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.
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.
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.
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.
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
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.
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.
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?
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!
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.
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.
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.
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.
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.
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
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.
[1] https://www.c-win.org/cwin-water-blog/2022/7/11/california-a...
[2] https://wikifarmer.com/library/en/article/almond-tree-harves...
How much is that in "Libraries of congress" worth of water consumption?
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.
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?
Stargate is closed loop.
[dead]
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!
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.
Because industry moves faster than policy.
Because governments and elections can be influenced.
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.
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.
Gift URL: https://www.nytimes.com/2025/07/14/technology/meta-data-cent...