, it’s not just about the prices. The problem for many urban areas in the west is not directly about pricing water to the end user, it’s about acquiring rights to the water in the first place.
If your legal system makes it impossible for you to solve your economic and environmental problems, I think you should change your laws.
These have been treated as property rights with eminent domain protections under the 5th Amendment. So, good luck changing that.
Eminent domain means the state can pay to take the property, no?
Yes, obviously. So the state of CA could buy out the farmer's rights (after a few court cases, one assumes) and reallocate them to suburbanites, or whatever. But that's a more difficult process than just increasing prices to market and doesn't help vis a vis Mexico or Arizona.
no court cases needed. this a totally normal practice.
If there's a dispute over market value, which seems relatively likely, it gets resolved in the courts.
"So the state of CA could buy out the farmer's rights"
This is what I'm in favor of.
Ok, but, if CA did that to the alfalfa farmers, CA would no longer have water problems, right?
Andy thinks the problem is that the state can’t get the rights to the water.
You, seemingly agreeing with him, note that Western states have to pay to take water rights.
But, that’s not really agreeing with Andy’s point I don’t think.
I'd be totally in favor of a federal bailout so California could tell the farmers to f off. I consider this to be one of the biggest governing problems nationally. The largest state in the country has an insane, archaic, and cronyist water system and nobody's really even trying to fix it.
The state can get rights to water and then give them to municipalities, but that doesn’t mean it’s easy.
By way of analogy, it is super simple in theory for the state to acquire land for high speed rail and then build it, but in practice it is extremely difficult.
There is also the issue of interstate compacts and other history and agreements such as the Colorado River compact, which is a water sharing agreement among 7 states, one that overestimated the amount of water would be available, among other practical problems.
Almonds, not alfalfa. Utah’s deal was alfalfa. CA water gets sucked up by almonds
Coase, I summon thee!
Imagine the Anton Chigurh "If the rule you followed brought you to this, of what use was the rule?" meme here, but instead of Anton Chigurh, it's Nobel prize-winning economist James M. Buchanan.
Who is the "you" is the critical question when it comes to the Colorado River. The rights in question here are multi-state and were negotiated in a compact signed in 1922. Changing that is the subject of ongoing negotiations and it's a really tough fight because there are lots of interests competing for diminishing water. Even if California wanted to raise the cost (or lower the share) of water for in-state agriculture, I'm not sure it would have a legal basis for doing that and the farmers could probably successfully sue based on the 1922 compact.
So it's a difficult multi-state issue that needs effective mediation and guidance from the federal government. Luckily we now have an excellent administration in place that will make that a breeze.
Do you have a cite? As far as I can tell, there's nothing stopping CA from doing what they want internally with the water they're allocated on an inter-state basis.
I haven't done a deep dive but apparently it's part of the senior rights of farmers in the Imperial Valley, e.g.: "Imperial’s farmers are protected by longstanding and increasingly controversial senior water rights. Even the largest cities in Southern California and Arizona have water rights junior to Imperial’s, which gives its farmers legal priority in times of scarcity."
https://calmatters.org/environment/2023/01/colorado-river-water/
That article is confusingly written, but the gist seems to be that this is about prices, not some kind of inalienable rights. The state apparently can and does buy the farmers out.
There doesn't seem to be any reason the state could not continue doing this and further reduce agricultural draw, save some water, sell some of the water saved to urban users, and resolve the state's water problems with no net fiscal cost.
Prices reflect exchange, not legal enforcement. It's because they have senior rights that the only way to reduce their share of California's water is to offer them something, in this case a buyout. In other words, they leverage their senior rights to get money.
It is about water rights. In Western water law, water rights are like property rights and get basically all the same protections. Yes, California could buy them, but the cost would be astronomical. Eminent domain is possible but would generate a political firestorm and years of litigation, plust the aforementioned astronomical costs.
There is an active market in water rights sales and transfers in California, although it is not a "free" market because almost everything has to be approved and permitted by California Water Resources Board. So there is undoubtedly some Abundance-relevant friction in the system that could be reduced.
One of my pipe dreams is that the federal government would decide to destroy the Glen Canyon Dam and drain Lake Powell into Lake Mead. This would drastically decrease the amount of evaporation and increase the total amount of water in the Colorado River. But alas, I don't think this will ever happen.
Yeah, I wish the Glen Canyon had never been built. It’s hard to imagine it will ever be anywhere close to full again.
*cue Monkey Wrench Gang reference 💣💣💣
Importantly though, desalinized water would be outside of the prior appropriation water rights scheme.
Yes, but that is expensive and really only helps coastal areas. It’s something that California, in particular, ought to do more of.
"but that is expensive and really only helps coastal areas."
Not really. Because a lot of inland water goes to the coast. So if you save that water on the coast you have more water inland.
Beat me to it. I'm regularly saying that Matt needs to spend some time learning about how much differently water rights operate out here and how big of a deal it is.
I don’t think it’s worth learning. The tldr is that western water rights are extremely inefficient. If you want to optimize within the system, sure, learn the rules, but the point is the rules should be changed.
It's easy to say the rules should change on a forum like this, but it will be extremely difficult to do in practice. Which is why it would be good for a smart, pragmatic person like Matt thinks to consider a way to practically work through the rules.
I'm not sure it's more inefficient than the scheme in the east, where water rights are tied to land rights. It's a different set of tradeoffs. The advantage is that there is so much water in the east that water isn't scarce, so leasing rights isn't a big deal.
I don't think this is true. The idea behind Coasean bargaining is that the initial allocation of rights doesn't matter, as long as the rights are alienable, you can still reach the efficient final allocation via market mechanisms.
As far as I can tell, it is in fact possible to buy and sell water rights. So I don't see why it should matter how the rights work.
Can those senior rights holders not sell their rights if they become valuable enough?
Yes, they can sell their water rights. But if you are a farmer or use those water rights for your agriculture (as is usually the case), then selling the water rights makes the land you work much less valuable at a minimum. It also often means you can no longer engage in agriculture. Here on the front range of Colorado, we've seen old ranches near urban areas sell their water rights, but also sell their land for (sub)urban development. But that isn't possible in most places.
varies by state and location.
in cases where there are restrictions on selling the water or water rights, you can still sell the property, which also confers the water rights.
The actual problem with water use in datacenters, which the NYT article doesn’t discuss but is a real issue specifically impacting that specific datacenter and community, is that they typically do use more water than a similarly sized warehouse or manufacturing facility and the watershed doesn’t adequately plan or augment for their impact. This can take some time to play out because the massive sites like Meta Stanton Springs generally take years to scale up to a significant utilization percentage of their space and power availability, which makes it even harder to accommodate if the water impact is unexpected. Newton County seemed to be caught blindsided about two years ago by how much water was being returned for treatment even though the facility first went live in 2018. At the same time, the industry is quickly adopting closed loop cooling systems, including ones that don’t use water at all (with questionable environmental impact!), and I suspect that in practice that water use won’t be the thing that folks complain about with their local datacenters in a few years.
I don’t understand why the data centers need a continuous stream of treated water.
Presumably, the water is not mixing with anything at the data centers, so why is it preferable to pump treated water in (and pay for the water and sewage fees) vs having water in a closed loop system that pumps it into the ground to cool it?
This definitely seems like a water pricing issue. I am skeptical the water is priced correctly if using that amount of treated water is more economical than utilizing the closed loop system.
The short version is that it's more expensive except for the largest and densest sites, and those sizes and densities didn't exist until the last couple of years, so in practice no one was going to deploy them. As I noted, that's changing and I expect that in the next 2-3 years this will be the norm, not that the exception.
Salt water will rust the metal. Sediment will gum up the pumps.
(It doesn't necessarily need to be potable.)
We do use closed loops, but usually the cooling water is cooled by heat exchange rather than by pumping the water into the ground. The question is, is it cheaper to use power to cool the ground water in a closed loop, or cheaper to just flush the water once it gets too warm and replace it? Often the latter is cheaper, but it will depend on power costs. As you say, its a water pricing issue.
I'm sure they'd already be doing this if the economics made sense, but it would be cool if they mollified the neighbors by offering free or cheap hot water for heating in the winter.
Curious, what is the environmental impact of the closed loop systems?
Not much if it's a water based system, any impact would be related to construction/installation which is minimal relative to the larger facility. My comment was in reference to some of the newer systems which don't use water at all, generally leveraging glycol instead. You don't want that stuff leaching into soil.
Every time I’ve seen rationally pricing water come up it drives a certain kind of left person insane.
I can’t recall the title of it but there’s an anti nestle bottled water documentary that makes it seem like the whole goal of pricing water is the screw poor people out of drinking and bathing water and force them to buy it from big conglomerates and agribusiness who benefit from super cheap water gets at most a by the way mention.
This certain kind of left person is also usually numerically illiterate, the source from which many related problems flow.
> Every time I’ve seen rationally pricing water come up it drives a certain kind of left person insane.
https://nitter.net/WinterPolitics1/status/1947120078606200939
I mean $8 for a bottle of water might set me off too, but the replies are full of people pointing out that Wrigley has free bottle filling stations. Which still get some people mad because someone might die if they don't get water delivered to their seat and poured down their throat for them, because they were taken at gunpoint to a Chicago Cubs game.
Man Disney World is looking judgmentally at Cubs games for having 8 dollar water. Incidentally also free water is available all over the place and yet people choose the worst possible bottled water over it.
The issue is that any way we can stop AI, ingenuous or not, is worth doing. If water works as a way to get them to stop building a data center in your area, use water. If electricity works, use electricity. These places *don’t hire locals* and take more than they contribute to every local economy they interact with, outside of buying off local politicos.
I’m generally against this kind of behavior when it comes to building useful things like housing, rail, even necessary LULUs, but when it comes to data centers, the “we shall fight on the beaches, we shall fight on the landing grounds, we shall fight in the fields and in the streets, we shall fight in the hills; we shall never surrender” approach is completely appropriate. It doesn’t matter how, stop the data centers.
Why are data centers any worse than any number of other businesses that can operate in an area? Is this based on fear of AI superintelligence killing humans or something?
I’m not particularly afraid of Skynet or murderbots swarming America, though maybe I should be, but that’s not something I have any control over. AI only benefits a few people, and they’re usually bad people like Elon Musk, Peter Thiel, and Sam Altman. And it harms basically everyone else. Through job losses, increased surveillance by our rogue government, data harvesting, enshittification, fucking up the education system, etc. We see arguments against AI here routinely, I agree with basically all of them. I’m for cutting the internet backbone most of the time.
At least when a new polluting industry builds a factory in town, it creates local jobs and probably makes a useful product. But a data center at most employs what, 20 people, usually from out of state? While driving up the energy prices for local users and possibly polluting or exhausting the local water? The juice isn’t worth the squeeze. Which is why Elon comes to places like Memphis to build his data centers because it’s too economically depressed and politically torn to fight it. And it puts our aquifer - the one real asset Memphis has - at risk (our water is the best in the US). If there’s a 0.000000000001% chance of a catastrophic risk or a $0.01 increase of a single citizen’s water or electric bill, it’s not worth it to allow a data center, because the costs outweigh the benefits even at those costs! They’re bads, not goods
They have almost no staff. Things aren't bought locally, they ship it all in. They just consume the electricity and water.
That's the argument as I'd make it, not sure I buy it, but an Amazon distribution center hires people and even with lots of robots at worst locals still get their packages faster.
Honestly, this sounds a lot like the usual NIMBYism. Data centers have diffuse benefits and concentrated costs. If a utility district is worried about water or electricity consumption, the right solution is to just raise the prices.
On the contrary, data centers have concentrated benefits (Musk, Altman, Thiel) and diffuse costs (the whole world)
You can stop data centers being built in your area, but you can't stop them being built. The value is going to accrue *somewhere* and to someone. I wouldn't worry too much about Thiel, etc... China is gonna wreck them because they backed a degrowth fascism that is inimical to innovation and progress (which isn't to say that Xi is a nice man).
I mean I’m not a degrowther, but I think AI is growth in name only
We could give an exemption for the first X gallons of water per person or something. The inability of lefties to think through scarcity issues is just baffling.
I don't know who came up with the phrase green on the outside, red on the inside but it is a problem. They don't want to solve the problem unless it means overturning capitalism.
My general impression is that things flow as follows: (1) libertarian types laugh at the left-winger for predicting that pricing will lead to capitalist abuse and monopoly rent-seeking, (2) the abuse and rent-seeking quickly follow, (3) libertarians don't ever discuss it again.
If you have a skeptical view of tech-capitalism in 2025, you're not necessarily correct. But you're also not going to be wrong more than you're right.
Huh? Are we still talking about water?
Most of the US has pricing for water. How many places have "capitalist abuse and monopoly rent-seeking"?
Are you familiar with water pricing in the United States? There is already plenty of rent seeking. Open this market up to newer entrants and markets, by all means, I’m sure it will be as rationally managed as Bitcoin and the nation’s housing stock has been.
Examples?
it's funny because the progressive twitter discussion on water pricing always seems to assume that some entity shoyld be forcing wealthy farmers to pay someone for their water when actual water markets involve someone buying water from farmers with water rights.
That might be true in some places, but its not true in most of California. Few farmers have water rights beyond ground water. Most of the water in the Central Valley is coming from the state via one of the large aqueducts.
that's still a water right, not a market. how much a water user gets is governed by right and the irrigation district charges a fee to cover the cost of infrastructure construction, maintenance and energy use.
the water market is when the rights holder sells their allocation to whoever offers the most money.
Its not a water right in the legal sense. If you have a water right, you can take water from that source without paying anyone. In California everyone has groundwater rights to their property, anyone with a lake or river on their property has limited riparian water rights, but usually when we talk about water rights those are appropriative rights, where people harnessed and diverted water not on their property for beneficial use, often for mining but also for farming, thus gaining the next most senior right to that water source. Water from the CVP and SWP isn’t any of those things - you pay a state agency to take water from their aqueducts. Farmers pay the same way (although a lower amount) than municipalities and industry. Its not really a market - its not like there’s any competition and the price-setting is bureaucratic, but it doesn’t follow the model you describe where municipalities are paying appropriative rights holders. That does happen, but mostly in older mining areas where municipalities use water sources originally developed for hydraulic mining.
I remember some anti-capitalist book about corporate “water wars” in Latin America lol. Same thing.
That said it’s not the lefties who are the problem, it’s the farmers.
Also the subject of the 2008 James Bond film, "Quantum of Solace"!
It seems like it would be relatively straightforward building a coalition around farmers if the lefties were just stock left neoliberals who wanted to make sure everyone could get enough water you know by pricing it reasonably.
“ The bad news is that the Trump administration is moving to allow more mercury, arsenic, and benzene into the air and water”
I don’t know what safe levels of these compounds and elements are, but could it be that we’ve already mandated that and are now regulating just to regulate? I ask because I have had experience with just that.
My career was in the oilfields for the most part. We took what came out of the well and separated it into its constituent parts, generally oil, natural gas and water. The water was oily, and in the old days they just dumped that overboard, which was bad. The government started regulating this. It started with “just don’t create a sheen”. Then they started mandating so many parts per million and kept lowering it past the point of absurdity. All it was doing was costing lots of money; it wasn’t saving a fish or helping a human in any way, unless he sold water treatment equipment.
There is an inertia to regulating, in which the regulation becomes the object, instead of the the safe air, water, or what have you. And we are way over-regulated.
I think this is an overly simplistic view (understandably) from someone who works in a regulated entity about how regulators actually go about making regulatory decisions.
I encourage people to actually check out regulations.gov and actually look at some rules. Someone mentioned mercury. OK, let's look at a mercury rule. This one is on chlor-alkali plants (chlor-alkali plants produce other chemicals using electrolysis):
https://www.regulations.gov/docket/EPA-HQ-OAR-2020-0560
After clicking on the link, click on the "docket documents" to review some of the supporting information to justify the rule. You can also review EPA's comments in response to comments made during the rulemaking phase from industry, other agencies (OMB), etc. Maybe you think this rulemaking is still arbitrary. Fine! But I think it's an uphill battle if you think EPA just invented some arbitrary standard unrelated to cost or health benefits.
Slow Boring commenters are generally pretty good, but they've also internalized "environmental regulations bad" (by which they actually mean NEPA is bad because it's the only environmental law people are broadly familiar with in the comments). It's mildly annoying.
I don’t know if these are good or bad, but the government regulators made bad regulations on PPM of oil in produced water. As technology came out that made it possible to reduce the PPM the regulations tightened and for no good reason. No one (fish, fowl nor human) was getting benefit from this.
If you had a bit more specific info, I could probably find it for you. Above, you mention "don't just create a sheen," which makes me almost certain you're referring to something called the Spill Prevention, Control and Countermeasure Rule. This has supplementary rulemaking, including something called the Discharge of Oil Regulation (aka the "Sheen Rule"). The Sheen Rule is not a regulatory standard; it's a reporting standard.
The Sheen Rule requires requires the person in charge of a facility or vessel responsible for discharging oil that may be "harmful to the public health or welfare" to report the spill to the federal government. This doesn't mean that no sheen is safe, it just means there is no responsibility for reporting. The "just don't create a sheen" standard your employer used was to avoid this reporting.
I've read that the clean air act generally relies on cost benefit anslysis to set pollutant limits, and nuclear regulation generally relies on "as safe as reasonably possible". Where does the clean water act fall on that spectrum?
https://www.epa.gov/laws-regulations/summary-clean-water-act
https://www.epa.gov/laws-regulations/summary-clean-air-act
Yes, the answer can't be lower is always better. Lower costs more. Sometimes that cost is worth it, something it's not.
Trade offs are real. This felt like a cheap shot blanket statement. Maybe true, but I want to see the work.
It depends on the community, but if regulations were increasing the cost of treating water too much, i think water would be more expensive
Great column, and thanks for the link to the very interesting Ritchie piece, but this seems a misstatement: "according to Hannah Ritchie’s calculations, we could desalinate the average American’s annual drinking water needs for $154.45." That number was for total domestic use. The number for actual *drinking* water is just $2.30 per person per year -- shown in the chart you included and stated in her piece.
Quote:
“If the average person in the United States uses 310 litres of water per day for domestic uses, it would cost them $0.42 per day. Or $154 per year.
In the UK, it would be similar: $0.43 per day, and $159 per year. British electricity is more expensive, but we also use less water so the costs balance out.
To get the WHO’s “minimum” domestic supply of 50 litres per day would cost around $0.11 per day and $38 per year.10
Here’s the surprising figure. Producing enough drinking water for someone — assuming 3 litres per day — costs just $2.30 for the entire year. That’s less than the cost of a single bottle of water in many countries.”
Yes I was confused by that too. Matt says that doesn’t include any of the industrial uses, but then what exactly does “domestic” use entail?
I understood Matt to be saying it would be enough water for a household’s direct uses (drinking water, laundry, shower, lawn, etc) but not indirect water uses (hamburger from store).
Sure, and meat is really important. It’s surely not popularist to advocate making meat more expensive.
I’ve done some research into your claim (had pastrami and corned beef for lunch) and I’ve decided that you’re correct
I thought it was clear (maybe incorrectly) that 'domestic' was what is used by all American households per capita for everything: drinking, bathing, washing the lawn, sprinkling the car, cooking.
It's not "total American usage divided by population" which is also important, but people really really fret about their *house* not having enough water.
1135 liters per household per day (per David) lines up with about 200 liters per person per day (per Matt's graph) if households have 5.6 people. That's probably big by 2x, but the numbers are so small that 2x doesn't matter too much.
I don't know for sure, but the EPA fact sheet linked in this sentence by Ritchie "The average American household uses around 1135 litres of water per day," includes discussion of outdoor use, including lawn irrigation. (https://www.epa.gov/watersense/statistics-and-facts)
Well I'm certainly better informed for reading this article. The two (very well presented) graphs were brutally conclusive.
Here in Australia I suspect that water scarcity is more of an issue than the US (especially over here in West Oz). I heard one expert say on a quality radio documentary that perhaps data centres could invest in ultra treated waste water (indeed sewerage) and wouldn't have the yuck factor that prevents it's residential use. He seemed to think it would be economically viable but I don't know how many data centres he'd built.
If there's a well informed reader of this who could tell me about the economics of waste water purification versus desalination I'd be very interested in his, her (or their!) comments.
Just build all the data centers in the Queensland rainforest.
If we build the data centers in Alberta, we can set up district heating and have the waste heat from the processors heat up everyone's house.
I misread this as “Atlanta” and was like, why would you do that, are you a sadist or just evil?
That's a great idea!
There is a strain of left wing folk beliefs that spread like wildfire online. One is housing is expensive because "private equity" is buying up all the houses. It's certainly not community resistance to change. Then you look at the numbers and all corporations including private equity own 500,000 SFHs out of 85,000,000.
Similarly "ai is using all the water!!!" then you look at the numbers - it's 0.2% or whatever.
Two things can simultaneously be true: (1) private equity can own a tiny percentage of SFHs and (2) *in a given market* private equity or other large capital pools intending rentals can be sucking up most of the available inventory and driving up the price of available houses.
But when you hear commentary online do you think most people are referencing a specific location where they know for a fact that happened? Or is it a myth they cling to so they don't have to confront their NIMBYism.
I mean, I’ve read at least one article about it being the case in various markets, and it read as plausible to me. Yeah, I don’t think it’s the case everywhere. Though a lot of people, wrongly in my mind, have an anti-landlord bias, so anyone buying-to-rent or building-to-rent is an issue to people.
Also, that’s the nature of the game with the internet. If town X has this issue the whole world will find out because there’s a NYT article denouncing Blackrock or whatever. Same thing as when one school in San Francisco does a stupid reverse racism or tries to make everyone trans or whatever and the whole world finds out because, internet. The internet makes the discourse toxic
I would simply drink wine. 😆
Beware of Big Wine. I haven't checked on this recently, but in Napa Valley at least a while ago, vineyards were scooping up all the available water and some cities were putting restrictions on how often people could flush their toilets (please don't ask me how they enforced this).
I know for a fact in Paso Robles (also a big wine region in CA) there are huge problems with over pumping and people's wells going dry
https://www.newtimesslo.com/opinion/the-grand-jury-is-in-and-the-paso-basin-is-still-in-trouble-14242788
Yes, too bad for them, and others in the wine producing areas of California, that they're not connected to the State Water Project and have to rely pretty much entirely on local sources of water.
Ironically using water to cool data centers could actually be used as a dual purpose desalinization plant.
Would that not require data centers to be near the coasts, where the land to build their large footprint is significantly more expensive?
There are closed loop water and non-water cooling solutions and design now that this is emerging is starting to tackle the somewhat lazy/naive earlier designs of Cool Everything
Not necessarily, take for example the great salt lake. There exists a lot of nonpotable water inland.
Isn’t that lake drying up creating a significant air pollution and other problems?
https://attheu.utah.edu/facultystaff/utah-is-making-progress-on-great-salt-lake/
Recycling water (reuse). and closed systems, it doesn't disappear although it can become non-potable.
The Agri usages are what is seeing such things dry up.
Evaportion.
Wouldn’t heated water exacerbate the evaporation problem or are cooling towers used?
I was merely pointing out a large body of nonpotable water. There's a lot of nonpotable water around. It accounts for more than 90 percent of all water.
Damn agriculture draining tributaries
Good point.
I don’t think you could do this cost effectively. Waste heat is difficult to manage, I don’t think data centers heat up enough to phase change water etc
Right. The residual heat removal from running a data center at design temperature doesn't produce the heat required for phase change. There is a large up-front water investment in data centers (depending on the design) but in a closed-loop system, there isn't much in the way of water usage - that water is constantly absorbing heat and then rejecting that heat, but under normal operating conditions, there isn't much in the way of lost water. As mentioned, the real cost comes from energy usage.
Even if your not boiling the water having water that's 30 degrees warmer than ambient will require less energy to bring to a boil.
Thought this was an Eddington review at first lol
All of America's water problems seem to be caused by growing Alfalfa in the wrong places.
https://slatestarcodex.com/2015/05/11/california-water-you-doing/
Why does it seem like all the major American AI companies are trying to brute-force their way to the most energy-intensive models and none are trying to develop ones that can run on less energy like DeepSeek did so that for example people in Africa where they have very limited energy generation can run it on their phones? Tech companies 10 years ago at least tried to virtue signal that they were going to make things accessible to the poor and close the “digital divide” so this seems to be another example of a vibe-shift towards a more selfish culture.
It’s also concerning how this increase in data centers is going to increase data storage and thus surveillance capacity, especially as people start relying on cloud AI for more and more personal stuff. In addition to working on cheaper models that can run locally, we should have a GDPR-like law in the US that limits what data can be retained and guarantees people the right to delete their data.
You don’t run AI on your phone. You use your phone to access the cloud. Any good AI takes at least 10 connected GPUs acting in parallel to work.
Startups would love to crater AI training costs. No one has figured out how to do that yet, but not for lack of trying.
Even so there’s a difference between a world where good AI is cheap enough to run on relatively small clouds that are widely dispersed across many countries and companies versus one where they can only be run on large clouds that are all controlled by a handful of US tech companies.
My understanding is AI can run on less than a million bucks in hardware — totally accessible to medium sized businesses and dedicated startups—. and the real constraint is getting the weights and IP.
The main labs are focused on developing agents (on the way to AGI), which is an energy arms race. Along the way, small, efficient models have come out of these efforts. Google makes Gemma; Anthropic makes Haiku; OpenAI makes a series of mini models; Meta has small models as well. Then you have labs focused on this, like AliBaba, Kimi, and Mistral. Together, all of them have made AI free to anyone with an internet connection, and the quality of the free option has been getting better continuously.
You mention DeepSeek, but that release actually did lead to drastic cost reduction efforts at the big labs. o3, the best model by my lights, has gotten very cheap compared to six months ago. That's a direct result of increased energy efficiency in response to competitive pressures.
Alibaba and Kimi are Chinese and Mistral is European. US companies don’t seem interested in doing this and if they do it incidentally it’s due to foreign competition. One reason US attempts to prevent diffusion of AI would actually be extremely bad for humanity if they succeed.
I'm not sure what 'interested in doing this' should look like if it's not models like Gemini 2.5 Flash or o4-mini. Granted, what you're talking about isn't the main object of these labs--but it wouldn't be the main object of those other non-US labs if they had the time and compute to compete in the main race. Competition gets the outcomes you want from the American labs, so I don't understand the concerns in this context.
Now if you're making a point about the US government's role, I think that's a different discussion and they could muddy what's been a very impressive global diffusion process for this technology.
Your understanding is on the order of Evil Oil Corporations Conspiracy to prevent 1000 mpg cars...
Phones on 3,4 or 5g bandwith are no where near capable of AI platforming (and neither do such people hae the energy access)
This is like on the order of "why don't we shoot our garbage into the sun" - it's "not even wrong."
AI platforming centers are getting more energy efficient as costs on energy explode and so initially lazy designs are rapidly getting evolved. However industrial scale physical infra innovation doesn't occur at internet commentary speed.
The AI services are currently massively losing money. They are planning on scaling up based on the current consumption patterns and growth holding. But to make that all a going concern, they need it to be widely used and profitable. We are at the point "then a miracle occurs" stage. As that miracle continues to not happen, much of the planned expansion is going to slowly evaporate. My need for agents and generative AI is close to zero and will be zero if I have to use a cloud provider.
My bet is on Google for what you're talking about. They pulled it off with TPUs and Google Translate - they invested a huge amount of resources upfront to make the dollars per query as low as possible, so that they could release it for free. So efficiency is a little bit in their corporate DNA.
Nobody has focused heavily on monetizing "good enough" models, because they believe spending effort on this will be obsoleted by wildly-improved near-future models. So far they haven't exactly been wrong about this. So everyone is racing to find out where the ceiling is, rather than optimizing for anything else. At some point (I presume?) there will either be a ceiling or we'll get the singularity, and this will change.
There are lots of cheaper models you can run locally. But by virtue of being smaller (which is what makes them cheaper) they aren't as good as the big models. But both of them get better with time -- the small models of today are better than the big models from 3 years ago.
It’s not ignoring ECON101. Societies operate by human consensus. Our wiring is such that there is no way 100% of our relations with fellow humans will be operated under market principles. History has shown that 0% is also the wrong answer. So various societies have experimented with various allocations. US has an unusually high share of human interactions governed under market rules. Europe has less (helpful framing for both a “rude” or just not subservient waiter as well as a guy who helps you put your car bumper back vs asking for your insurance info).
But even in the US it is awfully hard to get people to put life’s basic necessities into the category of market exchanges. So we might have to find a construct that produces a more optimal outcome without ECON101 re education camps.
"US has an unusually high share of human interactions governed under market rules."
But, unfortunately, not CO2 emissions, not traffic congestion, not (most) water use, not (most) parking, not air and water pollution.
And not how Medicare decides how much physicians should be paid for a long list of procedures and services.
https://www.nytimes.com/2025/07/21/upshot/primary-care-doctors-salaries.html?smid=nytcore-android-share
It’s true, market prices for water would stop you from allocating enormous amounts of water to growing alfalfa in the desert.
But even if you don’t like markets, I’m curious why your society can’t reach a non-market consensus that you should not allocate enormous amounts of water to growing alfalfa in the desert, as you are draining your aquifers dry and emptying your rivers? Most of you are not alfalfa farmers.
It's all about antiquated water rights in the west and no politicians have the balls to do something about it. Instead they put the squeeze on residents to curtail water usage which is lame because (in CA anyway) all of the residential/industrial/parks/golf course only accounts for 20 percent of water allocation. Cattle, directly and indirectly, account for about 50 percent(!) and don't add much to the state's economy. Almonds are popular to demonize but they are only 10 percent.
One solution may be to stop the unnecessary corn ethanol production and grow cattle feed in those areas which have much more water.