How datacenters use water – and why kicking the habit is nearly impossible The explosive growth of datacenters that followed ChatGPT's debut in 2022 has shone a spotlight on the environmental impact of these power-hungry facilities. But it's not just power we have to worry about. These facilities are capable of sucking down prodigious quantities of water. In the US, datacenters can consume anywhere …
looks out of window
what do I see
I see the SEA :o)
and I know MS were trialling putting small scale servers into an underwater containment system
just wondering if there will come a time, as in NOW, when they seriously need to consider putting the next gen data barns, if not UNDER water, at the very least right next door
we had Trawsfynydd nuclear power, sat on a lake, it got the water warm enough that the 'local' population of goldfish was the thing on myth and legend :o)
but it was a self contained coolant system, surely there has to be a way to utilise the biggest heat sink on the planet ?
The "biggest heat sink on the planet" has a finely balanced ecosystem contained within it. Together with global warming, which already has deleterious effects on marine ecosystems (read up on coral bleaching), it will only get worse if you suddenly decide to place every new data centre within pumping distance of the ocean. The marine scientists and environmental activists are already very concerned about how the rise in global ocean temperatures is faster than what corals and other vital marine life forms can adjust to.
Cut back on the shyte that is "AI" (there is nothing intelligent about using vast computing power to 'generate' imagery, art, etc), and cut back on general power consumption and let's talk again.
On the power generation side, this is also only going to get worse regardless of whether coal/gas is replaced with nuclear or not, because nuclear has similar water requirements on the cooling end as coal, oil/gas or biomass do.
This post has been deleted by its author
[NB. climate change deniers: don't bother posting.]
What, and let the reality deniers have all the fun?
I cannot come to the conclusion that such efforts are anything less than crimes against humanity.
Most sane people wouldn't come to that conclusion either, but then reality deniers may already be well ahead on the self-trepanation. I'd have to agree that much of the current 'AI' applications are, however much closer to crimes against humanity.
But the OP is a tad misiinformed, which isn't that unusual. This blue-green ball we sail the oceans of time and space upon have.. rather a lot of water. The article makes the point that water is rather useful for cooling, and we've known this for much of our history. Then again, the article also shows progress is perhaps moving in mysterious ways "the rest is used to flush out mineral deposits similar to those found when cleaning a humidifier.. Don't IT people use kettles any more?
But I digress. Problem is although there's a lot of water, much of it is in the wrong place. Fine if your DC is on the coast (rather than built on a swamp), less fine if inland in an arid location. Then it's much like the DC energy problem, cost. Most DC operators don't want to pay a sustainable price to build the resources they consume, and many don't have to when politicians are hooked on the idea that 'AI' will somehow create jobs. They can bring their state/county 'billions in investment', at a cost of their natural resources and energy.
So DCs could be built closer to water, but then we like living near the water, so land costs tend to be higher. Then as the article says, desalination or demineralisation adds cost in both a direct financial sense, and energy. On the plus side, DC operators could diversify and sell 'sustainable' sea salt. Remnants of ancient salt pans can be found along many coastlines from the days when salt cost more than an AI offload processor. But it's also why evaporation and condensors aren't used as often. Why bother when you've negotiated a cheap deal for water? Politicians could do more in a 'polluter pays' to make bit-barn builders pay more of the costs, but most are clueless and dazzled by the snake-oil from the AI promoters.
And then of course there's the general misinformation around 'warming the oceans', coral bleaching etc. Again, there's a lot of water so DC's heat contribution is much the same as peeing in the Pacific. On which point, see-
https://www.nature.com/articles/s43247-024-01683-2
Decreasing trend in destructive potential of tropical cyclones in the South Indian Ocean since the mid-1990s
...a significant decrease in power dissipation index has been detected in the South Indian Ocean basin since 1994, which is almost entirely due to a decrease in both tropical cyclone frequency and duration in this basin. The decrease in tropical cyclone frequency is influenced by increased atmospheric stability.
And it's much the same for the Atlantic & hurricane activity. Given both cyclones and hurricanes are heat transfer events from oceans to atmosphere, this isn't exactly the kind of result predicted by the reality deniers, or evidence of a slow boiling ocean. Then again, it's but a drop in the ocean. Data since 1994 makes for a standard 30yr 'climate' interval, but any changes over the last few years don't necessarily indicate any trend, other than one that's negative to the dogma.
That article about reduced tropical cyclone activity in the South Indian Ocean is a very cherry picked example. The paper itself says it is the only region in the world to show such a trend. In fact, while that particular measure of cyclone activity is lower now in the South Indian Ocean than it was in 1995, it's also been on the increase since about 2010.
It doesn't matter where you put the damn things, if they exist and use energy, they produce heat. We're a fairly closed system the only way to actually get rid of energy is radiation into space. But before we go all tree-huggy the energy we produce with hydro-carbons is energy that fell on earth from space (the sun) but was stored. That is if you believe oil, gas & coal came from trees which may not actually be the case. But as most believe that, let's go with it. So what is the answer? It IS to be more tree huggy and capture it again in vegetation. So why don't we do that, it would make a nicer environment. In fact it is happening naturally through raised CO2 levels, things are greening, we just need to restrain from chopping it all down and maybe help it along a little. Nuclear is cited as a solution but that could actually become a bigger problem because the cycle time for that energy is huge, in billions of years and its possible replenishment involves the destruction of the current solar system!
Forget all the NetZero BS that's coming from people who want to create a crisis to seize more power and control. CO2 is plant food, they love it and give us O2 in return. The level of CO2 is minute (0.04%), it's warming effect is trivial compared to water. There has been much higher levels of CO2 which tends to follow warming and life was not extinguished. Climate change is nearly all from our orbital characteristics, possible from changes in the core too due to it being molten.
>That is if you believe oil, gas & coal came from trees which may not actually be the case.
Where else would they come from? There's exactly three ultimate sources of energy on Earth: the Sun, residual gravitational potential energy from the Earth's formation, and nuclear energy from uranium deposits.
I see no sane way to argue that hydrocarbons come from high pressure alone or from radioactive decay, which leaves the Sun, which means plants. Probably with several additional steps, but ultimately plants. There's some organisms that can feed on the other two sources, but they're marginal.
Genuinely curious here, as the only other alternatives I can think of would be God, aliens, or not knowing what hydrocarbons are.
Genuinely curious here, as the only other alternatives I can think of would be God, aliens, or not knowing what hydrocarbons are.
Ah, well.. CHON! All of which can be found in rocks, and hydrocarbons can be synthesised in labs and chemical plants from the raw elements. Which leads to this-
https://en.wikipedia.org/wiki/Abiogenic_petroleum_origin
Mainstream theories about the formation of hydrocarbons on earth point to an origin from the decomposition of long-dead organisms, though the existence of hydrocarbons on extraterrestrial bodies like Saturn's moon Titan indicates that hydrocarbons are sometimes naturally produced by inorganic means.
So some combination of temperature & pressure leads to inorganic formation of hydrocarbons. The chemistry seems plausible, especially if there's abundant hydrocarbons on Titan. Unless of course Titans were also formed from dinosaur juice, like ours wasn't. Or mostly wasn't. Coal is easier to give a natural explanation for, especially when as a kid we'd find lumps of coal with fossil imprints occasionally. It's one of those interesting but controversial scientific theories, wrapped up in a bunch of politics and activism. So concepts like 'peak oil' may be pushed further into the distance, if oil and gas can be produced naturally. But then it's a timescale thing, ie oil & gas wells that have run dry should start to refill, if it's abiogenic.. Or refill very slowly if it takes time for oil & gas to be formed deep underground and then seep up to where it collects in the fields we've discovered.
The "biggest heat sink on the planet" has a finely balanced ecosystem contained within it.
Indeed. There are places that play with it - there is a district heat scheme in Copenhagen which uses the harbour as a water-source heat pump - extract heat in winter and cooling in summer. However, historic harbours are not known for their biodiversity;1 it's relatively small scale and they take heat back out in winter (of course tides and the mixing of water means this is not quite equivalent to storing heat in boreholes, nor environmentally benign, but they get away with it at small scale).
If we decided that we should put dozens of new AI datacentres east of London and dump gigawatts of heat into the Thames, it would rapidly degrade the marine and coastal habitats, which would really give Packham something to complain about (weird that he gets so het up about HS2 when the Lower Thames Crossing damages more woodland in it's 20miles than the full HS2 Y-Scheme, and impacts RSPB wetlands as well! Yet not a peep).
There's a very simple answer here - fuck AI. Silicon continues to develop at a pace. Watts/instruction are at an all-time-low. A 2020 6W chip will outperform a 95W chip from 2011. It seems hard to believe that genuine compute demand should consume more power (or water) than it did 10 years ago. At least not in Europe or the US. Our populations have not grown by an order of magnitude since then. There are many old telecoms buildings which have pivoted into data operations because as the electro-mechanical exchanges went solid-state, the floor-space and electrical requirements shrunk to a fraction of their former size, and they've let that space out for servers. Undoubtedly, useful compute - messaging, email, file hosting, webhosting - have all gone much the same way. Moore's Law has seen enormous growth in what you can do per U, along with growth in storage tech. Aside from on-prem data rooms going into "the cloud", much of the net demand for extra rack space seems to be analytics compute (of greater or lesser usefulness), and then the crypto and AI grifts.
Much of this amounts to make-work from an industry wedded to the concept of infinite growth, and then fads like AI which consume unfathomable resource for no real benefit. Christ, if you could point one of OpenAI's training clusters at Folding@Home for a couple of days you'd flatten years of crowd-sourced compute - and achieve something useful, like help cure cancer (same goes for the crypto clusters which were/are soaking up GPUs for years).
1. More the layers of pollution - oil and sludge on the bed. Go take some cores out on Southampton Water and off Fawley refinery in particular. A fun morning of poking stinking black anoxic layers of tar, compared with the marginally less smelly brown oxic layers you get from a healthy estuary.
If we decided that we should put dozens of new AI datacentres east of London and dump gigawatts of heat into the Thames, it would rapidly degrade the marine and coastal habitats, which would really give Packham something to complain about..
Packham will complain for cash. But 'we' have been delegated to TPTB and they should be advised by real environmental experts, who point out that this would be a bad idea. But they may get overruled by useless pillocks like Starmer who think the UK economy will be saved by AI. So any plans to dump millions of gallons of warm water into the Thames should get rejected at the planning stage, and probably would. I hope. So when there's a planning application for a new bit-barn, it should state energy & water consumption, heat production and ways to address those issues.
If it needs 50MW of electricity, it should show how it's adding 50MW of electricity generating capacity. If it needs 5 million gallons of water, and dump 10MW of heat, it should demonstrate how it's producing that water, and managing the heat. It's up to the developer how it does this, so desalination or whatever. It might do what other industries do and build cooling ponds or lakes. Developers could turn those into water features. So think of it as a 'Net Zero' or 'sustainable' approach.
Problem is bit-barns are often parasitic loads on both energy and water supplies, and the costs are socialised. Of course if developers had to pay their costs, bit-barns may not get built, especially as the economics around 'AI' are very much tulip shaped. But it's a solvable problem, ie create sustainable business parks around a DC, SMR power house and water features. Some waste heat could go towards heating office space, or even residential development. Call it New Slough or something.
I didn't object to almost any of your post, but you're getting a downvote for the last sentence.....
Ok, so perhaps it's not the best name. But could run a competition to name the new town, or sell naming rights. But I think it's a useful template for what not to do, or how we could do things better. I think there's also some historical precedent, ie the towns built by our industrialist ancestors for their workforce. But it's one of those things where some central planning may benefit. So look at the planning applications currently in the pipeline, available land in the government's land bank and tell the developers where they can build their bit-barns. Then based on demand for utilities and infrastructure, how much they'll be expected to contribute towards those.
If you figure on the balance sheets of the usual suspects, ie Amazon, Google, Microsoft etc they could afford to do this out of petty cash and could collectively get benefits by collaborating on infrastructure, services, and even providing attractive, sustainable housing to their workforce.
Likewise all forms of medical imagery that return vague and imprecise images - x-rays, CT scans, ultrasound. There is quite a lot that AI might be able to do for us, but I think current investor and techbro expectations on benefits are at least two orders of magnitude beyond what we'll eventually see. The techbro will drift on to the next gig, or surf the next fadwave, and a lot of investors will find that their hope and money were misplaced.
"The 'biggest heat sink on the planet' has a finely balanced ecosystem contained within it. ...it will only get worse if you suddenly decide to place every new data centre within pumping distance of the ocean"
True, but how much worse?
From the International Energy Agency's website*, total DC energy consumption was about 300TWh in 2022.
According to the US National Oceanic and Atmospheric Administration, there is about 1.3x21kg of water in earth's oceans.
The specific heat capacity of water is 4200 J/kg/K. So if all the DC energy consumption was turned into heat and dumped into the oceans (and spread about evenly), this equates to an annual temperature increase of about 66 nanoKelvin, which I suspect is not even measurable. Even if DC energy usage increased by an order of magnitude, it would still take about 1.5 million years to increase the ocean temperature by 1 Kelvin (or 1 degree centigrade if you prefer).
Of course, the heat wouldn't be distributed evenly and some areas of the ocean would be warmed a lot more than this.
I think the solution would be to stop building DCs in areas where their existence places stress on local water supplies in the first place. Nothing is impact free though.
*https://www.iea.org/energy-system/buildings/data-centres-and-data-transmission-networks
In Toronto the district heating system also provides cold water for air conditioning in summer. The municipal water system draws water from several kilometres offshore deep in Lake Ontario where the water stays cool (4 C) year round. This water is then run through heat exchangers to cool the water in the district air conditioning loops. This then provides air conditioning to several hundred buildings, saving about 75 per cent of the electricity which would otherwise be required for air conditioning. The water being used for cooling is already being drawn for municipal water supply anyway, so there's no additional water being used.
Toronto also gets the majority of its electricity from nuclear power, with more plants being built close by. The nuclear power plants also draw their cooling water from either Lake Ontario or Lake Huron. The Great Lakes are large enough that the amount of heat being discharged into them is insignificant compared to their size. In winter only a very small area outside of the cooling discharge outlet doesn't freeze, so it's easy to see that the amount of heat involved may be large on a human scale but is very small on a geographic scale.
If data centres need lots of electricity and cooling water then they need to start locating in places where these resources are abundant and stop building in places where they are lacking. As can be seen with Toronto (natural cooling water and nuclear power), there are solutions and they are practical even if they are foreign to Silicon Valley.
For you guys that are running small or edge Datacenters and are using DX cooling, you may want to look at the humidity in your DC to save a bit of money and energy.
The higher the humidity, the better the heat transfer from both the servers and the air conditioners.
From my experience, 20ton unit runs at 75% @ 20% humidity or 50% @ 45% humidity with each instance at 72deg room temperature.
When you initially increase the humidity you will see an increase in the utilization as the heat transfer improves.
"According to researchers at UC Riverside and the University of Texas Arlington, by 2027 global AI demand could account for the withdrawal of 4.2-6.6 billion cubic meters of water annually. "
That's predicated on increases continuing as they have done. What might be just as likely or even more so is plenty of these ventures going bust as the market saturates and the shiny has worn off of the party tricks.
The article is suffering from an Elonism, computing measured in Watts. That's very silly to say "XX Megawatts of compute". That's a meaningless phrase or a something that can mean many different things.
"XX Megawatts of compute". That's a meaningless phrase or a something that can mean many different things.
It's just following an old political tradition - measure the inputs rather than the putputs.
"We spend x% more per year on Y than Z did when they were in power." It should be an unequivocal statement and the only response for Z is to promise to spend even more. But it's a measure of input. Producing measures of outputs to prove value for money is harder so doesn't get talked about because Z would challenge it immediately.
Dealing in terms of inputs is even more important if you're in the habit of taking money form investors and haven't really anything at all to show for it.
""We spend x% more per year on Y than Z did when they were in power." I"
In horse medicine, the common prescription is to shoot the horse.
In politics, the common solution to any problem (real or imaginary) is to throw money at it. The important thing is to "be seen doing something" and money spent is how they measure that.
In politics, the common solution to any problem (real or imaginary) is to throw money at it.
Unless there's some risk of it achieving some useful objective - like building public transport, improving school facilities, or ensuring that teacher's pay keeps up with inflation (or paying them a living wage to start with, and providing classroom supplies - instead of expecting teachers to buy supplies out their salary, which is apparently a thing in some places!?!?).
If it's something useful, then it all sucking of teeth and "difficult decisions". But if Lyle Lanley comes along with a slick presentation about hyperloop, then suddenly there's hundreds of millions to spend on feasability studies when what cities actually need are some trams and buses!
"Unless there's some risk of it achieving some useful objective"
The only and singular objective of a politician is to remain in politics. They things you mention are awfully nice, but if they don't buy enough votes, they aren't useful to the money spenders. Who needs public transportation when one has a chauffeur and a parking space with one's name on it behind secure fencing?
"It could be worse. On the left side of the pond, big water is often measured in acre feet."
In the US, 100 years is a long time and in the UK, 100 miles is a long distance.
My water usage is measured in cubic feet since an acre/foot would have me sitting on the roof soaking my feet.
My water usage is measured in cubic feet since an acre/foot would have me sitting on the roof soaking my feet.
That's what I tend to think when I hear that unit. Also a measurement that makes sense given the origins of dividing up water rights for agriculture. Want to irrigate an acre with 1 inch of water a month? Easy to figure out water requirements, or how that would make water sources deplete. So..
In the US, datacenters can consume anywhere between 300,000 and four million gallons of water a day to keep the compute housed within them cool,
Is 1 to 12.3 acre-feet per day. Then the US's biggest reservoir is-
https://en.wikipedia.org/wiki/Lake_Mead
contains 28.23 million acre-feet (34.82 km3) of water.
When full, but due to a combination of drought, population growth and a desire to produce almond milk in a desert had dropped to only 7.5m acre-feet in 2022. WIth DCs using up to 4,500 acre-feet a year, it gets kinda easier to visualise the impact and what will happen if water demand from DC's exceeds the recharge rate of water sources like Lake Mead and other aquifers.
This approach may also not be appropriate for latency-sensitive workloads, like AI inferencing, where proximity to users is imperative for real-time data processing.
Thats old thinking. Latency is the new Genius. Delayed gratification is worth more. $2000/mo, in fact, for "Pro" reasoning ability.
Latency is almost totally irrelevant for AI inferencing, it's nothing like real-time.
An extra 50-100ms is completely invisible for all of these purposes, and is enough time to put the datacentre anywhere.
There are very few tasks where real-time is actually important, and even fewer where <16ms matters.
Yes there are some rivers and lakes in drought areas that may run low or even go dry, but there are plenty that never do. And obviously the ocean never does. Yes you might need to lightly treat the water to remove sediment/salt but that's cheaper than cooling solutions don't use water.
This whole thing is overblown. The mention of 10 gpm for a 1.5 megawatt datacenter really brings it into focus. That's like 4 people taking a shower. How many people do you think are simultaneously showering in the city where you live on average - and that's MORE expensive because the water runs down the drain and has to be treated and discharged. There are a lot of industries that use several orders of magnitude more water than that in a single factory and no one ever seemed to be worried about that.
I am starting to think that all the publicity around datacenters installing massive solar fields and other alternative energy sources have made the greens try to find another angle to get the public to be against datacenters. Especially with all the ire directed at "AI" datacenters of late. AI may or may not ultimately turn out to be fool's gold, but it isn't as if the massive datacenters to support people's social media addictions, becoming part of their couch as they binge endless schlock on Netflix or play the 10,000th iteration of the same tired FPS shooter online are any better uses of that electricity or water.
Part of the Columbia River issue is the water discharged by the data center is warmer, sometimes significantly warmer than the water that it intakes and than the water from municipal plants. The warmer water is detrimental to fish and aquatic life.
I don't understand this. If it is using evaporative cooling then there is no "discharge", the water is evaporated. This must be using cooling loops with direct heat exchange, which is a different technology. That would result in warmer water, but if they were located near an industrial customer that needs a lot of hot water they'd love getting prewarmed water to reduce their costs.
Maybe someone should put a laundry for a major hospital next door to it. The laundry for the major hospital where I live is located about five miles away from it so it doesn't have to have the whole hospital there, just close enough that the savings in heating the water more than pays for any incremental transportation cost for the laundry. Hospitals are always needing more space anyway, being able to repurpose an on site laundry facility for more beds/ORs whatever without having to buy more land could be a huge win (that's why the laundry for my hospital is located off site)
That’s not how evaporative cooling works in practice. *Some* water is evaporated into the atmosphere. Some is not. Some recondenses within the evaporator.
Assuming you can avoid the dogs/angry men go stand under a power station cooling tower and tell me how dry you are 30 seconds later…
Can’t speak for the US, but the UK has a 2C limit on cooling discharge net temperature for rivers. That is, whatever you pump back into the river can be no more than 2C warmer than what you pumped out. If you get the river too warm, algae gets into its groove, then wipes out pretty much all aerobic life in the river when it dies off. It’s not pretty.
If you get the river too warm, algae gets into its groove, then wipes out pretty much all aerobic life in the river when it dies off. It’s not pretty.
Algae is bad. Loch Neagh in Northern Ireland (the UK's largest freshwater body, which provides 40% of fresh water in NI) has now basically undergone total ecosystem collapse.
In 2023 the fly went, the eels are dead, blue-green algae rendered the lake largely anaerobic, and a year later things are looking no better.
The Loch Neagh collapse is not specifically due to warm industrial outputs.1 Nonetheless, it is a very visible reminder of what environmental science has been warning of for decades.
1. A variety of factors are at play, including general water management, sewage discharge and some agricultural runoff - not that the latter is as bad as often made out, because fertilizers are f-ing expensive these days and farmers don't apply more than they need - not like the days of arbitrarily spreading manure both as fertiliser but also to get rid of it! In 2022 Nitrogen spiked from £200/tonne to >£1000/t. Ukraine makes a lot of fertiliser.
So maybe they need a retention pond to hold that "runoff" before it is delivered into the body of water to give it time to cool down? If you make it fairly deep and pump water out the bottom of it instead of letting it flow off the top (which will be where the warmest water is) it seems like they should avoid that. Or are they already doing this and despite those measures the water is still coming out too warm?
If it is taking MUNICIPAL water then I can see the concern. The town is paying to treat and make drinkable water that does not need to treated let alone drinkable. I assume the datacenter may be "along" the river but not close enough to it for piping water directly from the river to be feasible?
Two things
- It is exceptionally unusual to use wholly untreated water for cooling systems, because scaling, silt and biologicals can eventually cause serious problems. Even old style power station cooling towers usually had pretty thorough water treatment systems
- Even if the cooling system takes its own water from lake or river, that usually reduces abstraction volumes possible elsewhere for municipal use.
DS999,
Legionella bacteria is present in something like 70% of fresh water. If that fresh water is also warm - then it's also going to be able to multiply - it likes it between 20-40°C. In the water it's pretty safe. My last Legionella course (20 years ago) said that it wasn't thought likely you could get infected just from drinking it. The way to infect people is to aerosolise it, particularly in a cooling tower.
You can of course pre-treat your water, but it may be cheaper to just buy from the local water company - who treat it more efficiently, at scale. If you are going to pre-treat river water, you'll need to filter silt and particles floating in it, and then you can either chlorinate it, or use UV. However industrial water filters need about 20-30% of the water that they've filtered to be used in order to backflush them - so you also need to reserve a decent amount of space for water storage - and of course your system will require regular maintenance.
Cooling tower water can become highly concentrated and extremely alkaline. All the calcium and magnesium salts build up in it. It corrodes the storage ponds - and the combination of uneven surfaces and the scale build-up can create nice hiding places where bacteria can start to breed - which bleach can't easily reach. Which is why you don't want to get Legionella into your cooling ponds.
"Yes there are some rivers and lakes in drought areas that may run low or even go dry,"
That has been a big issue in Germany. A secondary problem is raising the temperature of the water leading to an algae bloom or a destructive species taking over.
Burning loads more coal, CH4, Uranium or using dodge alternative energy to heat the air and water for entertainment is squandering precious resources.
Sort of, and we know what you mean, but an evaporative cooler that is cooling the air is not transferring heat to the air.
(Technically, evaporators transfer water vapor to the air. Water vapor can be hot -- live steam -- or cold -- evaporative cooling)
Maybe we can get the price of Fluorinert even lower like Seymour Cray did when he cooled the CRAY-1. Then immersion technology might prove helpful for cooling. Kinda like filling your car's radiator with "coolant". I suppose soon we will have coolant pipes running in & out of server racks.
Of course I miss my air-cooled Porsche.
High school chemistry taught me water can exist in three states: solid, liquid, or gas. So a data center converts liquid water in to steam. How is that "consumption" of water? It is just in a different form.
Sure it is converted but it doesn't cease to exist. The article compares oil to water. Oil is certainly consumed when used (burned), but water is not. Bad choice for an analogy.
So it really comes back to where is the liquid water coming from, and where does it go in a gaseous state? Well it turns to precipitation and falls back as rain....which goes right back in the fresh water supply.
If anything, data centers evaporating less-than-pure water are helping the environment by turning it in to distilled water and returning it to the fresh water supply.
> which goes right back in the fresh water supply
Actually, no, it doesn't. The rain falls in the ocean and on the land, and then it's no longer fresh water.
That is, if it rains, which in a big chunk of the SW US, it doesn't.
If you actually read to the end of the article, you'll see a quote:
Planet Earth has no shortage of water. What planet Earth has a shortage of, in some cases, is regional drinkable water, and there is a water distribution scarcity issue in certain parts of the world.
"So a data center converts liquid water in to steam."
They're using potable water when there's already a disturbing rate of aquifer depletion going on.
Datacenters aren't using "less than pure water". They are using water than can be used by people. farms, etc. Pure H2O is an artificial substance that takes energy to create and is only useful in a laboratory where there's a need to control contaminants. Dissolved solids in water are often a benefit to living things. Taking them out doesn't make the water better.
Climate change and droughts are for peasants. Their lawns will remain green and their data centers cool while the peasants must adhere to water conservation. They will fly to their climate summits in one of their luxurious private jets so they can tell the peasants how they have to walk more to reduce carbon emissions. They eat steak from private chefs while telling the peasants they have to eat bugs because cow farts and deforestation. If climate change isn't a hoax, then these people are achieving psychotic levels of selfishness.
This post has been deleted by its author
I was about to recommend the excellent reg online standards converter, but you can't enter gallons! Vultures, pls fix
An Olympic swimming pool is 50m long, by 25m wide, by 2 or 3m deep. 3m for multi-disciplinary pools - but seeing as we don't want to talk about spanking, I'm going for 2m deep.
Therefore 1 OSP = 2,500m³ = 2,500,000 litres +/- ?L of urine...
A US gallon (which we'll use) is 3.785L - the proper Imperial one is about 4.4L.
So 300,000g - 4,000,00g
=1,135,500L - 15,140,000L
=0.45 OSP - 6 OSP
I think it would be much easier if we worked in bushels. A much more sensible measure. At 36.4 to the litre for proper imperial. At which point 1 OSP = 68,681 bushels a much easier number to remember, I think you'd all agree.
A datacenter needs 4 million gallons of water ? Why should that be taken from the drinkable water supply every year ?
Build a tank that holds that water, and re-use it.
It will probably be more a bit more costly, but you'll only be draining the water once. Steam can be recycled endlessly.
Because they're evaporating it.
Some are using closed loops - 'dry' coolers or chillers. Those consume little to no water directly, but do use more electricity.
Others are using a local river or sea as a heat sink, which again consumes no water but may harm the environment by overheating the fishies.
But... Many DCs use evaporative cooling instead, where they simply let the water evaporate into the air and get blown away, to later rain into the sea. They tend to use drinking water for this because it reduces maintenance.
It's incredibly wasteful. They do it because it's cheap.
You have: 300000 gallons
You want: litres
* 1135623.5
You have: 4 million gallons
You want: million litres
* 15.141647
You have: 10 gallons/minute
You want: litres/minute
* 37.854118
You have: 10 gallons/minute
You want: litres/hour
* 2271.2471
You have: 126 million gallons
You want: million litres
* 476.96188
You have: 1000 miles
You want: km
* 1609.344
There wasn't any note that stated if the values were US gallons or Imperial gallons, and the article said "In the US", so I figured the values were most likely US gallons.
Both should simply stop recklessly using SI units with extra steps and use them directly, but nooooo.
If the units are Imperial gallons, here is the conversions;
You have: 300000 imperialgallons
You want: litres
* 1363827
You have: 4 million imperialgallons
You want: million litres
* 18.18436
You have: 10 imperialgallons
You want: litres
* 45.4609
You have: 10 imperialgallons/minute
You want: litres/hour
* 2727.654
You have: 126 million imperialgallons
You want: million litres
* 572.80734
We live on a planet with abundant water, This shouldn't be a difficult problem.
Maybe datacenters should be restricted in areas that have water issues with appropriate legislation so that they don't consume that needed for more important consumption such as food porduction and drinking. Then the datacenter must solve the problem without impacting others or be somewhere else.
DCs are industrial assets that output "compute" products and waste heat. Why not using that waste heat to reduce energy (and water) consumption in another industrial or residential domain? There was a short article somewhere that described a (small?) DC providing heat to the municipal heating system nearby. AFAIK the silicon wafers are extremely energy intensive to create, requiring sizable chunks of power just to melt the raw materials... Similarly, metallurgy and cement production could benefit from direct heat contribution to their processes...
To me it looks like the Capex is a major driver of decisions regarding the design and placement of the DCs, while Opex is to be transferred into the price of the final product, regardless of the long-term consequences.
I could rant along those lines a lot, but it would be another waste of water and energy to power the DCs and infrastructure needed to provide me with such a marvelous fondleslab :P
Why not using that waste heat to reduce energy (and water) consumption in another industrial or residential domain? There was a short article somewhere that described a (small?) DC providing heat to the municipal heating system nearby
Conceptually easy, and done in a few places. However, the financials are often pretty marginal, particularly because most process heat demand is based around high grade heat. You can convert low grade waste heat to high grade through heat pumps in serial, but with the chained inefficiencies it reduces the benefit significantly. The more modern (and minority) of district heating systems can take low grade heat inputs, but for both space heating and process heat there's a perennial problem about the dependability of these waste heat sources, because the waste heat provider probably won't guarantee the provision of heat.
Take an example of a hyperscaler DC: Loads of waste heat, could export that to a low temp district heat system. Conceptually easy, often both surprisingly expensive and involved to build. Moreover a hyperscaler DC in a region with highly variable energy costs will often be shifting loads to other sites depending on both processing demand and local energy costs. If local energy costs spike (eg cold winter's day, northerly latitudes) and/or processing needs decline (eg end of the working day) then they'll ramp processing down, and suddenly the heat output reduces at a time when heat demand is highest. What does the heat network or process heat user do? It has to revert to whatever Plan B is, and Plan B needs to involve assets that can reliably secure peak design heat capacity at no notice on the worst possible day. Same is true for more mundane interruptions like the DC gets shut down due a fire, or major comms failure, and the same interruptibility argument applies to other waste heat sources like thermal power stations or municipal waste incinerators.
So a heat network (or process heat system) needs to be designed to deliver 100% of capacity in the case of loss of waste heat. As a result the capex and maintenance of a waste heat supported system is actually a tiny bit higher than without waste heat (as there's all the 100% capacity assets plus the control and physical connection with the waste heat provision). It's also not just the capex and O&M - there's high costs of securing capacity of either gas or electricity at peak times. Those need to be contracted in advance with generators or gas shippers, and with the transportation and distribution systems, you can't just spring a sudden multi-megawatt demand on the local grid or pipelines at peak times without likely causing it to fail. It's also worth noting that the DC or any waste heat source needs certainty that it can offload its waste heat, and therefore needs the full capex and O&M of their own Plan B heat dump solution. In the case of a district heating system it's also worth noting these usually have trivial demand over the summer, and in some cases the systems are turned off outside the official heating system.
So all in all, yes this can be done, hopefully that (based on my knowledge working for a district heating company) will outline how a fundamentally simple concept with apparently obvious benefits can be more costly and less attractive than any intelligent layman might suppose. In that respect, it's like much of IT.
So all in all, yes this can be done, hopefully that (based on my knowledge working for a district heating company) will outline how a fundamentally simple concept with apparently obvious benefits can be more costly and less attractive than any intelligent layman might suppose. In that respect, it's like much of IT.
I think the biggest problem is the lack of joined up thinking. So take Slough as an example. Started out as a scrapyard, ended up as DC central. Consumes a lot of resources, wastes most of them and isn't very sustainable. It could utiilise some of its waste heat as district heating, but it was never designed for this, so the cost of trying to retrofit that would be collosal and uneconomic. Start with a blank slate, and the economics get a lot simpler. Policy is to build new homes, so build those with district heating to utilise 'waste' heat from datacentres or SMR clusters. Or do other controversial things like adding a CHP waste incinerator. Residents of New Slough would no longer have to sort rubbish into 5 or more wheelie bins, just stuff that burns, and stuff that doesn't. Or do that at the CHP plant instead.
"In the US, datacenters can consume anywhere between 300,000 and four million gallons of water"
Right, every time we read financial values here they are followed by ($ equivalent in US dollars).
Why is it different when the value is firstly expressed in US amounts?
This story is focussed on something other than money, but it's the same argument.
"xx million gallons": as others have queried, what sort of gallons? And how much is that in real measurement?
So now in addition to the general heating of global water resources from global climate change, we're now going to dump all the waste heat from AI into the mix.
Maybe in a million years or so something will evolve whose intelligence has survival value. Apparently talking monkeys weren't the best idea.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2025
