I do think it is good that this article is so detailed, but the core of the argument is so obvious, that even the need to say it points to something pretty bad.
AI needs much better critics. Any industry needs physical space and if humans are to make anything at all, they will need space for it. Compared to what industries exists Data Centers are *extremely* gentle on the environment. Do people not know how dirty e.g. the chemical industry can get and how much land and water they use?
If you are vaguely aware what industrial production looks like, you immediately recognise that data Centers are one of the nicest things you could have in your area.
I find it interesting that you brought up site moderation. The reason social media sites are overrun with trolls is because the site owners want them to be. They won't pay for serious site moderation because they cater to the trolls, not the people who want facts in the discourse.
A place where AI has already shown itself to be useful is in site moderation. Spoutible uses it and has a lively community because of it.
I suspect this will be controversial, which is kind of the point. But I appreciate you saying these things clearly because I suspect many of us policy nerds are a little bit too nervous about violating assumptions to make this point publicly. That I say this anonymously kind of proves the point.
Outside of the specifics, which are great, I was really struck by this:
“I try to use data centers as a prism through which to understand much larger but invisible environmental problems, hidden by our tendencies toward populism and localism that data centers offend.”
I’d go one further to suggest this is an even broader synecdoche for policy work beyond just land-use and the environment. It’s something that replicates in my area of expertise (healthcare/science) where valorization of the noble profession obscures the terrible terrible incentives in place, making it so we can’t explicitly critique conduct or fix problems and need to work on minor issues on the margins.
That, of course, leads to a lot of antipathy towards political winds that barnstorm into the room from the SMEs, when the pressure has built. You can see this with a lot of the attempts at reform currently in the news.
Brilliant work. Agree with 95% of it. The only place my allegiance switched was the small town losing its adjacent green space. I would regret losing that to any operation. Might change my mind if the architecture was such that the data centres were intergrated with and augmented the greenspace.
I think farmland is a far cry from green space. If it's genuine nature, sure. A lot of this land seems to already have been cleared for agricultural use.
Andy, have you written any posts yet on the worries about air pollution or (audible) sound? I'd love to see those steel-manned, so I could have something real to redirect my AI-will-cause-environmental-doomed friends' interests to.
You know what, fair enough because I've done the exact same argument in the vein of "liberal arts boy thinks the cloud is the white things you see in the sky / how do you think you can read this comment." Thank you for reminding me to argue better.
The issue I always felt was least likely to be fake or heavily overstated was thermal pollution of water - but I haven't heard as much as a peep from any critic about this, aside from that frankly bonkers "data centers are heating the air by many degrees" paper.
Of course, thermal pollution issues are local by definition, and solvable by just not building in specific vulnerable spots. They wouldn't advance the overall "AI has to go away" cause much.
Great read, I really appreciate your data-based insight at a time when AI fears are so widespread and major publications only seem to be contributing. Very effective at putting data center issues back in perspective and evaluating the pros/cons like any other land use decision.
That said, I do think your analysis of the land use impacts overlooks one thing: the aesthetic and cultural value land holds. Like some of the articles you (fairly) criticize mention, farmland is not just empty space awaiting conversion. In so many places the agricultural landscape is the defining feature of local identity, and individual farms are family heirlooms, at least in the older parts of the country. Urban land take erases that, and in the case of data centers doesn't really replace it with any new source of identity. And sure, the spatial extent of data centers may be small compared to other land uses, but even a single data center has the ability to transform how people perceive their landscape or their town. If it were the Nevada desert at risk I might be less concerned, but I truly believe that cultural landscapes matter and that farms represent far more than the physical land they occupy.
Less sentimental but in a similar vein, the US having so much land is not reason to waste it. I can agree with your point about the US having too much farmland, but that doesn't mean the "excess" should be freely urbanized. If agriculture degrades land (and it certainly does), conversion from agriculture to urban just seems like another step of degradation, and a far more permanent one at that.
Anyways, I agree with your overall conclusion, just wanted to add my two cents on one of the key reasons data centers are so feared. It may be subjective, but I think it's still important to acknowledge when making any land use decision, data centers included. (Also, I see I may be a bit late to the party here...)
Great and thoughtful points on land use - it’s the counter argument to the usual environmental critique.
To extend this further, when thinking about environmental impacts and energy transition, the grid bottlenecks may be a key consideration. For example, the hyperscalers in essence compete with a place on the queue for grid access. However, it’s not just a queue in the traditional sense. Their ability to raise finance, long dated PPAs and higher credit ratings - cheaper cost of capital, give them a structural advantage in the queue, outlasting smaller competitors, resulting in higher probability of grid access.
This brings me to my next point on additionality. They may be financing wind farms etc that may not have existed previously, but that’s to power additional electricity use that also may not have existed previously. Is this where we start to see holes in the argument of advancing energy transition, outside of second derivative benefits of advancing technology that may take many years to realise?
Thank you! The sympathy for the wealthy, rent seeking, animal torturing, water table breaking, massively polluting farming lobby is absolutely baffling.
It's based on a memory of what farming once was when it was mostly small subsistence farms dotting the landscape. That's a really wasteful use of land and resources, but it's been mythologized as noble and wonderful by people whose great-great-grandparents did everything they could to escape that life.
Nice work! What I read seemed quite good. But this article still seems too long for most people? Maybe move more of the details into appendices after the conclusion, like you did with "some recent bad articles.."
Great post! I’ve enjoyed reading all your posts on the AI issue. I’m curious - is there a reason that the land in New England going from farmland to forest is considered a net good?
The federal government and the state of Michigan apparently didn't get the same memo this week.
In Washington, the EPA launched its Water Reuse Action Plan 2.0 — repurposing cleaned wastewater for data centers and semiconductor plants. The pitch: AI needs water to run, and that's now a national infrastructure priority.
Meanwhile in Michigan, a utility just approved a one-year ban on water use by data centers. Full stop.
Same week. Opposite conclusions.
The question nobody is asking loudly enough: when water gets scarce, who gets it first — your lawn, your tap, or the server farm running your AI assistant?
Water rights are pretty settled in how that gets handled. The person with the oldest claim gets it first. I think L.A. had to make deals with people at some point to secure their supply of water because the population centers are relatively new compared to the farms along rivers on the western United States, so they will always have a lower priority claim to water rights.
I do think it is good that this article is so detailed, but the core of the argument is so obvious, that even the need to say it points to something pretty bad.
AI needs much better critics. Any industry needs physical space and if humans are to make anything at all, they will need space for it. Compared to what industries exists Data Centers are *extremely* gentle on the environment. Do people not know how dirty e.g. the chemical industry can get and how much land and water they use?
If you are vaguely aware what industrial production looks like, you immediately recognise that data Centers are one of the nicest things you could have in your area.
This. Absolutely this.
I find it interesting that you brought up site moderation. The reason social media sites are overrun with trolls is because the site owners want them to be. They won't pay for serious site moderation because they cater to the trolls, not the people who want facts in the discourse.
A place where AI has already shown itself to be useful is in site moderation. Spoutible uses it and has a lively community because of it.
I suspect this will be controversial, which is kind of the point. But I appreciate you saying these things clearly because I suspect many of us policy nerds are a little bit too nervous about violating assumptions to make this point publicly. That I say this anonymously kind of proves the point.
Outside of the specifics, which are great, I was really struck by this:
“I try to use data centers as a prism through which to understand much larger but invisible environmental problems, hidden by our tendencies toward populism and localism that data centers offend.”
I’d go one further to suggest this is an even broader synecdoche for policy work beyond just land-use and the environment. It’s something that replicates in my area of expertise (healthcare/science) where valorization of the noble profession obscures the terrible terrible incentives in place, making it so we can’t explicitly critique conduct or fix problems and need to work on minor issues on the margins.
That, of course, leads to a lot of antipathy towards political winds that barnstorm into the room from the SMEs, when the pressure has built. You can see this with a lot of the attempts at reform currently in the news.
Great work.
At least with all the problems with academia, scientists basically unanimously acknowledge them.
Brilliant work. Agree with 95% of it. The only place my allegiance switched was the small town losing its adjacent green space. I would regret losing that to any operation. Might change my mind if the architecture was such that the data centres were intergrated with and augmented the greenspace.
"Intergrated" lol. Cast or wrought iron grates, I'm not sure which would be best.
I think farmland is a far cry from green space. If it's genuine nature, sure. A lot of this land seems to already have been cleared for agricultural use.
If you look up Archbald PA on Google Maps you can zoom in and see that it's mostly forested lands.
Andy, have you written any posts yet on the worries about air pollution or (audible) sound? I'd love to see those steel-manned, so I could have something real to redirect my AI-will-cause-environmental-doomed friends' interests to.
It’s on the list! Should be out in May
https://andymasley.substack.com/p/to-be-clear-i-do-understand-how-sound
Would love to see where "raising local electricity prices" and "raising local housing prices" fits onto that axis
"City boy thinks food comes from the store"
You know what, fair enough because I've done the exact same argument in the vein of "liberal arts boy thinks the cloud is the white things you see in the sky / how do you think you can read this comment." Thank you for reminding me to argue better.
The issue I always felt was least likely to be fake or heavily overstated was thermal pollution of water - but I haven't heard as much as a peep from any critic about this, aside from that frankly bonkers "data centers are heating the air by many degrees" paper.
Of course, thermal pollution issues are local by definition, and solvable by just not building in specific vulnerable spots. They wouldn't advance the overall "AI has to go away" cause much.
I think Sabine Hossenfelder did a video on this, but the timescale for it beginning an actual problem is on the scale of hundreds of years.
Great read, I really appreciate your data-based insight at a time when AI fears are so widespread and major publications only seem to be contributing. Very effective at putting data center issues back in perspective and evaluating the pros/cons like any other land use decision.
That said, I do think your analysis of the land use impacts overlooks one thing: the aesthetic and cultural value land holds. Like some of the articles you (fairly) criticize mention, farmland is not just empty space awaiting conversion. In so many places the agricultural landscape is the defining feature of local identity, and individual farms are family heirlooms, at least in the older parts of the country. Urban land take erases that, and in the case of data centers doesn't really replace it with any new source of identity. And sure, the spatial extent of data centers may be small compared to other land uses, but even a single data center has the ability to transform how people perceive their landscape or their town. If it were the Nevada desert at risk I might be less concerned, but I truly believe that cultural landscapes matter and that farms represent far more than the physical land they occupy.
Less sentimental but in a similar vein, the US having so much land is not reason to waste it. I can agree with your point about the US having too much farmland, but that doesn't mean the "excess" should be freely urbanized. If agriculture degrades land (and it certainly does), conversion from agriculture to urban just seems like another step of degradation, and a far more permanent one at that.
Anyways, I agree with your overall conclusion, just wanted to add my two cents on one of the key reasons data centers are so feared. It may be subjective, but I think it's still important to acknowledge when making any land use decision, data centers included. (Also, I see I may be a bit late to the party here...)
Great and thoughtful points on land use - it’s the counter argument to the usual environmental critique.
To extend this further, when thinking about environmental impacts and energy transition, the grid bottlenecks may be a key consideration. For example, the hyperscalers in essence compete with a place on the queue for grid access. However, it’s not just a queue in the traditional sense. Their ability to raise finance, long dated PPAs and higher credit ratings - cheaper cost of capital, give them a structural advantage in the queue, outlasting smaller competitors, resulting in higher probability of grid access.
This brings me to my next point on additionality. They may be financing wind farms etc that may not have existed previously, but that’s to power additional electricity use that also may not have existed previously. Is this where we start to see holes in the argument of advancing energy transition, outside of second derivative benefits of advancing technology that may take many years to realise?
FYI the "The blockers on housing" link is a 404. (It looks like it got double-encoded: The spaces that were encoded as "%20" became "%2520".) Corrected link: https://www.urban.org/sites/default/files/2023-08/Land%20Use%20Reforms%20for%20Housing%20Supply.pdf
Just to be clear, it's not " the Senate’s overweighting of low-population states. " - that's from our founders, in the U.S. constitution.
Thank you! The sympathy for the wealthy, rent seeking, animal torturing, water table breaking, massively polluting farming lobby is absolutely baffling.
It's based on a memory of what farming once was when it was mostly small subsistence farms dotting the landscape. That's a really wasteful use of land and resources, but it's been mythologized as noble and wonderful by people whose great-great-grandparents did everything they could to escape that life.
Nice work! What I read seemed quite good. But this article still seems too long for most people? Maybe move more of the details into appendices after the conclusion, like you did with "some recent bad articles.."
Great post! I’ve enjoyed reading all your posts on the AI issue. I’m curious - is there a reason that the land in New England going from farmland to forest is considered a net good?
Weird week in water.
The federal government and the state of Michigan apparently didn't get the same memo this week.
In Washington, the EPA launched its Water Reuse Action Plan 2.0 — repurposing cleaned wastewater for data centers and semiconductor plants. The pitch: AI needs water to run, and that's now a national infrastructure priority.
Meanwhile in Michigan, a utility just approved a one-year ban on water use by data centers. Full stop.
Same week. Opposite conclusions.
The question nobody is asking loudly enough: when water gets scarce, who gets it first — your lawn, your tap, or the server farm running your AI assistant?
Water rights are pretty settled in how that gets handled. The person with the oldest claim gets it first. I think L.A. had to make deals with people at some point to secure their supply of water because the population centers are relatively new compared to the farms along rivers on the western United States, so they will always have a lower priority claim to water rights.
Why are the hyperscalers buying so much land and then only building on a small fraction of it? Is it just space for future construction?
I suspect a part of it’s to offset the data centers from the areas around them, I’m not sure
Surely part of it is the light pollution issues, setbacks probably help to avoid causing more problems with neighbors.
Incidentally light pollution isn’t listed on the top chart in the article? I’d love to see you look into this.