[00:00:22] Bridget Scanlon: I would like to welcome Landon Marston to the podcast. Landon is a professor in environmental and water resources engineering at Virginia Tech University. And his research focuses on developing solutions towards more sustainable water management. And he has quantified how and where water is used in the economy and the related environmental impacts.
And today, Landon, we are going to focus on data centers, which are on a lot of people's minds. So thank you so much for joining me today.
[00:00:59] Landon Marston: Yeah. Thanks for having me, Bridget.
[00:01:00] Bridget Scanlon: So I was very impressed with your paper that was published in 2021 in Environmental Research Letters about data centers. And it really felt like it helped me understand the water impacts, and the carbon impacts of these centers, and you provide a lot of context for that. And with so many communities concerned about these environmental impacts and footprints, maybe you can just describe a little bit what motivated you to do that study, and what you felt like maybe were surprises from the work.
[00:01:33] Landon Marston: Sure. Yeah. I mean, this research really started, maybe 2018 or 17, is when I started thinking about this. Just started a new position as an assistant professor at that time at Kansas State University. And seeing some just popular media articles talking about the energy consumption of data centers and being a kind of a person with expertise in water resources and working at the water energy nexus, I knew that if you're using lots of energy, you're also going to be likely using lots of water.
And so just thinking about what the implications of that on growth of data centers is going to be in terms of water. At the time, this isn't going back, eight years or really a decade ago, there was this thought that this, the growth of data centers, and specifically the electricity demand of data centers, was going to continue to grow exponentially.
There were some media articles that were loosely based off scientific studies, a few years prior to that, that suggested by this time, 2025 or so, a third of the global energy demand was going to be for data centers alone.
Now, what we found, one of my collaborators in 2016 published a paper that showed the upward trajectory from 2000 to 2010 and data centers growth, but also their electricity demand was growing quite rapidly, right? And so that, the assumption, I think, in some of those earlier studies was that growth is going to continue on indefinitely, right? But because energy's such a large bottleneck, as we're seeing right now with the expansion of AI, but also because it's quite expensive in the operation of these data centers, there was a lot of inefficiencies that were found.
And so what we saw over the course of 2010 to around 2020 or so, 2018, about a 6 and a half times growth in data center computation, but only about a 6% growth in electricity usage. And a lot of that stems from the fact that, again, trying to reduce the inefficiencies at the specific site. And so this means, more efficient cooling, server operation, and just really trying to lower the overall kind of electricitydemand of these centers.
And so you start to see a shift from like these kind of local, in-house data centers to what we're seeing more and more now, which are these hyperscale data centers, which can take advantage of these economies of scale. And so in 2018, it's really about when we started the study at that point, globally, data centers, the most recent estimates at that point were about data centers consumed about 1% of the global electricity demand, which is about a little over 200 terawatt hours. And so for the audience that's listening, you might not be familiar with terawatt hours. You're probably familiar with the kilowatt hours, which are going to be, for your light bulbs, 60 kilowatts. And if you run that for an hour, that would be a 60 kilowatt hour. So just to kind of put that in perspective, the terawatt hours. That's about enough electricity to power a hundred million homes for an hour, right? Or I'll put another way, an entire state of California for about 1.5 weeks. So we're talking huge amounts of electricity that's being consumed by data centers, and this is continuing to grow over time.
Now, where I'm from in Virginia, this is like the data center capital of the world. 30% of global data center capacity is here in this, in the state of Virginia, located close to Washington, DC, Northern Virginia area for the most part. They have a lot of key things that data centers are looking for, cheap land, key fiber optics reliable power sources, skilled labor. Access to clients, particularly the federal government and some of the contractors there. And so we're seeing a lot of growth in that area. And I talked to folks in the area concerned about data centers coming in, and electricity use, and noise and water consumption.
So back to the electricity use specifically. We're looking at how this is growing in the US specifically in 2018, which is kind of the base year for our, when our study was done in 2021, about 76 terawatt hours of electricity was used to power data centers. And this represents about 1.9% of US electricity use.
My colleagues and my former PhD student, as well as several of their co-authors published a report recently for the Department of Energy, specifically Lawrence Berkeley National Lab, showing in 2023, so five years later, that 76 Terawatt hours had jumped from almost a hundred terawatt hours. So going from 76 to 176 terawatt hours in the matter of five years.
So now we're looking instead of 1.9% of U.S. total electricity as dedicated data centers, as of 2023, that was 4.4%. Now their projections into 2028, is 325 to 580 terawatt hours, so roughly 6.7 to 12% of US electricity demand, looking now about three years from now, will be dedicated to data centers if these projections hold.
Now there's a lot of uncertainty and kind of in that range. Largely because no one, even the industry themselves, are not really sure how the next few years are going to play out. We're seeing rapid growth and expansion of data centers, new technologies are emerging. And so it's really unclear trying to make these projections. We're shooting in the dark.
And that in fact, talking to some, the authors of this report, my collaborators, and again, former PhD student, they can only make it to 2028 really because of that significant uncertainty. And so they don't want to really project too much beyond that because we don't really know what that's going to look like.
But the key takeaway here is that data centers use a tremendous amount of energy. And as we'll talk about later in this conversation, because of that, have significant greenhouse gas emissions related to their operation as well as water use.
[00:07:01] Bridget Scanlon: Right. So, that was really nice to set the stage and put it in context, when you based your study that was published in '21 on 2018 data, it was about 2% of US electricity and then more than doubled in about five years to 4.4%. And then the projections then are about from 7 to 12% in 2028.
When you produce electricity, you oftentimes use a lot of water for cooling and stuff like that. And one thing about data centers, you could think, well, maybe they could just operate during the day when you've got renewable energy, or they could operate at night to adapt to availability.
But these data centers they operate 24 7, 365. So, uninterruptible power supply. So that's another challenge, I think, for the data centers, you can't just shut them down if you don't have the electricity or the water readily like you can do for other sectors.
[00:07:56] Landon Marston: Right, exactly. And I mean, I think there is some flexibility, and that's an area of research or looking to like what workloads could potentially be shifted during different parts of the day or different times of the year, to perhaps reduce that electricity demand and environmental footprint. And really for data center operators, cost is really what the main thing they're considering. But for the most part, we think about data centers, think a lot of people immediately go to AI because that's kind of the hot thing right now.
But this Zoom call is using a data center. We go home tonight and watching Netflix data centers a lot of our everyday life, is everyone is really dependent upon data centers, even if we don't realize it, just because it's the backbone of our IT technology, and all of the things that we take for granted each and every day. And those things are always on demand and always available to us.
And so, it's really important not only for critical tasks, and like AI and for work, storing enterprise level data, and being able to access that. But even for, like I said, some of the.. I don't say trivial things, but some of the entertainment things that we enjoy every day those are also going to be serviced by data centers and often readily available 24 7 whenever we want it.
It's there on demand. So those data centers have to be up and operational to cater to that demand.
[00:09:05] Bridget Scanlon: Right. And you mentioned that Virginia is the data center capital. And Northern Virginia, they, I think they use the term data center alley, but the US is the global leader. And it's kind of hard to get some of these numbers. I mean, some results when I was looking up, they were saying it's very difficult to quantify, but statistics suggests that maybe the US in 2024, as of March 24, had about 5,000 data centers.
And the next country was Germany with about 500. So 10 times less. And Great Britain with a similar 500. So US is the leader and within the US, then, Virginia is the leader. Maybe it's not something you want to be leading in.
As you say, you're next to DC and you've got good fiber optics and all of that. It is not clear to me, how critical it is to be next to major urban areas, or the streaming and all of that sort of thing, how critical that is, or if you could put them out in the middle of nowhere. But you wouldn't have probably the transmission in all the other infrastructure to support it. I guess that's part of it also.
[00:10:07] Landon Marston: Yeah. As, as far as the location of data centers, and this is one of the things that we talked about in our study as far as reducing the environmental footprint, one way that we could do that is by simply thinking more strategically about where we place these data centers. I know we're going to get to that a little bit later in our conversation, but as far as the decision making and placement of these data centers and how companies go about doing that, we've had some conversations with some of the large multinational [companies] that have data centers globally. And at the time, this was in shortly after our paper was published in 2021. Perhaps things have changed, but they said they had a list of about a hundred different factors that they consider when trying to site a data center, and water wasn't on that list anywhere.
And so this isn't really a key consideration, at least at that point. I think that's probably changed since then. Many companies, including that one, have put public facing documents that have claimed, aim to be water neutral or even negative by 2030, or they always set some date farther into the future.
But there's, so I think it's at least on their radar now. And communities have started to push back in terms of water and water utilities, what that means for rate payers and things like that. So, those are starting to be considered. Now, going back to just the siting issue in general, a lot of times some of those things that come, that do come into play, obviously, cost of land, we talked about fiber and major infrastructure, so electricity, also could be, water infrastructure and increasingly more and more. Workforce, that's sometimes a role.
But what you're alluding to earlier, Bridget, was proximity to major cities and that can be really important for certain types of data centers where they're always transmitting data back and forth. So that's a lot of things, again, back to the Zoom call, things like latency is a big factor.
And so they want to make sure that they're able, any consumer in these population centers can ding their data center. And that comes back in, in a short amount of time. The longer the distance you have, the longer that latency is going to be. And so that's going to have implications on their, their operations and kind of customer satisfaction.
And so those type of, many of those data centers that the large populations like we're talking about, it's their benefit to put them close to large population centers. And again, usually it's in the suburbs where land's a bit cheaper, kind of out in the outskirts a bit, but close- but still fairly close to the large city.
Now some of these other types of data centers that are, for instance, the ones that are being used to train AI models, they don't necessarily have that requirement. You're seeing sometimes these being put in more remote places simply because they're there to train these models, not necessarily for the latency times near major population centers.
And so that would be one kind of key distinction. What I've probably been talking about at this point is like these hyperscale, these really large data centers, probably what most people envision, like warehouse size data centers that are holding, servers for Meta, or Amazon, or Microsoft, or Google.
But you also have like enterprise, or smaller type of data centers that would be located for, like, for instance, Virginia Tech here has their own data center. I'm sure Texas, UT has their own data center. Large corporations, they'll have their own data center, which are usually much less efficient and much smaller.
But there's some advantage to have those on site, as far as data security and privacy and things like that.
[00:13:05] Bridget Scanlon: Right. Landon, you mentioned earlier on, maybe 2010 to 2018 or 16 or 18, that had a big increase in the computational workloads and stuff like that, about more than a six fold increase in that. But then on the electricity consumption only increased by 6%.
And you also refer to, these in-house or on-premise data centers, which is kind of what we have at the university and stuff. So these are small data centers. And then the other type, I guess is co-located, or cloud centers, but then the hyperscale centers that you mentioned, these are the biggies. And the type of data centers has been changing over the past decade and a half.
Now they talk about accelerated data centers that focus on AI and machine learning and CPUs and GPUs and all of these sorts of things. So it's really rapidly evolving. So are the energy requirements and the related water requirements changing, I guess.
[00:14:01] Landon Marston: Right. Absolutely.
[00:14:03] Bridget Scanlon: And in your study then in environmental research letters, can you describe how you calculated the electricity usage?
Both the internal and external, and then the related water use?
[00:14:16] Landon Marston: Absolutely. Yeah. So. I think the first thing to start out with, and I should maybe preface this by saying this isn't a unique problem to data centers. This is really for any sector. We don't have great water use data. Like when I say great water use data. I mean metered water use that tells us exactly who's using the water, how much they're using, when they use it.
There's not a comprehensive national database. Or even, or for that matter, I'm not aware of any nation that has that type of data publicly available for any and every industry. And so, what we had to do to estimate both the water use as well as electricity use, which is also not publicly available.
We had to use different coefficients. So one of those is the power usage effectiveness. Some, it's often called PUE. And so this is the total power supplied to the data center. And so this encompasses everything from lighting, air conditioning, even like the bathroom facility and things like that, divided by the power consumed specifically for the IT equipment.
And so an ideal situation would be a value of PUE of one. That essentially means all the electricity, all the energy going into that system is for the IT equipment, the servers and whatnot. But realistically that's never been achieved. The lowest values I've seen, these are sometimes even like more just kind of research studies haven't been actually done at scale or like 1.04, Google's shown 1.07.
But most of these hyperscale centers have values of 1.2, maybe 1.3. Basically, that means about 30% of that total electricity usage is not for the IT equipment. It's for other things, right? But when you get to like, these enterprise or kind of in-house data centers, you might be looking at values closer to two. Meaning about half is going to the IT equipment, half's going to other things.
And so we use these PUE values that come from the literature. As well as information on the area of the data centers. And so at the time, I haven't looked at this recently, but my, the PhD student, AB Siddik going to give him a shout out here. He was looking, we're trying to figure out, okay, can we site, can we figure out where these data centers are located?
We talked about hyperscale already. And so those are those big players and these are these massive data centers. And so a lot of times you can find for some of these larger players, at least I don't know about now, but at least at the time you could find these things online. And so we were able to locate these and get some basic information on them.
We also talked about the in-house data centers. Those are a little bit tricky. Our collaborator, Armand Shababi, gave us some data from Lawrence Berkeley National Lab and some of his previous research to get these kind of smaller scale data centers. And then the third category is co-location. And so these are basically data centers that rent out space within their warehouse or data center compound to different companies to place their servers and IT equipment. And so because they are client facing they're trying to recruit people to come in, they have a lot of information online that we were able to extract. We were able to basically do some web scraping and pull all this information and then get information on things like the square footage, right? So this is what one thing we needed mixed with the PUE and then the values basically how much electricity is used per square foot, right? So we're taking all the information, and that gives us an idea of how much electricity, knowing the square footage of the facility, it's PUE based on whether it's a co-location, or hyperscale, or on site. And then able to translate that into expected electricity usage.
[00:17:42] Bridget Scanlon: Right. And that's great. I mean, that takes a lot of forensics, Landon.
[00:17:47] Landon Marston: it's hard now. It's even harder now because since that study that some of these companies have pulled things where we can't access it in the same way as we could before. So I don't know if that was because of our study or not. But it's even harder to get these data now 'cause a lot of things are behind paywalls and not made public facing as they were, five, six years ago.
[00:18:04] Bridget Scanlon: Right, right. And so then you have the electricity that you direct use to cool the system in the facility, but also the indirect energy use to supply to the system. And that's oftentimes coming just from the grid. And then also the energy that you use to treat the water and the wastewater.
So maybe you can describe a little bit how much of the electricity is directly used for the system from your analysis, versus indirect electricity, and then the related water consumption.
[00:18:34] Landon Marston: Sure. So, so most of electricity is used at the data center itself, and so you're alluding to the scope or the boundaries of our analysis. And so if you think about a data center kind of in the middle and we would have different kind of resources coming in, right? So one of those is going to be the power plant sending electricity to the data center.
Another would be the water treatment plant. So it's going to be sending water to the data center, but it takes a tremendous amount of electricity to treat and pump that water. And so you can think about the electricity also going to the water treatment plant, which is then could be attributed back to the data center because that treated water again, requires electricity, but that treated water is for the data center.
Same thing with wastewater. That wastewater is leaving the data center, being treated by the wastewater facility, and then being dissipated. But that treatment of that wastewater also requires electricity. And so if we were to attribute that electricity use of the wastewater plant to the data center, based on how much ever wastewater it sends to that wastewater treatment plant, we're able to account for the total electricity use of the operation of that data center.
And so having said all that, this broad scope looking at not only the direct electricity going the data center, but also this indirect electricity going through the water and wastewater treatment plant, the vast majority of electricity use for data center is actually attributed that direct linkage from the power plant to that data center itself.
But there is some, there is a- I can't remember the number off the top of my head. A couple, a few percentages of electricity use are coming from that wastewater and water treatment plant.
[00:20:04] Bridget Scanlon: Right, and then to estimate the water usage, then you have to figure out what power plants are contributing electricity to the data center and what the energy source is, the fuel type, and then the water consumption of that to estimate the related water usage.
[00:20:23] Landon Marston: Absolutely. And so as you alluded to, there's different types of power plants that produce electricity on the grid. So solar, wind, thermal, electric, which encompasses natural gas and coal, and even nuclear. That thermal electric component that I just mentioned requires a significant amount of water to run those facilities. And that's used for, they withdraw a lot of water and depending on the coolingtechnology they use, they might consume a decent amount of that water as well. And so what we wanted to do is to be able to connect those data centers. Not only to the kind of the broad grid, but to specific power plants to be able to then relate that energy usage of that data center back to the power plant and then ultimately to the water that it consumed.
And so what we did there is we looked at what's called power control areas. And so these are basically... you think of electricity grid taking up the entire United States and there's kind of different subsections of that grid that are operated together. And so we had data from EIA that looked at not only the amount of electricity generated in each of these areas, but also imports and electricity outputs from each of these regions. And so using all that information as well as the location of where the data center was at, which power control area it was actually physically located in. We were able to determine kind of how much electricity it's getting from different power plants. And then from that, estimate its water consumption based on the power plant type.
And we actually have pretty good data on the water consumption associated with these different power plants at different times of the year and whatnot. So we were able to get estimates of total water withdrawals and water consumption associated with each of these data centers' electricity use.
And we can also then look at those wastewater and water treatment plants. And do the same thing, connect that back to the specific power plants, figure out the water usage. So get that indirect water usage as well. And so this is the indirect water use, and then we can talk more about the direct water use.
But I think you maybe had a question, Bridget?
[00:22:15] Bridget Scanlon: Oh no. I mean, Energy Information Agency has a lot of data on the electricity from power plants generated by different power plants. And then if they're above a certain size, and I think a hundred megawatts or whatever, then they report the water usage. So, so that's a great source of information.
The data have been improving tremendously over the past decade or so. So, that's really nice. And so then, tell us Landon, how much water was used for the data centers in 2018 then, U.S, wide, and how that compares, to other sectors?
You've done a lot of work on water use for different sectors, and we both know that irrigated agriculture is usually the elephant.
[00:22:56] Landon Marston: Right. Yeah. And so, I mentioned earlier about the indirect water use associated with electricity and also these electricity associated with the wastewater and water treatment plants. Then we have the direct water consumption. So this is the water that's used directly within the data center itself, and this is primarily used for cooling. And so these data centers generate a tremendous amount of heat. The servers generate tremendous amount of heat, and so they have to they often use water to help cool that. It's a relatively inexpensive method of cooling, and reduces the electricity demand.
And so since water is relatively cheap, especially compared to energy, many data centers will opt to use a more water intensive, but less energy intensive cooling method. And so we see that quite often. So now to your question of how much water was consumed by data centers. We estimated in the year 2018 about 0.51 cubic kilometers.
And so that's about 0.42 million acre feet. And so those units probably seem, maybe hard to like wrap your head around, even in water and sometimes I even struggle with this. So maybe to help put this in context of the total economy's water consumption.
So I should, before I get into that, I guess I should make one clear distinction here.
Water withdrawals is how much water is diverted from a source. So a lake or a river or an aquifer. And some of that water might be consumed. And so when we talk about consumption, we're talking about the amount that is no longer going back to the water source. So it's going to usually evaporate, or transpired, or just leaves the basin.
And then you have a portion of those withdrawals again consumed. The other portion might be return flows. And so this would be what would return back to the source, maybe through a wastewater treatment plant, for instance, or in other contexts it might just run off a farm field, for instance, and make its way back to the river. And so oftentimes the consumption amount was always going to be equal to or smaller than the withdrawn amount because again, some of that's going to return back to source.
Alright, so now with that, put aside, let's talk about the consumption of water by data centers.
Again, 0.51 cubic kilometers, 0.42 million acre feet. To put this in context, in kind of that 2010 era, we published a paper in 2018 in water resources research that looked at the water footprint of the US economy. And those numbers are based off around 2010, 2012, kind of in that timeframe.
And we estimated that the total US water consumption was 128 cubic kilometers. So again, remember data centers in 2018, 0.51 cubic kilometers. US total economy, 128 cubic kilometers.
So we're looking at around point less than 0.4% of US consumption in 2018 could be attributed to data centers.
Now if we look at total water withdrawals, about 445 cubic kilometers of water was withdrawn in 2015, according to the United States Geological Survey. Some studies have suggested that about 80% of, give or take 10%, about 80% of the water that's withdrawn by data centers, is, evaporated for cooling.
If we assume that that's mostly true, that means we're looking at 0.6 cubic kilometers, maybe 0.7 cubic kilometers of water is withdrawn or delivered to data centers. So again, putting that in perspective of the 445 cubic kilometers of water that's withdrawn nationwide for the entire economy, we're looking at 0.2% or something like that, 0.1%. So a relatively nationwide, really small percentage of our overall water consumption and water withdrawals can be attributed to data centers.
[00:26:25] Bridget Scanlon: Right, and so, then you would think, okay, that's a necessary but not sufficient condition for water to be a problem. Right.
[00:26:32] Landon Marston: Right.
[00:26:33] Bridget Scanlon: Because, okay. It's small, on the large scale of the U.S. But then the problem is then these data centers are located in different areas, and so at a much more local scale, then that can be a much higher percentage of the water that's being used.
And then the Berkeley report also had some numbers on water use, which were similar to what you had. And also projecting water use increasing in the future with the increased electricity demand. And I think in your report also, you looked at water intensity.
Well, you have similar water intensity for the different types of data centers, like the onsite or internal ones versus the hyperscale. Oh, but the water intensity per unit of workload was much higher for the inefficient, internal onsite data centers. About six times more than your hyperscale centers.
[00:27:24] Landon Marston: Right, right. Yes. And so that reflects the fact that these internal data centers, so again, these enterprise data centers that are within, inside of an individual company or university, typically much less efficient, don't operate at full capacity, but still have similar cooling demands. Not quite the same, but there you obviously see economies of scale as you have these data centers go from these kind of local data centers, to now these warehouse size data centers.
And because of those economies of scale, that's not only for number of workloads, but also for electricity demand, and ultimately for water and carbon intensity as well, which we haven't talked to them about that as much. You don't see those scale quite the same way as you do for water, but you do see significant improvements of that intensity per workload.
[00:28:08] Bridget Scanlon: So one nice thing about that article, then, in environmental research letters, trying to distinguish the water footprint, which is the water consumption. And thanks for describing the difference between water withdrawal and water consumption. 'Cause oftentimes we get those confused.
So, the water footprint. And then in your paper you talk about the water scarcity footprint. So the water footprint may be low, but if you have no water in the region or very limited water supplies in the region, the water scarcity footprint might be pretty high. So maybe describe that a little bit.
And you did a really nice map of the watersheds in the US, and looking at the water footprint and the water scarcity footprint. That was excellent.
[00:28:49] Landon Marston: Thank you. Yeah.
Water footprint is basically the consumptive amount that we were talking about earlier. And it can also not only be the direct consumptive amount that might be happening at the data center, about 25% of the water footprint is happening locally at the data center.
So again, mainly for cooling, and about 75% is going to be this indirect water footprint. So again, water consumption that's happening at the power plant, or where it's receiving its electricity. And as well as this, a nominal amount also for the water utilities that are providing the data center water.
And so, if we look at kind of these values and comparing the water footprint to the water scarcity footprint, we start to see some divergence between these, and that's because some areas are, perhaps inefficient or you consume a lot of water, but water's relatively abundant in that area.
The water footprint metric doesn't account for whether or not there's ample water supplies locally to meet those demands. And so whether you're talking about the water that's consumed or withdrawn at a power plant, highly localized, right? Like it's withdrawing a large amount, kind of from one location.
Water footprint will encapsulate that volume, but it doesn't tell us anything about whether there's, again, ample supplies available for other uses and for including environmental uses. The water scarcity footprint on the other hand, gives us a little more insights, not only to that water volume, which is incorporated or kind of encapsulated in that calculation, but also the relative scarcity of that water in that particular location, again, whether it's coming from a power plant, withdrawing water, and then generating electricity, which is sent to the data center, or it's direct withdrawals that are going to the water treatment plant and then eventually to the data center itself and then used within the data center.
And so what we found is that, we kind of see really these data centers are scattered across the entire United States and they're withdrawing from almost every basin with watershed within the United States, both either directly through their own deliveries from the water treatment plants or indirectly through the power that they're pulling from, their different energy sources.
And, there's obviously hotspots even though they're kind of spread across the nation. You see hotspots around major cities. We already mentioned Virginia and kind of this corridor that really stretches down to North Carolina. And you're seeing it all the way down to Atlanta, really you're seeing a lot of data centers in those regions.
In Southwest, you see several up in the the Washington, Oregon region. You see several data centers as well. And so again, part of that's cheap electricity, siting, all the things we mentioned at the beginning of this conversation are some of the reasons why those data centers are located in the places that they are.
But as you can imagine, like, we've got ample water and for the most part, in most of the places in Virginia that we're talking about, this data center alley that you mentioned, which you're looking down into along North Carolina, and Atlanta, and those regions.
But when you go out to the Southwest, it's a very different story. And so, while they might be using similar amounts of water, perhaps more, it's going to be amplified by the fact there's less, much less water available there. And so because of that, that's going to make this water scarcity footprint, this metric, be much more elevated compared to data centers using the same amount of water in Virginia, for instance. And that's, again, reflecting the relative scarcity in these two different locations.
And so we wanted to, in this study, we felt it appropriate to look at that, to use this water scarcity metric in addition to water footprint. To not only tell us the volume of water that's being consumed by these data centers, but what its impacts might be at a local level on the particular watershed.
[00:32:15] Bridget Scanlon: Yeah, and I think that was extremely valuable. And while the efficiency of these data centers may be improving and stuff over time, and that might help with the total water footprint. But the water scarcity, water is just much scarcer in the Southwest U.S, even if they are using less water.
And so, maybe I'm not sure if I'm correct, so maybe 75% of the water is indirect, and it's used in the electricity at the power plant. And so it also depends on what type of power plant you have, have a thermal electric cooling versus a renewable energy and things like that.
I was looking at some stuff and they were talking about sustainable data centers or green data centers and stuff. And so, that would suggest maybe they would try to use more renewable energy, and then that could reduce the water use. But, a lot of the renewable energy, the solar you would think would be in, warm climates where the water could be scarce. Or you could make it green by buying credits and offsetting your energy.
So talk to us a little bit about the trade offs between water and energy and where you locate them. I know you included that in your analysis.
[00:33:28] Landon Marston: Yeah, certainly. And so to kind of touch on your first point that you're making about these kind of green data centers and relying on renewable energy sources, whether that's wind or solar or hydropower, geothermal, there's another one. Wind and solar requires very little water. Hydropower, depending on how you account for it and that, this, not to get too academic about this, but like, there's different ways to attribute the evaporation that's happening on these large reservoirs behind the dams that are generating hydropower. And so depending on how that attribution is made, you could have a really small water footprint or a really, really large water footprint.
And so this is very difficult to make those attributions clearly. So it's a little bit of kind of preference in some ways. And so what we did in our study, we actually kind of did a sensitivity analysis where we looked at these kind of two extremes of a way of attributing that water consumption evaporation that's happening behind that dam to attribute these to data centers. And again, kind of the two extremes to kind of put an envelope around it. But there are data centers that are, you're right, like having contracts with power providers to secure clean energy with solar, wind hydropower as well. And I think one of the things that we'll see moving forward, or I'm hoping at least, is that there will continue being this transition within our energy portfolio toward more renewable sources of energy.
And as we do, that's going to, in some ways, take care, help take care of the exceedingly large electricity demands, and with that water consumption, because as I noted earlier for solar and wind, don't really require a lot of water use. And so as we get the co-benefits of not only reducing the carbon emissions of our electricity grid, and this is again, data centers are one of the largest users, electricity of any industry.
So we haven't really talked about the carbon piece here much, but that's usually what a lot of people think about when they're thinking about moving to renewables on the grid. But we're going to have this co-benefit of also reducing our water consumption when we move to more solar and to more wind.
And that's going to lower this water footprint of data centers because as you noted, Bridget, 75% of the water consumption associated data centers is indirect. Going again, back to the power plants and it's water consumption. And so by changing those sources of energy, we're going to be able to effectively reduce its water footprint at the same time. And so, I think that's one strategy that we're going to see moving forward.
Now, another thing, as far as the trade-offs that we might see. Let me, before we get into the water and the carbon, those type of things, let me just talk more broadly about infrastructure and design.
I think what we're going to start seeing more of is data centers, particularly these large hyperscale data centers that are training AI. We're seeing this in the news headlines and these companies are coming out directly and saying it like energy is a key bottleneck in their advancement and their kind of build out of data centers.
And so the Department of Energy is working with some of these big players in the data center industry and AI, to secure cheap land, maybe on federal lands or oftentimes in close proximity to existing, planned or perhaps rehabilitated data centers. And this has a couple advantages beyond maybe just cheap land and perhaps cheap power is not only, we talked primarily about power generation. But you also need power transmission to get that energy from where it's generated to where it's ultimately used. And that's really expensive. And frankly, our infrastructure isn't built to handle this expected capacity. We talked at the beginning of this conversation going from, 1.8% to 4.4% of US electricity demand going to data centers with the next couple years going to six point, whatever it was, 6.5 to like 12.2.
We're going to be expanding this rapidly and our, it's not only our generation capacity, but our transmission capacity as well. And so a lot of these data centers are thinking, can we put our data centers right next to a power plant? And that would then lower the amount of, build out with this infrastructure, these transmission lines and transformers and things, all this stuff that's needed in order to deliver that electricity.
And so I think that's one thing that we'll see. And another advantage of that is it's usually in close proximity to large bodies of water, these especially thermal electric power plants, I should say. And so, we'll, that might help kind of with securing secure water supplies. It depends where you're at in the country. There's whole water rights. We don't get into all that stuff, but those are things that you'll see I think play out as well.
So now back to your original question, which is the trade-offs between water, energy and carbon. So I alluded to the fact earlier that there's different types of cooling technologies that are used in data centers. I won't go through all of those, but just summarizing it briefly. Basically there are some cooling technologies that use, that require tremendous amount electricity to operate, but use little to no water. And then on the other extreme, you have some types of cooling systems that use a lot of water, but don't require a ton of electricity, right?
And so that in itself involves complex trade-offs because as we already discussed, you not only have the direct water usage by the data center itself used primarily for cooling, but you also have the water use associated with electricity. And it really depends on what type of power plants are being used. Solar, wind, very little water. Versus thermal, electric, natural gas, coal, nuclear, those type require quite a bit of water. And so depending on what the source of electricity is, your cooling system, you say, well, I don't really want to, I really care about water a lot, so I'm going to use this, less water intensive cooling technology.
It's going to use a little more energy, but a lot less water. You actually might, in the end be using more water because you're using more electricity and that electricity is very water intensive. And so it's really on a case by case basis. We have to kind of think through these things and look at a local scale.
We've recently published this was the end of 2024, beginning of 2025. This was B Siddik. So again, the PhD student that led our data center paper in 2021. Armand and Prakash and other guys, both those guys at Lawrence Berkeley National Lab, published a paper that created not only this academic paper and methodological framework, but a website hosted by Lawrence Berkeley National Lab, where you can go in and type in any, your, at your personal address. You can type in the address of the data center, and it will give you an estimate at that for whatever time period that you specify, of not only the electricity, if you're going to use a one megawatt hour of electricity, what is the corresponding water scarcity footprint? What is the water footprint, and what is the carbon footprint associated with that electricity demand? And so tools like that I think will be really helpful to help understand these trade-offs, and help companies, my hope is help companies try to make better decisions about where they build, how much they build, and what source of energy that they're relying on.
Because these, again, it is not a one size fits all. We have to cater this to the specific grid, and the specific conditions that we're seeing at a local level.
[00:40:23] Bridget Scanlon: Right. And that's so important to get to the local level. And I was reading that just even recently, that Phoenix is putting up some regulations, the city, related to data centers. And I think when we were chatting the other day, you mentioned Dales in Oregon and reading up about that, and Amazon helping to improve the water infrastructure in the Dales region, by providing aquifer storage and recovery well and other things, paying the full price of the water. So it is really difficult for communities then. Do they have the infrastructure to support the electricity usage and the water supply? These are difficult questions. And so getting to the local level would be ultimately very important.
[00:41:06] Landon Marston: Absolutely. Yeah. We talked at the beginning of the conversation at a national scale, the total water consumption, water withdrawals of data center is relatively small, less than 1%, but less than 0.5%. At a local scale though, it can be quite significant because these are highly concentrated water users.
And so one thing, to put this in context, let me see if I can pull this statistic, I thought I had it handy here. Google, in the year 2023, reported their total water usage globally, and they let's see, they put that at 29 billion liters of water withdrawn, and 23 billion liters of freshwater for onsite cooling. And about 80% of which was potable water. So sounds like a lot of water and in fact it is. But this is one company, one of the largest players in data center operations. The water used spread across the entire globe. To put that in context, that's roughly the same amount to irrigate around 4,000 acres of alfalfa in Arizona.
Right. So, and then there's about, in that same year, there's about 369,000 acres of alfalfa was grown in Arizona. It's about 4,000 acres of Arizona alfalfa irrigate is by the equivalent to all Google's water consumption globally. And again, there are about 369,000 acres irrigated. So this is what I'm saying, as far as like, a drop in the bucket compared to these large water uses, which are going to be agriculture.
But the difference, the key difference here between agricultural water use and what we're seeing with data centers is again, highly concentrated. So you have just kind of one intake, and it's often relying on potable water.
So there's going to be some additional costs with that. There's also going to be local communities that might be impacted because their infrastructure, as I noted at the beginning, often these data centers are located outside of city, so usually smaller communities with smaller water systems that might not have the capacity to fully absorb all of the data centers. Especially if you have several data centers moving in, it might constitute 10, 20, 30% or more of their total water demand. And so that usually requires major investments in updating or expanding their water infrastructure to accommodate these data centers. And the question is, who pays for that?
Is it, as you noted earlier and maybe the case of Amazon where they fully paid for their expansion? But in other cases, that's not always the case where they might be subsidized in some way. And really the rate payers are kind of helping pay for some of these deliveries to these data centers.
And so that's again, really a local question that needs to be resolved, but something that water utilities and local communities should be thinking about is okay, who's footing the bill for this expansion? Because they are going to use a lot of water. And especially again, in small communities, it might be a substantial portion.
And what happens if we subsidize this, we give them tax credits and say, okay, we're going to allow you to tap into our water system. And we're going to, build a new water tower and it's going to cost several millions of dollars to help make, ensure that you have a stable water supply. And then, the bubble bust in the data center closes five or 10 years later, and they've made these decadal investments in the water infrastructure, and now 20 or 30% of their water demand is gone in an instant.
And so these are things that I think, again, while not a big deal at the national scale, the local scale need to be thought through and, sometimes these water utilities, especially these really small water utilities might not have the technical capacity to fully assess these trade-offs because data centers are now, it's not just a Northern Virginia thing anymore.
They're, as I noted earlier, spread across the country. You're starting to see 'em everywhere. And so you might have a lot of places that aren't accustomed to dealing with data centers, might not know the right questions to ask, and really have a full understanding of those water demands.
[00:44:54] Bridget Scanlon: Right. And I think that is so important. I mean, what is the data center going to bring to a community? Is it going to bring increased economic development? I mean, there are not that many jobs associated with data centers.
[00:45:04] Landon Marston: Not many jobs.
[00:45:05] Bridget Scanlon: And so, when you mention the comparison with agriculture, these agricultural communities, there's a whole industry related to that, which is economy and equipment and all sorts of things that support agriculture.
So, these are important decisions. And if the data centers start helping with the infrastructure and contributing to those costs, then I think that would be a big help to the whole thing. And then people are talking about reusing waste water, other things, so there many different things to consider.
[00:45:38] Landon Marston: Right. Certainly.
[00:45:39] Bridget Scanlon: Yeah. Well, I really appreciate your time, Landon, and I'm very grateful for the paper that you and ab put together. And they're very informative. And a lot of centers were putting out, sustainability reports and stuff like that, and most of the emphasis, of course, had been on carbon emissions, greenhouse gas emissions.
But then maybe after you guys started to emphasize the water aspects, maybe the sustainability reports then are reporting more on water. But as you say, transparency is complicated and it's nice that the Department of Energy is also developing tools and stuff for groups to evaluate many aspects related to data centers.
So thank you so much. I really appreciate it.
And, I'll also put together a website, with links and highlights and stuff of the discussion so people can get more information related to this.
[00:46:29] Landon Marston: Thanks for having me. I really enjoyed our conversation and hopefully the folks watching online are able to enjoy it as well.
