A data center like Meta’s, which was completed last year, typically guzzles around 500,000 gallons of water a day. New data centers built to train more powerful A.I. are set to be even thirstier, requiring millions of gallons of water a day
I naively assumed these were closed loops. Where does the water go? I would think it just gets warm and does not evaporate.
tpmoney · 8h ago
That whole paragraph also seems completely unrelated to the issue as well. It doesn’t sound like water supply is the issue so much as sediment in the water breaking pumps and clogging the infrastructure.
I guess the theory here is that the amount of water being cycled is stirring up sediment somehow? But if that’s the theory they don’t really say that or talk to anyone who says why or how that’s happening. Is the consumed water being returned to the aquifer somehow and churning up sediment with a lot of added turbulence? Is the volume being consumed creating some sort of suction effect that’s pulling sediment up? Was this project one of the ones that required “dewatering” as described in the article? Is the theory that is the thing that caused the problem and if so, does that mean the approving process for that needs an overhaul?
Not to say there aren’t issues to be addressed here, but the big “gallons of water” number seems to be tossed around a lot in these discussions with no quantification about what that actually means. The solution to the problem is different if that means gallons of water being pulled from the ecosystem entirely , or if it means gallons of water being heated and having effects on the ecosystem, or it means gallons of water burning through processing and treatment plant resources faster.
toast0 · 5h ago
I mean, given that the datacenter is only 1,000 feet away from their home, it may be that just the earth moving and heavy equipment for construction disturbed the waters they were tapping into. I don't see an indication in the article of the position of their wellhead or the depth of their well.
Small changes can make a big difference, I had to replace my submersed well pump, and even though it should be at the same depth as the old one, I still get a lot more sediment, even years later.
I'll say that it's pretty shocking that a data center was built so close to at least one home. I'd expect there to be more of a buffer between industrial and residential, especially in such a low density setting.
tpmoney · 1h ago
It's certainly possible, though I wonder is building a data center like this significantly more "destabilizing" to the ground water than say building a housing development? Many folks have had massive housing communities built at high speed 1000 feet from their property and we haven't seen these sorts of stories about well issues with that (that I'm aware of).
I think this is something I find immensely frustrating with the NYT and major media these days. Even the most basic sort of follow up questions never seem to be asked by their reporters. We're told "500,000 gallons of water" can be consumed, but noticeably lacking is any information about whether or not this particular data center consumers that much water. We're told that they use "about 10%" of the county's daily use. But we're also told that a pending request for 6 million gallons is "more than the county's entire daily use". So for ease of numbers let's say that the county's entire daily use then is a nice round 5 million gallons. That would indeed put the data center usage at 500,000, but if that's the true number, why not say that? Also the article starts by telling us that "Months after construction began in 2018" their well problems started, but the end of the article tells us that Meta bulldozed the forest by their house in 2019, and that their troubles started after that. The article also says that while construction began in 2018, the facility wasn't finished until last year. Even if the facility is consuming 500,000 gallons of water a day today, there's no possible way it was dong that "months" after construction started if construction only finished last year.
Also, if Google satellite imagery is to be believed, the part of the property that Meta built on near their home only just started getting cleared in 2019, and was still just clear cut land and dirt roads in April of 2021. Buildings started going in sometime between that time and February 2023. In late 2019 when they were already replacing appliances, the nearest heavy construction on the data center property was a half a mile from their property.
Of interest is that in that same area, just south of the new data center is also a brand new (as of 2022) water reclamation facility. A facility on which construction started in 2020, but for which there was already an existing pumping station and plans for work there had been in the works since around 2005. (https://www.covnews.com/news/new-newton-water-reclamation-fa...). Again if google satellite imagery is to be believed, work on that pumping station looks like it started in 2013 sometime, and ground breaking and clearing for the new expanded facility also started in 2019. Did that cause problems with the ground water? It seems like it treats water and discharges out into the river basin so maybe not? But also seems like the sort of thing a reporter might want to follow up on.
None of this is to dismiss the very real problem these people are facing, but at the same time, this reporting is frustratingly vague about both A) what the actual timeline of events and proposed mechanism of action is and B) any external evaluation of those claims and examination into other possible causes.
> I'll say that it's pretty shocking that a data center was built so close to at least one home. I'd expect there to be more of a buffer between industrial and residential, especially in such a low density setting.
This actually got me wondering, is a data center considered "industrial" for zoning purposes?
exmadscientist · 9h ago
Option A is to build a closed-loop system, that runs a million gallons of water per day through chillers and recirculates it. You need to find chillers that can handle that load continuously and enough power to run them.
Option B is an open-loop system where you run a million gallons of water through exchangers, heat it up, then dump the hot water and find a way to get a new million gallons of water from the local municipality.
Option B is cheaper, so they do that. Higher water prices would change that equation, but that's not what we have now, and it's hard to pitch an Option A project if anyone else is willing to offer rates that make Option B work. The Prisoner's Dilemma strikes again.
tomatotomato37 · 8h ago
I find it hard to believe continuous consumption of potable municipal water is cheaper than running chillers or exchangers cooled by a river/ocean, especially considering powerplants and the like have been doing the latter for decades
Veserv · 6h ago
Why? It is only a measly 500,000 gallons a day.
That is only ~2,000 m^3/day (~2 acre*foot/day). Even if they exclusively used the most expensive source of water, seawater desalination, that would only be ~800 $/day.
Your average almond tree uses 3-4 acre*foot per year [1]. So the yearly water consumption of the data center is ~200 almond trees. Your average almond tree produces ~50-60 pounds per year [2] and ~4500 pounds per hectare (2.5 acres), so that is the water consumption of a tiny 5 acre almond farm producing ~10,000 pounds of almonds per year.
The internet indicates the wholesale price of almonds is ~2 $/pound, so you can either have a data center or 20,000 $ worth of almonds.
How much is that in "Libraries of congress" worth of water consumption?
</SCNR>
exmadscientist · 6h ago
It really shouldn't be, but part of the site selection process for these things is finding a place with cheap enough power and cheap enough water that you can rip them off by dangling the "jobs!" carrot. So it's not exactly random. And there are enough locations in the US that view providing cheap utilities to their citizens as a benefit (which, when things weren't getting arbitraged on a national scale, was probably a reasonable policy) that they can always find someone.
pxeger1 · 8h ago
Maybe building a heat exchanger in a river requires loads of environmental / planning permits, but just producing millions of gallons of (warm) "sewage" doesn't, because it's already allowed?
FL33TW00D · 8h ago
Stargate is closed loop.
sellmesoap · 8h ago
Reminds me of the time the backup generators at my colocation provider overheated during a power outage. The reason? The fire at the nearby substation needed a lot of water to cool off the electrical fire and the generators were cooled open loop off the same potable water system. SRE has to cast a wide net to be effective!
Saris · 8h ago
Why isn't the focus on the local government who is allocating that much water without caring about the effects?
Yes AI is wasteful, but if they couldn't get water they wouldn't build there.
yifanl · 7h ago
Because industry moves faster than policy.
srean · 8h ago
Because governments and elections can be influenced.
tpmoney · 6h ago
Seems like all the more reason to put the responsibility and blame on the government. You will never eliminate “influence”, and especially the more power the government has, the more value there is in spending on “influence”. The only possible solution is to hold the government and the representatives responsible for taking actions to the detriment of their constituents. If we give them a pass because “elections can be influenced” we might as well just disband the government and allow governing by the highest bidder.
MangoToupe · 6h ago
Eh, even the distinction between private enterprise and government is largely irrelevant. At the end of the day, forces too large to fight conspire to make peoples' lives miserable.
I guess the theory here is that the amount of water being cycled is stirring up sediment somehow? But if that’s the theory they don’t really say that or talk to anyone who says why or how that’s happening. Is the consumed water being returned to the aquifer somehow and churning up sediment with a lot of added turbulence? Is the volume being consumed creating some sort of suction effect that’s pulling sediment up? Was this project one of the ones that required “dewatering” as described in the article? Is the theory that is the thing that caused the problem and if so, does that mean the approving process for that needs an overhaul?
Not to say there aren’t issues to be addressed here, but the big “gallons of water” number seems to be tossed around a lot in these discussions with no quantification about what that actually means. The solution to the problem is different if that means gallons of water being pulled from the ecosystem entirely , or if it means gallons of water being heated and having effects on the ecosystem, or it means gallons of water burning through processing and treatment plant resources faster.
Small changes can make a big difference, I had to replace my submersed well pump, and even though it should be at the same depth as the old one, I still get a lot more sediment, even years later.
I'll say that it's pretty shocking that a data center was built so close to at least one home. I'd expect there to be more of a buffer between industrial and residential, especially in such a low density setting.
I think this is something I find immensely frustrating with the NYT and major media these days. Even the most basic sort of follow up questions never seem to be asked by their reporters. We're told "500,000 gallons of water" can be consumed, but noticeably lacking is any information about whether or not this particular data center consumers that much water. We're told that they use "about 10%" of the county's daily use. But we're also told that a pending request for 6 million gallons is "more than the county's entire daily use". So for ease of numbers let's say that the county's entire daily use then is a nice round 5 million gallons. That would indeed put the data center usage at 500,000, but if that's the true number, why not say that? Also the article starts by telling us that "Months after construction began in 2018" their well problems started, but the end of the article tells us that Meta bulldozed the forest by their house in 2019, and that their troubles started after that. The article also says that while construction began in 2018, the facility wasn't finished until last year. Even if the facility is consuming 500,000 gallons of water a day today, there's no possible way it was dong that "months" after construction started if construction only finished last year.
Also, if Google satellite imagery is to be believed, the part of the property that Meta built on near their home only just started getting cleared in 2019, and was still just clear cut land and dirt roads in April of 2021. Buildings started going in sometime between that time and February 2023. In late 2019 when they were already replacing appliances, the nearest heavy construction on the data center property was a half a mile from their property.
Of interest is that in that same area, just south of the new data center is also a brand new (as of 2022) water reclamation facility. A facility on which construction started in 2020, but for which there was already an existing pumping station and plans for work there had been in the works since around 2005. (https://www.covnews.com/news/new-newton-water-reclamation-fa...). Again if google satellite imagery is to be believed, work on that pumping station looks like it started in 2013 sometime, and ground breaking and clearing for the new expanded facility also started in 2019. Did that cause problems with the ground water? It seems like it treats water and discharges out into the river basin so maybe not? But also seems like the sort of thing a reporter might want to follow up on.
None of this is to dismiss the very real problem these people are facing, but at the same time, this reporting is frustratingly vague about both A) what the actual timeline of events and proposed mechanism of action is and B) any external evaluation of those claims and examination into other possible causes.
> I'll say that it's pretty shocking that a data center was built so close to at least one home. I'd expect there to be more of a buffer between industrial and residential, especially in such a low density setting.
This actually got me wondering, is a data center considered "industrial" for zoning purposes?
Option B is an open-loop system where you run a million gallons of water through exchangers, heat it up, then dump the hot water and find a way to get a new million gallons of water from the local municipality.
Option B is cheaper, so they do that. Higher water prices would change that equation, but that's not what we have now, and it's hard to pitch an Option A project if anyone else is willing to offer rates that make Option B work. The Prisoner's Dilemma strikes again.
That is only ~2,000 m^3/day (~2 acre*foot/day). Even if they exclusively used the most expensive source of water, seawater desalination, that would only be ~800 $/day.
Your average almond tree uses 3-4 acre*foot per year [1]. So the yearly water consumption of the data center is ~200 almond trees. Your average almond tree produces ~50-60 pounds per year [2] and ~4500 pounds per hectare (2.5 acres), so that is the water consumption of a tiny 5 acre almond farm producing ~10,000 pounds of almonds per year.
The internet indicates the wholesale price of almonds is ~2 $/pound, so you can either have a data center or 20,000 $ worth of almonds.
[1] https://www.c-win.org/cwin-water-blog/2022/7/11/california-a...
[2] https://wikifarmer.com/library/en/article/almond-tree-harves...
Yes AI is wasteful, but if they couldn't get water they wouldn't build there.