Data centers may be the most controversial objects in America. They’re blamed for draining water reservoirs, stressing electrical grids, and driving up utility bills. Communities throughout the country are pushing back and seeking moratoria.

But Andy Masley says this is all wrong—that data centers aren’t especially resource-hungry. Through his dogged research and insistence on putting numbers in context, Andy has emerged as the most influential debunker of data center myths.

He joins Jon Bateman on The World Unpacked to compare ChatGPT prompts with microwaving a burrito and explain why a bestselling book overestimated AI’s water use by one thousand times.

Transcript

Note: this is an AI-generated transcript and may contain errors

Jon Bateman: Andy Masley, welcome to The World Unpacked. Yeah, thanks so much, Jon, really happy to be here. A lot of people out there right now have questions and concerns about the environmental impact of AI and data centers, the resource use, the effects on local communities. As far as I know, you are the guy out there on the internet who has done more than anyone to compile data on these issues and develop almost like an encyclopedic knowledge of that. But first of all... How did that come to be? How did you become this guy?

Andy Masley: Yeah, so in January of last year, my own personal use of chatbots was really starting to kick off. And some additional context before going into that story is I'd been pretty fired up about climate since I was a teenager back in the 2000s, had double majored in philosophy and physics, and was pretty interested in both the science and ethics of climate change and how we can make our big complex society transition to entirely renewable energy in a way that's both fair and pluralist, but also takes the threat of climate change very seriously. And when I was going to these parties and bumping into people I didn't know, I would very often have people say like, oh, you use chat GPT, that's so terrible for the environment, why do you do that? And I would be like, oh, why did you believe that? What do you believe the actual environmental impact is? And they would say something like, oh, it's 10 times as much as a Google search. That was a really popular line two years ago. And my first thought was like, oh, 10 Google searches is so. Tiny as an amount of energy. Even if I said like, oh, I did a thousand Google searches today, I don't think the average person would be like, oh, do you not care about the environment? You did a 1000 Google searches, that's insane, you know? And so I would ask people this, and I was finding that the general conversation was pretty bad from the get-go. They would be, oh why are you defending this? You're like a tech bro, like what's going on? And it was very hard to get through to them at these parties.

Jon Bateman: So you're not a tech bro, you're more of like a climate bro, if anything else, and yet you were just left with these unanswered questions and confusion, and so you started digging into it.

Andy Masley: Yeah, so basically I wanted to know like surely someone else has done this already, like someone somewhere must have written about how this doesn't really add up, like cutting a computer program, even if it's something more resource intensive, like AI is just not going to make a dent in the average person's emissions or water use. And I was finding that everything I had read about it was actually very strange. Like the first article I'd read about AI in the environment, I think it was from and they had included the statistic about how like ChatGBT in total is now using two times as much energy as a whole person. And like my thought there was like, oh, like, you know, so many other popular apps use so much more energy than that. This really gives us nothing. And I was like okay, you now, I can't find anything satisfying here, so maybe I'll just write my own personal case for why I don't think using ChatGBt is a bad environmental choice. I had a blog at the time that had like 50 loyal readers, they were most of my friends, and I was mostly writing these like long form pieces on things I thought were. Important that I would want to give someone if they wanted to know my deal, but I didn't expect it to take off at all. And so this was the same way. I kind of wanted something to keep in the back of my pocket to use at parties, to be like, Hey, this is my deal. If you disagree, that's fine, but this is where I'm coming from. So I'd written this and I included a lot of graphs about like, hey, look, you would have to do like 200,000 chat GPT prompts like per year to equal the emissions that you would save from say like recycling or driving just a little bit less and stuff like that. Um, And after I wrote this, within a day or two, it was really blowing up online in a way that nothing else that I had written had. I think the hosts of Hardfork were using it to argue on their show almost immediately, and I was like, whoa, that's pretty crazy. I feel the niche there. And so, as I started to look into this more, I realized just how strange the state of the debate was. And for a little while, I was just focused on individual impacts of your personal AI use. And I was kind of just deferring to people saying. Oh, but you know, these data centers are so bad. Like there are these huge buildings that use huge amounts of electricity and water. And at the time I was pretty deferential. I was like, oh, you know I know that data centers concentrate such huge amounts of individually tiny prompts and activities that on net they're probably as damaging as people say because surely people can't also be getting that wrong. And then around like April or May of last year I started to look way more into them and realized one that like almost all of the Reporting on data centers also had a lot of the same kind of strange moves that I had learned to be wary of in my own time of reading about climate. The big one I think is just reporting gigantic numbers without context. Like, oh, this thing uses millions of gallons of water or it's using like however many gigawatts of electricity in total and stuff like that. And basically I'd learned through, you know, like years of reading about this stuff that anything that we do at the scale of society. Is gonna use a lot of resources from the perspective of an individual and like you need to contextualize that with all the other ways that we use electricity and water and things like that to understand how it's actually impacting the environment.

Jon Bateman: Yeah, I think one of the biggest insights that I've gotten from reading your work over the last couple years is that AI is an industry. And so if we're thinking about its water use and electricity use, the right comparison is to other industries. And so that idea like unlocks a lot of different interesting questions and data, which I wanna get into. But before we get into that, as you said, people have a sense of personal investment in this. They worry. About their own contributions to environmental and resource issues, pollution, the like. You know, my mom has sent me a bunch of articles over the last couple of years about she's worried about her chat GPT use and is it responsible. So just what is your bottom line explanation to the average viewer or listener out there who wants to know whether their personal use of an AI chat bot each day is a significant part of their environmental footprint.

Andy Masley: I'm basically not aware of any major researcher who's finding consistently that people's, the cost of individual AI chatbots is adding up to something that we should worry about if we place it in the context of all the other ways that we use energy and water. So I might actually start with water because I find that that's something that people worry even more about. There's this really popular statistic that. One individual chatbot prompt can use an entire bottle of water. And the researcher who came up with that statistic, who had reported it to the Washington Post, who popularized it, no longer agrees with the statistic at all. At the time, he thought it was a reasonable estimate. But we now just know a lot more about what the models at the time were like and what current models are like. And he agrees that the current models use probably something like 30 times less water per prompt than that. But this specific statistic has stopped. And if you just look at that in a vacuum and say like, oh, you know, even 30 times less than a bottle of water could still be a meaningful amount. Like, this can always sound like a lot in a vaccuum. I think it's very shocking to people to hear that their computer use uses any resources at all. But if instead you just ask like, how many AI prompts would it take to raise my personal water footprint by 1% or raise my personal carbon footprint by one percent? I usually try to start there as a way of helping people understand. And when I do that, I usually find that. The average person's chatbot prompts probably would require something like 200 to 1,000 prompts to raise that person's emissions and water use by 1% specifically. And this is partly due to the fact that computing is just so incredibly resource efficient across most things that it just very rarely adds up to being a big part of our emissions and our use. So yeah.

Jon Bateman: So what would you compare it to? What's another thing in the typical person's life, like an American, for example, that might be in the same realm or in the order of magnitude in terms of water or energy use as chatbot use?

Andy Masley: Yeah, so I usually like to compare to using a microwave for a few seconds at a time. Like the average chatbot prompt is probably something like a few second of a microwave. And if you have a whole conversation, maybe that's like a minute of a micro wave. And like that's definitely not nothing, but it's also like, oh, like I would personally use a microwave for one minute if it could generate this gigantic, useful conversation for me.

Jon Bateman: Or just reheat a burrito. I mean, most people, those who are kind of conscious about their energy use, they're typically focusing on things like travel, plane rides, car rides, how warm you heat the home in the winter. I actually don't know anybody who is really rationing their microwave use.

Andy Masley: Yeah, and I think, like, in any other situation with computers, I think this would be read as very strange. Like, I haven't read a climate expert say, like oh, you should ration how many Google searches you use, or like, yeah, again, like exactly how many seconds of a microwave you use. Like, maybe some people would say like, oh, don't use your dryer quite as much, but like, I don't know anyone who is, like trying to optimize exactly how many seconds their microwave uses for the sake of the environment. And the reason they're not doing that isn't just because like Oh, this would be difficult and it would be a worthy sacrifice if I could. Like there's a more complicated reason where if you start to make these incredibly tiny cuts, like this also, also costs like your time, your energy that you might not then spend on more meaningful cuts or other things you could do for the environment or other important things for your life. And at some point, like incredibly, incredibly tiny cuts to your personal lifestyle. Just don't really trade off or even matter very much or even imply that limit less at all because again like I think These incredibly tiny cuts are likely to be swamped by other things that we do.

Jon Bateman: You mentioned that other forms of computer use also use water and power. Sending an email, doing a Google search, storing an image in the cloud, watching Netflix, listening to this podcast. You guys out there, cover your ears, because right now you are burning coal and evaporating water in some form. How significant is any of that compared to AI use?

Andy Masley: Yeah, so AI use is definitely moderately resource-intensive by the standards of other things we do online specifically. But the thing is that other things that we do online are just such an incredibly small part of our missions from the get-go that even a large multiple of those can still be very small. One example I use a lot is that an analog watch uses literally one millionth the power of a digital clock. If you buy a digitalclock, you're multiplying your time-telling energy use by a factor of a million. And does that tell us anything about how bad it is to buy a digital clock? The answer is it doesn't, because even though the analog watch is so much more efficient, these both round to zero in an absolute sense. And so with AI specifically, it's a little ambiguous exactly how much, say, Netflix watching or other things use in comparison to AI, just because Netflix is actually very efficient. Best I can tell right now is it might be something like 10 to 20 minutes of Netflix for every AI prompt. So that's actually a moderate amount if you're just looking at the data centers themselves. And so our computer use in general, if you're thinking about what's happening in the data center, AI is often slightly more resource intensive than like other comparable ways of spending time. If you just look at the resources of actually running your laptop or your computer, then AI actually just becomes like running a few minutes of your laptop normally. Like every few minutes, your laptop will use about as much energy as the AI prompt in the datacenter and stuff like that. And so like, again, if I met a friend who was like, I'm trying to, you know, like reduce my emissions as much as possible. And to do that, I'm ending my laptop use like three and a half minutes early every day. I would just be like, oh, this is just so hopeless as a way of affecting the climate. It's gonna get zeroed out by all these other things you do. There are just all these other changes you can make instead that are much better. And it just seems like the sad waste to cut your digital activity as a ways of reducing your climate impact specifically.

Jon Bateman: So probably a lot of people are out there listening to this and thinking, OK, this could be compelling if true, but where is he getting his data? Is this data coming from independent scientists who have rigorously studied the issue? Is it coming from industry sources that might have an incentive to skew these numbers? Is a lot this just informed speculation that actually we should have relatively little confidence in Where's the data coming from?

Andy Masley: Yeah, so there are a bunch of independent experts that I personally defer to, partly because they're much more critical of AI's use than me. Like there's this one organization called EcoLogits who estimate their best guesses of AI is both energy and water use based on what they know about both the software, the hardware that it's running on, the grid that data centers are built in and things like that. And so they're one source that I like to defer to partly because they're really not coming in trying to defend AI's resource use. Like if you look at the bottom of their page, they make a lot of points about how, oh, if everyone stopped using AI, we would save this many plane flights and stuff like that. But the numbers they use mostly converge on the same estimates from a lot of other sources as well. And so I don't really try to update too much on any one source. Like if I see one source that's way lower than the others, I don't go, aha, like this is the one I'll use, you know? Like instead I try to figure out like, what's the general spread of estimates from a lots of different places? Based on this common knowledge we have about roughly the algorithms running the models, the hardware that they're running on, and they usually do seem to converge on a general range. It's not exact, and there's a lot of variation in the energy that different lengths of prompts will use. We can get into that separately, but just starting with just the median prompt and what the average person is sending. I haven't found any source anywhere that's implied that the median chatbot prompt. Is using such a large amount of energy that it will noticeably impact your carbon emissions or water use specifically.

Jon Bateman: I guess what that means is if people are coming away with different views on this issue, it's not that they actually have different data or are making different assumptions about the data or have different levels of skepticism about different information sources. It mostly would be just based on the framing of the information and kind of how people might be telling stories around this information. And what do they use as their comparator? What do they consider to be a large or small amount of water or energy? Is that your assessment?

Andy Masley: Yeah, there's that. I think there are a few just blatantly wrong facts floating around, like the bottle of water per prompt fact. I think a lot of people also just don't know how much water and energy their everyday lives use. So like one big problem in the debate is that when people hear like, oh, AI uses a bottle of Water, they immediately lurch to like, oh, well, I don't like drink so much water every day. And so this is using almost as much water as I drink. And the amount of water that you drink is really a rounding error in your total daily water footprint because water is just so useful. And plentiful in America anyway, that almost everything that you do actually has a pretty significant water footprint, both the electricity that you use, the products you buy, the food that you eat, and if you add all those up, it's very hard to get AI above a completely marginal amount of water.

Jon Bateman: Can you paint a picture for people about the way that a consumer product might have upstream water use in some kind of industrial process? What is that process? And how is water being used? And then how does that result in some huge amount of water as compared to me just taking a shower?

Andy Masley: So one really obvious one is like say you want to buy a pair of jeans like jeans require cotton Which is grown out of the ground and if that cotton is irrigated at all That's going to involve huge amounts of water being dumped into the ground And it either becomes a part of the plant or it evaporates or some might return to the local water system But still a significant amount of water is lost from the local Water system in the process of growing the cotton that eventually become your jeans or like other materials for other things that you might buy specifically. And if we compare this to AI in comparison, where what's happening with AI is that water is being flown in the vicinity of computers that heated up a little bit because they had to answer your prompt, and that water has then flown to a place where it's evaporated, or some of it's evaporated and then it circulates back. I would just estimate, without thinking about it, that a process that involves dumping huge amounts of water into the ground to grow a physical object is just gonna be a way bigger part. Of my water footprint than the thing that cools my computer specifically. And if you actually look at it, each pair of jeans probably takes something like a million AI prompts worth of water to develop. And so there's a lot of embodied water cost. I think the other key thing that a lot of people don't know is that almost all of the ways that we generate electricity in America involve heating water to spin turbines specifically. And there's lot of ways that water can be evaporated and lost that way. A lot of water is also lost by all like, the dammed rivers that we have, where hydroelectric dams are used to generate electricity. Those block off water, which is then evaporated at a much higher rate, because it's sitting still in the sun. And so a huge amount of water cost happens through our electricity use in that way as well. And so one example I'll use a lot is that your alarm clock, if you have a digital clock that uses something like half a liter to a liter's worth of water, it consumes that much water in power plants on average if you're drawing from the average US grid. And so. Even the electricity that you use has a big water cost. And so everything that you do online has a water cost, not only because data centers use water to cool them, but also because all that electricity and the electricity you're using in your computer also has a cost to generate as well.

Jon Bateman: I think this would be a good point to help people understand what is a data center, what's happening inside of it that requires water, where is the water coming from, what is happening to it inside that building, and does it leave again, where does it go. Like, when we're talking about data center water use, just at a physical level, What is that?

Andy Masley: Yeah, so a data center is a really gigantic building with a bunch of computers inside. Like some people use the metaphor that the data center is itself a really gigantic building sized computer that's being interacted with by thousands or tens of thousands of people at once. And there are two kinds of data centers that we can think about. They're kind of more traditional like internet oriented data centers and then more data centers that are built for AI specifically. Like we can kind of get into the difference there. But data centers have existed since the internets existed and before that even like way back into the 50s I think is the first time the term data center was actually used and Basically, it's just very useful for a lot of different reasons to concentrate a lot computing in one place It's really useful to have computers that can very easily connect to each other You get economies of scale and the energy used and the heating and cooling systems that they have And so basically every time you interact with the Internet and you download a file that's stored on the internet or you open up a webpage that's sort on the Internet, you're interacting with the data center almost the way that you would a computer. Like the data centers out there somewhere and you're accessing something inside of it and it's bringing it to you.

Jon Bateman: Yeah, I mean, in other words, data centers have existed long before the advent of chat GPT, right? We already have thousands of data centers dotting the landscape in different parts of the United States, elsewhere in the world. Now because of the rise of generative AI and large language models, data center are being built bigger and faster than ever before. Before we get into that though, just help us understand the water use specifically. What is it used for and what does that physically look like?

Andy Masley: Yeah, so there are two parts of the water cost of the data center, the actual water that's used inside of the datacenter, and then the water that is used to generate the electricity off-site in the normal power plants that use water. And usually most of the Datacenter's quote-unquote water cost is actually in that off-site water cost just generating its electricity, just because all electricity that we use usually involves water unless it's coming from solar wind.

Jon Bateman: Okay, so in the same sense that my using my alarm clock actually uses water, data centers primarily use water in that way, just through drawing power.

Andy Masley: Yep, exactly, in the same way that others do. And that's definitely like a problem, but I think this can already lead to a lot of misconceptions about like where this water is actually being used and whether it's flowing through the building. And so a smaller fraction of that flows through the data center itself. And there are some things about data centers that make their water use a little bit different and strange. Like one aspect is that most water data centers use has to be treated municipal water in a way that power plants don't. Like power plants could say, just like draw from a lake. And then send it back. Whereas data center water needs to be at a much higher level of just cleanliness because it needs to able to flow through pipes in the data center without getting it dirty specifically.

Jon Bateman: And the idea then is that as water flows through these pipes and the pipes are sitting next to these server racks and the servers are generating heat, just like your laptop generates heat. And my laptop has a little fan that sometimes gets really, really loud when it's working really, really hard, particularly if it's using AI. But in the data center world, they're not relying on these fans so much as they're relying on cool water flowing past in order to absorb that heat.

Andy Masley: Yep, 100%. So yeah, basically the issue for data centers is just like your laptop gets hot, all the gigantic amounts of chips and computers inside also get hot as energy flows through them and is ultimately dissipated as heat. And the question is, how do you move that incredible amount of heat in the data center away from it? The largest data centers right now can get truly astronomical, and the very largest data center can use as much energy as whole cities. And all of that energy is being dissipated as heat inside and needs to be transitioned out of it very fast, specifically. And so the data center has different options for how to move that heat. And water is just a very energy efficient way of doing that, basically, because water can pick up heat very well. It could then be moved somewhere else. And if you just allow it to sit in place outside, it can evaporate. And so it takes that heat away with it. There are other ways of cooling a data center. Like one of them is like what's sometimes called closed loop, where water is transitioned to what's basically like a refrigerator. And cooled down, and that has more of an energy cost, specifically, and so there's this interesting problem where data centers that use less water per unit of energy will require more energy to cool the servers instead. So like, another option for some data centers is air cooling, where instead of flowing water, they just flow cool air instead, and it takes more energy to actually get that air moving and get it cooled to the point where it can actually cool the data center, basically.

Jon Bateman: So to some extent, we're going to be jumping back and forth between water and power here. But as you just said, some of the largest data centers now are so massive that they could use as much power as a small city. It's just intuitively, when someone hears that, how could that type of facility not be a massive drain on the power and water available in a specific local area? You're listening to The World Unpacked with John Bateman, where we dive deep into pressing global issues and make sense of the big forces shaping our world. Now, have you learned anything so far that intrigued or surprised you? Let me know in the comments or just give us a like. And if you wanna hear more of my conversations with the world's most informed and interesting people, you can subscribe right here. Now, back to the show.

Andy Masley: A lot of people will correctly point out that man, like some of these really large facilities can use millions of gallons of water per day. How is that sustainable? And I think people are especially upset that water is being used on kind of a transient digital product that will just like disappear, especially if they don't see any value in AI in the first place. And so it's understandable that they react to this. My question here is to not so much compare this to like, oh, millions of gallons of water. That sounds like a lot from my perspective. But how much does this actually compare? To the way that cities or other industries or farms use water specifically, because the way that residents in a city use water is usually actually a small minority or a small fraction of the total ways that water is used in a region. Almost all of American water consumption happens via farming and other industries rather than households specifically. And so my first move here is to just say, yeah, obviously we should be. Questioning these really gigantic facilities that are popping up everywhere. It's not that I want people to just close their eyes to them. But I do want to say, instead of just reacting like, oh, this is millions of gallons of water per day, I want us to just get our footing in, how does this actually compare to other things happening in the area?

Jon Bateman: Yeah, so would the better comparison be like, OK, a data center wants to be built in my municipal area. How would that compare to another light industry coming in, whether it's an Amazon fulfillment center, a car manufacturing plant, I guess that'd be heavy industry, some new agricultural operation, animals, plants, whatever you're growing there. Like Is that the right way to think about it? And if so, what does that tell us?

Andy Masley: Yeah, so my general rule, my request for thinking about this is to try to imagine which comparisons would be useful to you if you were like a city planner or something like that. Like if someone came to you and said, this industry is gonna use like two million gallons of water per day or it's gonna use like five swimming pools per day or something, I'm just kind of like making up those, I don't exactly know the conversion of swimming pools to gallons here. But I think that giving this number in swimming pools or gallons of water would actually leave you kind of blind. To how this would actually affect your area, because you wouldn't know how that would compare to other things in the area, where if instead the person said, hey, this data center is gonna use about 2% of the region's water, and it's gonna be about as much as the golf course, or it's going to be about as much like the car manufacturing facility, or this farm or something, those comparisons would give you a lot more understanding of, oh, this is roughly how I expect the data center to interact with the water resources that we already have in the areas, and the things already using water, and My general claim here is that once you contextualize data centers that way, they just kind of fade into the background of other normal ways that we use water. Like in basically every instance I can find of data centers controversially using a lot of water, they seem to behave as, you know, about the same amount of acreage of like corn farms nearby using water or to like nearby factories using water. And they don't, they very rarely stand out as like, oh, this is such an unimaginable amount of water. That the community is going to have a hard time meeting it. There are definitely a few places where data centers use very large fractions of regional water. And that can be a problem. But here, they still seem to be mostly behaving like other industries. And then the question is, like,

Jon Bateman: Yeah. So I guess if I'm this city planner, I'm sort of thinking about the data center or any other industry coming in as almost like a machine that transforms the local inputs I have into some sort of local outputs or consequences. And so on the input side, you know, the data center or the golf course or the power plant, it's going to be taking water, electricity from the grid, land, and then on the other side of it, it is going to be um, tax revenue, you know, jobs, but also negative externalities, um, pollution. So, if you were to just add that all up... Do you want to have a data center in your community or not? Maybe I'll just put the question to you. I mean, you mentioned to me before the show, you're from a small town in the rural Northeast. If you were the city planner of your hometown and someone came to you and said they wanted to build a data center, but also you had some of these other options available, maybe a golf course, maybe a mall, what would you say?

Andy Masley: It's actually very hard for me to come up outside of like a few specific cases of places where data centers haven't significantly helped local communities. And this is incredibly controversial and kind of like on an extreme end of this. But I find consistently that like, if I just ask the question, would citizens of an area be willing to sacrifice this amount of tax revenue or pay this amount of tax or revenue to avoid the specific externalities of the data center? It's often very hard to believe that they would like there are specific extreme situations with either. Like air pollution or noise pollution, where specific data centers have caused harm. And I don't think that the locals would actually make that trade. But in most places I can find across the country so far anyway, if I look at like, oh, like data centers generate way more tax revenue per unit of water than almost any other way you can use water specifically.

Jon Bateman: Help unpack this tax revenue idea. How much money are we talking about? We're in the DC area right now, pretty close to Loudoun County, Virginia, data center alley. If you drive through Loudoun County, you will see tons of signs almost everywhere you go, arguing against data centers, new transmission lines. It's a huge political issue there. Does Loudoun county get significant tax revenue from those data centers?

Andy Masley: Yeah, that county alone, so excluding all of the state level taxes, Loudoun County alone receives something like $1.3 billion every year from data centers, specifically. And so, Loudon County is already one of the wealthiest counties per capita in the country, and this specifically allows their, say, property taxes to be a few thousand dollars lower every year, specifically, if you just compare this to, like ... Oh, like the county is using about 10% of its water on the data centers specifically. So, and they're using something like 3% of the land in the county. And in return, they're generating what's effectively like a money cannon that's spitting out a $100 bill into the tax revenue like every other second, basically. And like, if you look at like their total tax base, something like half of it comes from the data center's specifically, yeah. And so like, I think that Loudon is a weird case where there's more data center capacity there than anywhere else in the world. And so I can imagine that here it's probably butting up against the point where people are upset and they are starting to encroach on people's property. There have been some famous issues with noise in Loudon we can talk about later. And so it's not that people are always wrong to be worried about them, but I think that if people had shut them down way earlier, they'd be missing out on such incredible amounts of tax revenue for their specific area that I don't really think they'd willing to sacrifice. For the sake of using a little bit less water. I did kind of a joke calculation a little while ago to try to figure out, okay, at the rates that Virginia pays in other places, if Loudoun County wanted to build a desalination plant 70 miles away and pump all that water back into Loudoun, to make up for the amount that the data centers are using would require something like 2% of their annual revenue, specifically, to just get all that back, if that makes sense. And so like... There, if we were just asking like, should we sacrifice 50% of the town's revenue to save this much water, or to avoid 3% of our land being used? I don't think many people would say yes.

Jon Bateman: That's interesting. So you're saying Loudoun County gets so much tax revenue that the county itself, if it wanted to, could fully backfill all of this water using just 2% of the financial benefits.

Andy Masley: Yeah, this is kind of a hand-wavy back-of-the-envelope calculation, so take it with a grain of salt. But like, I don't know, I think that the county doesn't even choose to do that. They choose to spend it on like, schools and stuff. And I think like, if you ask the average resident without naming data centers specifically, and if you just ask like, would you be willing to use a few additional percentage points of water? Like, for the city to draw a few additional percentage points of water, and in exchange, your schools get unbelievable amounts of funding, and your property taxes go way down. Like, is that a worthwhile trade? I think the average person would say, yeah, it's OK to use a little bit more water for that purpose. But I'm worried that when people hear the word data center, they get all these very negative associations about things they perceive to be happening. And the upsides, the potential upsides of just how much funding these things provide everywhere, including to communities like mine, who are both like, poorer and also like maybe don't want a lot of like new people moving in for the jobs specifically like that combination of like kind of like something that doesn't provide as many jobs as you would like but does provide way more tax revenue is often very Appealing to like local communities that want to mostly stay as is but like do maybe want more tax Revenue for the locals who are there. Yeah, I mean all of this is make

Jon Bateman: me think people are kind of like anti-building data centers or maybe pro-building datacenters as a sort of general principle, but how much of this is just a matchmaking problem of finding the communities where the specific sets of pros and cons that data centers represent are appealing to those communities versus not. I mean maybe like you said Loudoun County you could make a case that it's really benefited them, but then for some of those citizens they feel like reached. A saturation point and they don't want to go any further. They feel like they're already getting the tax revenue that they want and any further development is kind of not worth the benefit. Maybe there's some other community that doesn't have a data center yet and is willing to sacrifice a little bit of esthetics and some water in order for that tax revenue. Is that right? And if so, why can't we do that matchmaking better? Or is that actually happening?

Andy Masley: I think that this is actually happening at the level of local government officials who are faced with the full trade-offs that data centers are offering them. There's this very noticeable pattern where local politicians and just local officials will see the trade-off of the data center and get really excited about them. Like it's very, it's actually very easy to reach your handout. And find local officials who are very, very excited about their current or future relationship with the data center. But I'm worried that there's this kind of separate contingent of people who are growing more and more powerful who have read these very alarming, but I think kind of very misleading, popular coverage stories of data centers and are incapable of actually making the normal trade-offs where they're thinking like, oh, no matter how much tax revenue it adds, it's gonna destroy my community specifically because I've read these stories of datacenters coming in and doing something awful. And my claim is that if you actually poke at these stories even a little bit, a lot of these claims kind of fall apart. There are, again, specific places where they have been harmful, but it's way less common than people think.

Jon Bateman: So why don't we talk about a couple of these stories one set of stories came in the book Empire of AI by Karen how we have to talk about this is one of your big moments Yeah, so this was one of the biggest AI books of the last year Karen how is a well-known reporter. She wrote a kind of inside story about open AI its rise and development interviewed tons of people in the industry You know, and she took a critical eye toward a lot of aspects of the AI sector, including its water use. You read the book, you took issue with it, you wrote about it. Tell us about that and what happened next.

Andy Masley: Yeah, I do want to clarify before I start that I don't have any like hostility to like Karen's work outside of this like I think like I do worry that the AI labs do behave as if they have imperial ambitions. So I don't want to say like, oh, everything else is fine with them or that like the rest of the book is bad or anything. Basically, what it happened was, as I was writing a lot about AI and water use, I had more and more people reaching out to be being like, Oh, hey, Andy, we basically agree with your take on America anyway, where we don't think this is a big issue in America. But we are really worried about how data centers are interacting with other poorer countries that might not be able to resist them or might have more water scarcity. What do you think about that? And we, when I would ask people about this, they would usually say like, oh, the place I'm getting my information is Empire of AI specifically. It just has this really compelling chapter on water use in other areas. And to be clear, I don't actually still know the deal in a lot of other countries. Like the world's a complicated place and I don't wanna like come in steamrolling some other country and being like, it's fine actually, because I genuinely don't know. But basically, I'd opened up Empire of AI and to the chapter on water specifically and the environmental impacts. And almost immediately was finding a lot of the same mistakes or strange assumptions that I was finding in other places. There was like at the opening of her chapter, she had described AI as consuming half as much water as the UK or something like that. And if you look into her source, almost none of that water was actually consumed. Something like 97% of it was completely returned to the source unaffected. This is kind of a separate conversation, but- that had raised some alarm bells that the rest of this conversation or the rest this chapter specifically wasn't going to be entirely on the level that I wanted it to be at basically. And there's this one specific chapter where she tells this dramatic story of this local community in part of a city in Chile that is resisting a local data center tooth and nail because as she describes it, the data center is going to use a thousand times as much water as the city of 88,000 people uses. And like that sounds like a complete apocalyptic disaster. You know, like if an industry came in and used a thousand times as much water as your city, like it makes sense that the locals are fighting to the nail to prevent that from happening. I couldn't believe that that was the case when I first read that because I had just been reading a lot about data centers at the time. And I just knew that there was no building in the world using a thousand time as much water as a city of 88,000 people anywhere. And so alarm bells were going off for a little while. I was like, maybe she did something weird with words here. There's some misinterpretation. And then I had gone to the back of the book where she explained the calculation and I just run the numbers myself and realized that like the number she was using for how much the average citizen of the city used implied that each person was living on like two milliliters of water per day or something really small like that. They basically have to be living in Dune or something for this to work out basically. So.

Jon Bateman: So long story short, I think what happened here is after you brought these issues to light, she went back to her original sources and confirmed that there was the wrong unit of water was being used.

Andy Masley: Yeah. So like if something is as clean as a thousand X error, like I had to assume there was a unit error somewhere. And I realized that like the city she was drawing from literally always recorded its water in cubic meters. Um, and she had reported the water use in liters instead in a cubic meters, a thousand liters specifically. And so I was like, Oh, this seems like a promising place for where the error came in. And that turned out to be right. Basically what had happened was she had emailed the city asking for the city's water use and leaders. And I think whoever responded was just super lazy and was like, here's the number, and didn't bother to check that the unit was wrong. And so that was one error. There was a second issue where she had looked at the maximum per second water draw of the data center, like the most it would ever be allowed to draw per second. And this is for like very extreme emergencies specifically. And she had taken that absolute maximum water draw per second and multiplied it by the number of seconds in a year. And I think that also really massively overinflated the number by another like four or five X or something like that. And so by the time you make all the adjustments, it comes out to something like 10 to 20% of the city's water use. And like, that's not nothing, it's still significant, but there are a lot of other industries that also do that. There are like farms, there are factories and things like that, and the data center kind of recedes into one of many things that are using water in the area. And I Think like there, the question isn't like, oh, everything's totally fine. We should never worry about this. But. I do think we should take a more critical eye and say like, well, there could be a lot of ways the community is benefited by this as well, via tax revenue or jobs or other things like that. And so we need to consider this as like a trade off that the community is making rather than this like imperial thing that's moving into the city and like sapping them completely dry of their water, if that makes sense.

Jon Bateman: This is, I think, one of the moments where a lot of people stood up and took notice of your work, because your series of blog and social media posts documenting this issue actually resulted in Karen Howe going back and getting her publisher to issue a correction on the book, which is a rare event, not something as a journalist or author you'd ever want to do. And it's kind of an admission that a guy on the internet that has looked into these facts and figures very, very carefully, can actually get to ground truth. Sometimes better than someone else out there who maybe lacks the depth and is perhaps being influenced by narratives in order to, I guess, cause them to make some lazy mistakes they might not necessarily make in another area. I think that's one of your takeaways here, right? That there's a sort of cognitive bias that's crept in around the presumed nefariousness of data centers. You've called it a moral panic. Um, where we may judge information about them through a different lens than we might otherwise.

Andy Masley: Data centers make a very useful enemy for a lot of people right now. It's either an outlet for a justified distrust of like big tech companies and what they're doing to society more broadly. It's also just very, data centers are also just very strange and unintuitive buildings where you see something that's using so much power and it's very hard not to assume that what's happening inside is very inefficient and wasteful. And I think for a lotta people, it's kind of hard to imagine that like thousands or tens of thousands of people could be interacting with this building at any one time. And to the point where everything individually in the data center is actually quite small and even resource efficient compared to most other things that we do.

Jon Bateman: Yeah, like it's the case that these massive data centers are popping up all over the country, and in some cases, the world. It's also the case, that ChatGBT now has like a billion users. And so those two things are connected, right? Like a billion people are using these buildings. That actually suggests that they're accomplishing quite a lot with a relatively small amount of space.

Andy Masley: Yeah, and chat GPT is just one part of the full ways that AI is being used. My general question for people who are so incredibly critical of AI's general climate and water impact is, if you have any other product that's being used by a billion people every single week, what is a reasonable amount for it to emit? And what's a reasonable for it use water? And I don't expect them to have an answer on the spot. That's a really complicated question. But I would expect that if we have these buildings that Concentrate the thing that is serving that in one incredibly tiny specific place, which is a data center, which is large by the standards of buildings, but kind of tiny compared to the billions of people who are potentially interacting with them. We have to ask how reasonable is this from that perspective for the same reason that if I'm judging the environmental impacts of Coca-Cola, I also need to take into account how many people are drinking Coke around the world, if that makes sense.

Jon Bateman: Well, it's interesting, because in the climate debate... Activists who are concerned about carbon emissions and fossil fuel do sometimes converge on particular projects that do represent a concentration. Like for example, a new pipeline, right? So there were several new pipelines being built or proposed to be built in the United States and Canada. Keystone XL was a famous one. I think there's another pipeline having to do with Alberta. That then become sources of huge controversy where people do raise their hand and say, yes, I'm trying to stop global climate change. And that means stopping climate change in particular areas. And when it is concentrated in one thing, like a pipeline, that's the hill I wanna die on. I mean, is that a useful way of thinking about this? That there are people out there who think, well, yeah, we are in a kind of environmental crisis. Having to do with either global climate change or perhaps local drought in particular areas. And so therefore, if someone wants to come in and have a big new project, whether it's serving a large audience or customer base or not, like this is the hill I wanna die on, no more.

Andy Masley: My two requests for people who are thinking about which industries to cut or target for the climate specifically is one, are you considering just how many people are using the industry? Because that does seem to have some normative weight. I can talk about that more in a second. And then the second one is, are you unfairly targeting an industry just because it's new specifically? Like I do worry that there's this general narrative that no matter how individually resource efficient any new industry is, there's this implied idea that we can't have any new industry or any quote unquote new emissions until we've completely solved the climate crisis and the green energy transition. And I'm sympathetic to that as a first reaction to the situation, but I think this really normalizes a lot of the normal ways that we emit. In some ways, all emissions that we create are quote unquote, new emissions. Every time I get into my car, my car is old, but the CO2 that it emits is new and will stay in the atmosphere for a very long time. Every single time I use it. And I guess I worry that this like hyper focus on any new industry that's popping up isn't super promising as a way of thinking about the climate compared to like, oh, like we should think about this more as in like, how does this compare to all the other industries? Is this actually the most promising lever to pull? Yeah, new.

Jon Bateman: Yeah, new industries and, like, unvirtuous industries, right? I mean, it's interesting to compare to agriculture. So, you know, if you go to Loudoun County, where all these data centers are, you will also see these so-called CAFOs, Concentrated Animal Feeding Operations, commonly known as factory farms, right. If you're not familiar with these, you might not notice them right away. They are these often long, cylindrical buildings where... Thousands of chickens or hundreds of pigs might be at any given time. You might notice them if you roll down their window and then you would smell huge amounts of fecal matter drifting by. So it's interesting, right? Physically, esthetically similar to data centers. Massive footprint in terms of water use, pollution, energy use. And yet, these are valorized because it's farming. And farming and farmers are part of the American cultural fabric and are presumed to be basically good. Whereas AI is not only novel, as you said, but associated with a lot of, for many people just dark, dark ideas that they're concerned and afraid of it. And yeah, I guess then that does inflect how they view a data center.

Andy Masley: I definitely think that industrial animal agriculture is among the very most evil things that we're doing. So you've chosen an extreme example in the other direction for me. I do really worry about turning climate into kind of a cudgel to use for separate, unrelated conversations about what's valuable and what's evil in society. If a bunch of Catholic environmentalists got together and they were like, we've discovered this huge source of waste in our society, which is Protestant churches. They use so much energy and water. We need to shut them all down. Like, I would worry there that, like, they're basically making this claim that, like, we can't waste any emissions on something that is bad. And then smuggling in this contentious claim about what's bad, specifically. And so, with AI, like I'm totally fine if people, like hate AI and want to use it less and want other people to use less, too. Like, that's an open empirical claim or an open normative claim. But I really worry when people lurch from that to then say, and therefore it's bad for the- the climate because it literally adds no value at all because you know a billion people are using this stuff every week and enough smart people are using it like something like half of all you know computer programmers will say that they use it like daily or something I don't have the exact stat but it's shockingly high and my worry here is that people are smuggling a contentious ethical claim that's definitely open and I want them to be able to debate into a very empirical climate emergency sounding claim if that makes sense and like. I'd like to be able to separate those and just ask like, oh, given that some people love this and some people hate this, how much should it be allowed to emit? And I don't think the answer is zero.

Jon Bateman: Yeah, in a way, the incredible unpopularity of data centers is a test for market economy principles. These resources exist. They can be purchased in the marketplace, subject to local regulations. And so if a data center shows up and offers to purchase those resources in line with local, regional, and national regulations, then I guess the question is like, are we cool with that?

Andy Masley: I don't want people to just blindly trust massive data center companies and be like, oh, they always have my best interest at heart or whatever, you know? But I do want them to look seriously at where have these externalities actually occurred? I definitely agree with data center critics that Elon Musk's data center specifically added what I think is an unacceptable amount of local air pollution to already very overburdened, poor communities that didn't really have the political power to resist them. That's a clear, very bad externality. Like a more immediate local one is in Ashburn, Virginia. There's this ongoing issue where a data center's gas turbines are producing a huge amount of noise. For the locals, we can talk about that. So they're definitely clear places where local governments have in fact failed to protect people from specific negative externalities of data centers that have been pretty bad. But like, I worry that a lot of people are really lurching from this to then say, oh, all the other hundreds of data centers that are being built. Right now must also come with these secret externalities, and they're just steamrolling local governments into being allowed to do because they're making some backroom deals or the government see other advantages. And my request for the debate is again, like to just treat this as other industry. Like if we were looking at any other industry, would these few objectively very bad cases mean that we can never build these things again? Or do we have to understand like, what's the actual rate of this happening? Like how reasonable or unreasonable is it to expect that like. If something moves into your town and uses 1% of the total water draw, how much will that actually affect your water access and stuff like that?

Jon Bateman: I'll say the noise thing is something I'm pretty personally sympathetic to. I'm sensitive to noise myself. I happen to live in some place with a lot of ambient noise from nearby traffic. It really bothers me. I would be really upset if any kind of industrial facility, data center, or otherwise, showed up and started adding to the ambient noise in my environment. Now, I guess the solution to that is local regulators should have some kind of. Basic standard they apply to how to measure noise of different frequencies, including low frequency noise that it's like some of the most bothersome and kind of long, sort of can travel over long distance and difficult to abate. And then just apply that to all comers, right? Data centers and otherwise. I don't know if that's what's happening though.

Andy Masley: Yeah, I've had a few different city planner friends who have specifically told me that, man, we are just so bad at regulating noise right now specifically. There are very clear success stories, but data centers occupy this really unfortunate situation where, to be clear, not all data centers make huge amounts of noise. Quite a few of them are mostly completely silent. I've been around a lot of them while they're at full blast. And so there are different parts of the data center that can make noise. But yeah, I mean like. When a data center is making noise, the noise is often very constant. It's low enough to kind of get underneath specific like thresholds of when it's actually regulated. And like you said, it can sometimes be lower frequency and that means like traveling far distances or be less open to being regulated as well. And, you know, like I'll deal with noise pollution. Like I live in the middle of DC. Like there's some construction happening next to me right now. And also like my neighbor's dogs keep barking and that's like annoying. But I think that if you added a low constant hum to my life, that would make me go crazy. It's like, oh, it's just this 24-hour thing forever. I can see why that's specifically so terrible. And cities definitely do have ways of avoiding this. And some of this is downstream of specific plans for data centers falling apart and data centers not being hooked up to the grids quickly enough and this just bad intersecting failure that happens. And so, like... I definitely think this is one of the most serious problems that I worry about the most with data centers is like the noise issue, but also there are like clear regulatory ways that we can avoid this. And also just like, if we just say, oh, you know what, there were some serious bad noise issues, that means we should never have this industry at all. I think that would just cause people to miss out on, among other things, just like billions of dollars in tax revenue for their regions and stuff in some places that could really use them. And utility revenue. I think it's also sometimes under discussed how many of America's water problems don't actually come from like a net lack of raw water. It comes from like old aging water infrastructure, like either lead pipes or other things that they really need a large, consistent, very wealthy buyer to come in and like provide utility revenue to fix basically. So like, yeah, I basically worry that a blanket like, oh, because these specific things have happened, we can never have these things again, like really steamrolls a lot of the positives that can happen.

Jon Bateman: So, we're getting toward the end here, but I think you're raising an important question about how one big industrial player coming into a community and drawing on that resource could affect the price or availability of that resource for others. So a lot of the concern that people have has to do with their water bills and their electricity bills. And the theory or intuition there is that there's now more competition, more demand for this scarce resource, and therefore everyone's gonna be paying more for it, including local residential customers. Is that true?

Andy Masley: For water, I'm pretty convinced that hasn't happened almost anywhere for two different reasons. One is that in a lot of places, water use is actually pretty positive some because if the utilities get more revenue, they can build out more of their service area, they can improve their infrastructure and stuff like that. There are a lot places where data centers have actually reduced the cost of new utility builds that would have otherwise been passed on to consumers specifically. And then water markets specifically and electricity markets to some extent or, um. Separated by like commercial versus residential use. And so if a data center does start to compete with other water uses in the area, it's gonna cause like other commercial or industrial uses, like their bills to go up, which is, it can be a problem, but even there, I'm not really clear on like almost anywhere where that's actually caused a problem. I'm basically aware of one spot in the country where data centers have actually raised people's water bills so far. There's one specific place in Georgia. Where they accelerated the need for this new water treatment plant that would have been built anyway, but just a few years later. And that specific cost of accelerating the bill got passed on to residents specifically. So like, I think the idea that they raised water bills is very like, you know, it's in the water. Like a lot of people will be like, oh, obviously when a data center moves to town, your bills go up. But at least in terms of water, I'm not seeing a place where that's happened yet outside of this one specific Georgia community. And then electricity is much more complicated. Electricity markets are just so hard to understand. From so many different directions. There, the Lawrence Berkeley National Laboratory had studied the general relationship between data center builds and residential electricity prices at the state level anyway. And they found this really weird pattern where the states with the largest new data center bills with the most new wattage going to them also had the lowest increases in residential bills, basically. So residential bills have been rising everywhere, partly due to inflation. But the states with the most new data centers actually had some of the least additional cost to electricity. So, yeah, a lot to say about that. I think that both electricity and water markets are actually not at all zero sum. They don't follow a simple supply-demand curve because the supply itself is also changing a lot. And so, again, I'm not asking people to never be critical of data centers, but to just look at what do we actually know about these things? Do they actually raise electricity and Water bills in the ways that... Is often reported, and I think we have a few examples of them raising electricity bills a little bit in specific regions, but it's not anywhere near what people will normally talk about. And I think, again, I worry about losing out on the upsides of like, oh, you know, all the tax revenue for your electricity bill going up by a dollar per month.

Jon Bateman: One of the things I admire about your work is that you often invite your critics to come into the picture and offer you any contrary data points. So on behalf of Andy, I just want to say to our audience, if you are aware of anything that contradicts, anything that Andy has said, let him know. And I'm sure we will see that in a future blog post. I'd love to hear your closing comments here. You have spent the last couple years of your life diving deep. Into a very arcane and data-rich area where most of the public and the common tariat simply has bought into myths and misunderstandings. What has that taught you about the nature of the world? What insight does that give you into things beyond data centers?

Andy Masley: Yeah, I think that there's a lot of motivation for everyday people to kind of construct this elaborate picture in their head of what expert communities believe. Like I think a lot People have this understanding in their head that all experts who look into this understand that AI is uniquely bad for the environment and data centers are uniquely bad for the environmental compared to other things. And I find that that mostly evaporates. There are specific experts who will say this, specific experts to disagree with them, but there's not anything like the consensus that people understand. And so I think this has made me, I was always pretty deferential to expert opinion. And I think if anything made me more deferential to expert consensus where it does exist. But also much more cautious of people's general motivations around constructing the perception of an expert consensus where it might not specifically. And I think there's just so many reasons why people would wanna do that. It feels good to be able to steamroll your enemies and identify someone you hate, like say the big tech companies or AI or something, and feel like you have this really powerful weapon against them. And I that it takes a certain amount of emotional restraint. To not start to construct this elaborate story in your head without like grounding it first and what's actually happening beyond like your immediate perception of like, oh, well, all good people say this. So it must be true. Like I think like it's pushed me to like seek out where the expert communities actually are and like what they're actually saying and like take things with a little bit less of a grain of salt than I did before.

Jon Bateman: Andy Masley, thanks for joining us today.

Andy Masley: Yeah, thanks so much, Sean, I had a great time.