I'm just not sure I understand what point they're trying to make.
Honestly, 6500 households of water a day is nothing. Individual cities often write more than this off in a day when rain overwhelms their watershed systems.
Even with perfect watershed recovery our ability to treat, utilize, and transport it back without losses is a far bigger concern. The average age of pipes in the US is approaching 50 years old. I've seen reports project the US will need to invest nearly 1T in the next 15 years to sustain the current system and rate of growth.
Hell, there was a 54 inch water main break in Detroit this winter that flooded ~120 homes. A conservative estimate for a pipe of that size leaking all of its water for 5-6 hours is 20-25,000,000 gallons of water. We're going to be seeing a lot more of that.
Another thing I see overlooked constantly is that many of these DCs are augmented with their own filtration systems which allow them to primarily consume greywater for a net gain.
Municipal water drawn from the Great Lakes is mostly borrowed for a short time. There's energy costs to treating it for potability and then treating the wastewater, but not really a change to how much water is available.
That completely misses the point, it's a harbinger. This area is seeing it first because it is largely floodplains on top of clay coupled with a decent amount of rain and a consistent freeze-thaw cycle.
Merely building a functioning bypass for a pipe of this size at this depth takes the better part of a year. Replacing it? Out in the suburbs some of the projects to repair pipes from the 40s that leaked and caused sinkholes are still ongoing, they started over 20 years ago.
> Many data centers rely on evaporative cooling, or “swamp cooling,” where warm air is drawn through wet pads. Data centers typically evaporate about 80% of the water they draw, discharging 20% back to a wastewater treatment facility, according to Shaolei Ren, an associate professor of electrical and computer engineering at the University of California, Riverside.
The refrigerant is closed loop, but the condensers are sprayed with water which are much more efficient at removing heat than a fan + air.
I did some research on additional cost of water + increased efficiency of the condenser, but it's not just that. The water needs to be treated, otherwise it leaves way too much sediment.
AI data centres rarely actually need to resort to using water because they're using DTC and not air cooling, so we're looking at cooling 60 degree Celsius water down to ~40 which you can do compressorless with dry coolers alone. In practice many will use cooling towers instead or aswell but they still don't need to evaporate nearly as much water as a normal low density data centre would per KW.
By the way "spray the condenser" technology is very very rare because it's pretty much worst of both worlds and nowhere near as good or efficient as a cooling tower but much more maintenance than a dry cooler + compressor. Typically for high efficiency modern sites you'd be looking at cooling towers or dry cooler + compressor for water facilities or spray the water directly onto the air (direct or indirect adiabatic) for air based facilities.
To complicate things many (typically city DCs/Colos/ enterprise facilities) are air based facilities that convert to water by using heat exchangers in the datahalls.
I'm not inclined to endorse continued data center construction for much of anything, AI especially, but I was a little curious about how they determined what areas qualify as "high-stress" water conditions? Because they show red squares where I live (near Charlotte, NC) and we are pretty firmly within the normal range of precipitation and water levels. If anything, it's been wetter in the last 10-years than at any point in my lifetime.
Well I guess they should not be building these data centers in areas with no water and high ave Temps. Where they are building having these issues are a surprise to no one.
There are areas with lots of land, water and much lower ave air temps in the US and some have empty buildings that can be retrofitted for use as data centers. But no, lets build in places that allow the company treat their employees as slaves.
Get rid of the corn subsidies and production will shift to more sustainable crops that don't result in high fructose corn syrup being overused and increasing rates of diabetes and obesity.
Have you considered you have cause-and-effect backwards? Any human advancement hurts the planet and then the burden is on environmentalists to bring awareness?
Why would a for-profit not hurt the environment for more profit if it isn't illegal and won't get them sued (for more than they profited)? There's a rich history of them doing exactly that. Similarly, governments have done horrific environmental damage and it was up to environmentalists to create the awareness to make it stop.
I honestly don't know how you can look at a world with record numbers of wildfires, communities fighting over water supplies, and for-profit companies say that water beyond subsistence levels isn't a human right and think "oh yeah, environmentalists are looking for a reason why this water-consuming thing is bad".
Especially when the same technology could be powered in a way that doesn't pollute as much and cooled in a way that doesn't consume as much water -- if only the environment was more important than shareholders.
Yeah because the crypto world underwent a bunch of disastrous rug-pulls and large scale scandals that scared people off. And frankly I'm happy about this, the moment I started (and I'm not joking here) taxi drivers and hairdressers talking about NFTs I was pretty certain (and hopeful) that the bubble was about to burst.
I'm just not sure I understand what point they're trying to make.
Honestly, 6500 households of water a day is nothing. Individual cities often write more than this off in a day when rain overwhelms their watershed systems.
Even with perfect watershed recovery our ability to treat, utilize, and transport it back without losses is a far bigger concern. The average age of pipes in the US is approaching 50 years old. I've seen reports project the US will need to invest nearly 1T in the next 15 years to sustain the current system and rate of growth.
Hell, there was a 54 inch water main break in Detroit this winter that flooded ~120 homes. A conservative estimate for a pipe of that size leaking all of its water for 5-6 hours is 20-25,000,000 gallons of water. We're going to be seeing a lot more of that.
Another thing I see overlooked constantly is that many of these DCs are augmented with their own filtration systems which allow them to primarily consume greywater for a net gain.
Municipal water drawn from the Great Lakes is mostly borrowed for a short time. There's energy costs to treating it for potability and then treating the wastewater, but not really a change to how much water is available.
That completely misses the point, it's a harbinger. This area is seeing it first because it is largely floodplains on top of clay coupled with a decent amount of rain and a consistent freeze-thaw cycle.
Merely building a functioning bypass for a pipe of this size at this depth takes the better part of a year. Replacing it? Out in the suburbs some of the projects to repair pipes from the 40s that leaked and caused sinkholes are still ongoing, they started over 20 years ago.
I was under the impression data centers use closed loop cooling.
I had the same impression, but apparently not.
> Many data centers rely on evaporative cooling, or “swamp cooling,” where warm air is drawn through wet pads. Data centers typically evaporate about 80% of the water they draw, discharging 20% back to a wastewater treatment facility, according to Shaolei Ren, an associate professor of electrical and computer engineering at the University of California, Riverside.
The refrigerant is closed loop, but the condensers are sprayed with water which are much more efficient at removing heat than a fan + air.
I did some research on additional cost of water + increased efficiency of the condenser, but it's not just that. The water needs to be treated, otherwise it leaves way too much sediment.
AI data centres rarely actually need to resort to using water because they're using DTC and not air cooling, so we're looking at cooling 60 degree Celsius water down to ~40 which you can do compressorless with dry coolers alone. In practice many will use cooling towers instead or aswell but they still don't need to evaporate nearly as much water as a normal low density data centre would per KW.
By the way "spray the condenser" technology is very very rare because it's pretty much worst of both worlds and nowhere near as good or efficient as a cooling tower but much more maintenance than a dry cooler + compressor. Typically for high efficiency modern sites you'd be looking at cooling towers or dry cooler + compressor for water facilities or spray the water directly onto the air (direct or indirect adiabatic) for air based facilities.
To complicate things many (typically city DCs/Colos/ enterprise facilities) are air based facilities that convert to water by using heat exchangers in the datahalls.
It's much cheaper just keep pulling in cool water than it is to cool it yourself
Welcome to Gelman amnesia.
I'm not inclined to endorse continued data center construction for much of anything, AI especially, but I was a little curious about how they determined what areas qualify as "high-stress" water conditions? Because they show red squares where I live (near Charlotte, NC) and we are pretty firmly within the normal range of precipitation and water levels. If anything, it's been wetter in the last 10-years than at any point in my lifetime.
The conflation of AI and use of natural resources / energy use is mostly a political choice and a straw man.
So put datacenters in Tennessee. We have tons of water and hydroelectric power.
Surprised there isn't more of a push to align compute with suitable geography frankly.
A lot of stuff doesn't need 10ms latency. Why not move it somewhere that has geothermal, or abundant water, or say an ocean to dump heat into.
Seems like the approach is always to bring the resource to the datacenter instead
https://archive.ph/tf090
Well I guess they should not be building these data centers in areas with no water and high ave Temps. Where they are building having these issues are a surprise to no one.
There are areas with lots of land, water and much lower ave air temps in the US and some have empty buildings that can be retrofitted for use as data centers. But no, lets build in places that allow the company treat their employees as slaves.
why not built near the ocean and use saltwater or brackwater as coolant?
Obligatory link about AI water/energy use https://open.substack.com/pub/andymasley/p/individual-ai-use...
Eh, I'd focus on agriculture first. AI is a drop in the bucket compared to what it takes to grow a cow or an almond.
Step 1 is simpler. Get rid of farming in the desert first.
This a fight going on in Idaho right now:
https://www.idahostatesman.com/news/local/community/boise/ar...
https://archive.ph/X9nTS
If you look at a google map of Ada County Idaho, the only viable agriculture land requires watering from the Snake and Boise Rivers.
https://maps.app.goo.gl/TFPgA6nG6hMBfPcx5
I often wonder about this. Is it actually cheaper to farm near water, or farm in a hot climate and import water (e.g. CA & AZ).
You can't eat a hyper realistic picture of a dog in a suit.
Would you concede the point if I produce a photo as proof?
Yep. Agriculture uses roughly 2-4K times more water than the infrastructure that powers data centers. https://www.unesco.org/reports/wwdr/en/2024/s
I'd rate basic necessities like food 4000x more important than data centers too
I don't know. I suspect most people rate data centers higher than almond milk...
Of course! But what about all of the agriculture that gets exported? In CA, it is over half of almonds and ~15% of alfalfa (cattle feed).
Get rid of the corn subsidies and production will shift to more sustainable crops that don't result in high fructose corn syrup being overused and increasing rates of diabetes and obesity.
If we could just get rid of these darn living creatures we would use much less water.
That's what I'd love Markov bullshit generators make these crawlers implode with metacircular nonsense. That and Gzip bombs.
Artificial Intelligence is for people without enough natural intelligence.
I’m hugely pro conservation, clean air and water, etc.
But modern environmentalism is a religion and this is how it works. Any human advancement needs a narrative on how it hurts the planet.
Have you considered you have cause-and-effect backwards? Any human advancement hurts the planet and then the burden is on environmentalists to bring awareness?
Why would a for-profit not hurt the environment for more profit if it isn't illegal and won't get them sued (for more than they profited)? There's a rich history of them doing exactly that. Similarly, governments have done horrific environmental damage and it was up to environmentalists to create the awareness to make it stop.
I honestly don't know how you can look at a world with record numbers of wildfires, communities fighting over water supplies, and for-profit companies say that water beyond subsistence levels isn't a human right and think "oh yeah, environmentalists are looking for a reason why this water-consuming thing is bad".
Especially when the same technology could be powered in a way that doesn't pollute as much and cooled in a way that doesn't consume as much water -- if only the environment was more important than shareholders.
We already went through this with crypto mining; the oceans didn't boil, the world didn't end.
Etherium also went to proof of stake. Apparently it's going swimmingly.
Tell me more about your impression of the timescale that climate change would progress at.
Yeah because the crypto world underwent a bunch of disastrous rug-pulls and large scale scandals that scared people off. And frankly I'm happy about this, the moment I started (and I'm not joking here) taxi drivers and hairdressers talking about NFTs I was pretty certain (and hopeful) that the bubble was about to burst.