[00:00:00] Max Herzog: Welcome everyone to the Water Data Forum. I'm Max Herzog, Deputy Director of Programs and Partnerships with Cleveland Water Alliance, and I'll just be getting us kicked off here today. We are talking today about water management for data centers. But this is part of a broader series on water data questions presented by Cleveland Water Alliance CUAHSI, and the Water Environment Federation.
[00:00:37] So please do stay tuned for future sessions, and we'll mention that again at the end. We're very privileged today to co-present this session with the Water AI Nexus, which obviously has particular interest in this topic, and I'll just be doing a little bit of housekeeping. I did see a raised hand.
[00:00:55] Please, if you have comments drop them in the chat. We won't be enabling audio comments just to make sure that we keep the session rolling as we are on a tight schedule today. So just one piece of housekeeping before we dive into the panel discussion today. Want to encourage folks to share their questions.
[00:01:15] We should have about 10 or 15 minutes to tackle some of those at the end. We probably won't get to all of them, but we'll try and get to as many as we can. And as you do share those questions, please use the Q&A function in your kind of toolbar at the bottom. I know it's easy to confuse with the chat, but we will be going to the Q&A for questions first, as it's easier to navigate and then resolve.
[00:01:38] And then depending on how much space our panelists have as the conversation happens, we also do have some capability then to respond with text throughout, so we may be able to get to a couple additional questions that way. So again, please do submit your questions throughout using that Q&A function.
[00:01:56] And now it's my great pleasure to introduce our moderator for today. John Ikeda is the Chief Mission Officer with the Env- Water Environment Federation and has been working really closely on these issues, so we're very fortunate to have him here today to walk us through our discussion.
[00:02:14] John, I'll hand it over to you.
[00:02:15] John Ikeda: Great. Thank you, Max. No, and it's a pleasure to be here. So I'm John Ikeda. I'm Chief Mission Officer at Water Environment Federation. We are a leading voice for the water sector, representing over 30,000 water professionals around the world as well as one of the co-founders of the Water AI Nexus which is a groundbreaking collaboration between the water and AI communities to advance both what we're calling AI into water.
[00:02:40] How can the water sector take advantage of AI to solve big water challenges? But also what we're talking about today water into AI. What is the role of water in supporting AI economy, include data centers and semiconductor manufacturing, and how do we ensure that there's enough water to support the AI economy, but also everything else that we use water for in the country?
[00:03:05] So I'm very excited about the discussion today. We've got a fantastic panel including Chris Carreiro, who's the Chief Technology Officer of Park Place Technologies. We have Landon Marston, Associate Professor of Environmental and Water Resources Engineering with Virginia Tech. And finally, Kelsey Semrod, a Senior Water Resources Scientist at Pacific Northwest National Laboratory.
[00:03:31] A really, I think, important group today because there's such a complex narrative right now going on around data centers and water. And I think it's probably an understatement to say that AI is a hot topic of conversation in the country right now. And, within that, the role of data centers in supporting AI is a robust public discussion.
[00:03:57] And at the front of that- serious and real issue, and it is something that I think we in the water sector need to need to help shape and help ensure that we're being proactive in supporting that, but also that we're making sure that data centers are being designed in such a way that we're managing, they're managing water effectively, and and that it's not overtaxing America's strained water infrastructure and and our water re- water resources.
[00:04:30] So I want to turn this over to the panel, and before we get into more of the details, I think there's such a broad understanding of water in data centers right now. Can you give us a sort of high level overview? When we talk about how data centers use water, what does that look like?
[00:04:51] Maybe I'll, throw it out first to Chris from the perspective of the technologies that are used for cooling in data centers. What's the, 101?
[00:05:02] Chris Carreiro: Yeah, 101 is ultimately everything is getting hotter. These chips are A few years ago we were talking about maybe three KW cabinets.
[00:05:13] A lot of the kind of references are more so power-hungry systems, hotter chips, all of these things are related to basically the innovation of racks going from three KW now up to 50 KW and some even higher. Some of the Nvidia announcements are talking about 500 ca- KW cabinets. All these things, the more power it uses, the more water it's gonna waste.
[00:05:40] The systems are basically indirectly cooled by either an air conditioning or some kind of heat exchange in the building. And some of those systems aren't really efficient for for water water waste or just kinda just water usage in general.
[00:06:00] John Ikeda: Good. Kelsey, can you talk a bit more about, about that water usage?
[00:06:04] We've-- It's a lot of times when people talk about how data centers use water, they're looking just at that direct level of consumption. But bigger picture, how do data centers use water?
[00:06:15] Kelsey Semrod: Sure. Yeah as Chris was mentioning water is crucial, and water management's crucial for the cooling of these systems, and there are certainly trade-offs, water and energy trade-offs with the different cooling types.
[00:06:30] Dry cooling is possible. It's not like... Air, cooling is possible. It's not as commonly used necessarily, and it requires more energy. And as you mentioned, there are different types of water use. We have the, direct cooling, where the water is directly supplying the systems, and we refer to those as scope one water use consumption.
[00:06:54] And and then there's indirect cooling. So that's what we refer to as scope two, and that's more associated... That's that indirect water consumption that's associated with power generation or transmission and treatment of the portable water or wastewater. And as Chris mentioned, just the landscape is, it's constantly changing with technology improvements and technology development, like the NVIDIA chip, for example.
[00:07:21] And just requires us to be nimble in terms of from a science perspective and a planning perspective to just alter our models and planning assumptions accordingly. So again, trade-offs. But a, lot of that water use is not actually necessarily directly for the cooling, but it's also that indirect requirement as well.
[00:07:45] John Ikeda: Lane, can you talk a bit more about that?
[00:07:46] Landon Marston: Yeah, sure. And I'll build first off what Chris was mentioning with respect to these, racks moving from three kilowatt hour or three, three kilowatts to 50 or more now. A- as he was noting, when energy comes in, some of that energy is used for compute, but some of that energy is actually manifests itself as heat, and that heat has to be removed or else the servers don't operate efficiently.
[00:08:09] And so as he was noting, there has to be some cooling system to move that heat out and dissipate it in some way. And so historically, you go back a couple decades ago, you would have these closet or enterprise data centers. And so these would be things that would maybe be in an office building, maybe take up one or two rooms within a traditional office building.
[00:08:26] These have been around for decades. And these have traditionally used the same type of cooling system that, that your office building would have used. Mostly air side cooling, very little water associated with that, at least directly. Back to Kelsey's point, indirectly, it's using a lot of energy, and that energy has water use associated with that.
[00:08:44] And so when we, A- as we've seen this kind of shift within the sector, not only from these server racks that are low-density energy to now more high-density energy, you're also seeing another shift, which is moving from these kind of enterprise data centers to, in the last decade or 15, 20 years, these co-location data centers, which you can think of maybe the best analogy of a rental unit, right?
[00:09:07] You would go, and you would rent... One company owns the rental facility, but you would go and rent a space in, that facility and store your things there. S- likewise, people would store their data on the servers and... but they wouldn't own the entire facility. Now, increasingly, you're seeing these hyperscale data centers, and so these are gonna be these really massive buildings, sometimes can be m- multiple buildings that might be the size of a small college campus.
[00:09:32] And I think it's important to make this distinction between different data centers. Data centers are not a monolith. It's not every data center is exactly the same. Some data centers can be, again, encompassed within a traditional office building. Some are gonna be the size of hundreds of acres large.
[00:09:49] And so I think as we move through this conversation, it's important that we make that distinction because these have very different water use profiles associated with them. So that's one of the, key points that I want to bring up there. And then again, back to the, cooling technologies. That heat has to be dissipated.
[00:10:05] As we move to these higher density racks more heat's generated, and it... water is much, much more efficient at dissipating that excess heat than air side cooling. So that's why you're seeing this increase in, water-based cooling in many cases. But as Kelsey had mentioned, there's some advances that have been happening over the last several years, including just last week, that, that might shake up this this kind of way of, looking at this relationship between water and energy and, also the heat that comes from these servers
[00:10:37] John Ikeda: Can you talk a bit more about that?
[00:10:39] Landon Marston: Yeah. About the last piece of the- Yeah ... what came out last week? Yeah. So NVIDIA just, and I haven't gotten into the details of this, Kelsey might know more than me NVIDIA just released an announcement about 10 days ago, a week and a half ago, something like that, that said they have now developed a chip and kind of this AI framework where they can raise the, temperature, the threshold of which the cooling water can come in- into the system and dissipate that heat.
[00:11:05] So now it's somewhere like 45 degrees Celsius, about 113 degrees Fahrenheit. So this is a, pretty notable jump in the heat tolerance of that cooling system. So what that practically means, if you're using like a closed loop system even though the people often think this closed loop means zero water, that heat in that loop still has to be dissipated in some way.
[00:11:26] And so now you can use these air side cooling perhaps as opposed to water-based cooling, effectively lowering the evaporation associated with dissipating that heat, which effectively from a water management perspective means less water over the course of the year. But you still have this significant variability throughout the year in the water usage, especially those hot summer days, you're gonna see larger amounts of water.
[00:11:50] And so this is something we can maybe unpack a little bit throughout the course of the conversation, is this, operational variability, which can present challenges.
[00:11:59] John Ikeda: And, maybe to Chris or Kelsey as we lay out the broad brushstrokes of what we mean with water and data centers closed loop cooling is a term that, that people are using a lot now, and I think- It feels like people are talking about it in different ways.
[00:12:18] When you, define closed-loop cooling, what does that look like for you?
[00:12:23] Chris Carreiro: Yes. So I'll take this or at least the first half of this. So you have the heat dissipation in the building. We call that a facility water loop. That's basically the large piping, that's the cooling towers that's basically anything outside of the building.
[00:12:42] And it could be a closed loop or it could be an open loop generally. Closed loop would be glycol and water just floating through the system, removing heat from within, in this case the data center from every single rack that's insid- inside of that data center. It would somehow touch that either through a secondary loop or something.
[00:13:04] But it would be a loop that passes heat from the device and then all the way ultimately out of the building. If it was an open loop that would be something like an evaporative cooling tower where you have that water still circulating through all the I- all the IT cabinets and then taking heat and then it leverages evaporation to remove some of that heat.
[00:13:32] On a, cool day on a regular day, let's say it's below say it's below 80, 90 degrees outside and not very humid, you'd get a lot of evaporation from from that which is highly efficient for for cooling a data center. So it's a really good way of cooling that data center, but you waste a lot of water.
[00:13:54] So you have to replenish that loop with additional water to continue keep all the air out of that system and then continue to get the heat removal from the building. So those are the two I guess kind of simple ways of, looking at it, and that's just on the compute side. There's also that for the power equipment in other parts of the building that usually IT doesn't get involved with.
[00:14:17] But specifically why the data center's front and center is because we're putting more and more systems, more dense equipment that basically is evaporating more and more water. I think there was an example once of every every Google search is something like a tablespoon of water that would be evaporated.
[00:14:37] And your ChatGPT search is like a 500-milliliter bottle that's being wasted every time you do a, query on ChatGPT. So those are kinds of the examples, the differences between traditional searches and, something more with GPU-focused AI searching.
[00:14:56] John Ikeda: Yeah. I will say that I've, seen any range of different numbers on how much water is used depending on the query.
[00:15:04] And I think sometimes it, it raises more questions than it answers. And this is where maybe moving beyond the numbers is important. And- and Kelsey, maybe a question for You Landon talked about this isn't a monolith, right? It, depends on the data center what kind of data center it is what kind of cooling technology, the time of year.
[00:15:28] To what extent does geography play a part in this as well? Does it matter if a data center is in here in Virginia, where we have a huge cluster of data centers versus in the arid American Southwest? Is that more or less the same thing, or does the water availability, geography, climate impact that?
[00:15:48] Kelsey Semrod: Yeah, certainly. And as Landon was speaking, I also wanted to foot stomp that importance of considering regionality or those state specific requirements, region specific requirements because every data center developer region is unique. And we can't necessarily treat water management the same in Virginia versus the, desert Southwest.
[00:16:16] With that there's, there, there are certain statistics that and modeling that PNNL has been involved in through our Department of Energy funded open source modeling project. It's called IM3 Integrated Multi-Sector Multi-Scale Modeling Project. This is a human earth systems model, and it looks at impacts across water, energy, and land systems at more of a a regional scale and more of that considering those, factors that you talked about keep humidity viability of different cooling technologies siting implications for different data centers.
[00:16:58] And I mentioned scope one and scope two water use earlier. And Virginia is relevant. I'm based right outside of DC, and Virginia is considered is, what we call a, primary market in the data center field, which means that it has an existing demand of over eight hundred megawatts. There are a few regions in the country that are considered primary markets, and Virginia is one of them.
[00:17:23] And some of the initial modeling results out of PNNL w- in this IM3 program indicate that direct water consumption that's going directly to, cooling, those scope one cooling those scope one water uses are less than one percent in all states, except for Virginia, where it's about six percent.
[00:17:43] But those local uses are, much higher. But really it's all about, like from a large scale, it's really all about the indirect requirements, those scope two requirements. And in Virginia specifically, again, that's my lens and where a large cluster of these data centers are we have about sixty percent of the the electricity in Virginia comes from other, states in the region.
[00:18:08] And for direct water use the models are estimating by twenty sixty we'll see about ten percent of total water use just for data centers in, to cool data centers directly, up from six percent currently by twenty sixty. But then for indirect use in twenty twenty-five, we had about twenty-six percent of the water use in the state was associated with indirect water use for data centers.
[00:18:35] And this is projected, again, based on current trends, current technologies this is projected to rise to eighty-seven percent by twenty sixty So this just points to the importance of- Yeah ... That long-term capacity planning and those local considerations about regional water requirements now and into the future.
[00:18:57] John Ikeda: Which is so Virginia I don't generally think of us as particularly water stressed, but with that much of a, shift assuming there's not a huge step change in water use and energy use efficiency with data centers can our watersheds sta- manage something like that?
[00:19:22] Kelsey Semrod: It, I think the answer is, and this gets maybe into an opportunity area, the answer is it probably depends, and it, requires us to be diversified in the types of water we use in terms of alternative water use. Loudoun County we, hosted them at a resilience workshop we had last last spring, PNNL and Department of Energy.
[00:19:48] And they mentioned that in 2025 they had 117 additional data centers approved at the start of the year. Again, that's just approved, not necessarily going through to full development. But that would require a lot more capacity for them to handle they said four- 40 million gallons per day.
[00:20:08] Whether that comes from from, directly from the Potomac River or from wastewater or from o- other types of sources, I think just requires us to be more, more diversified. And to your point, yes, a lot of us think of us being very very water not water stressed in this region.
[00:20:32] We have plenty of water from the Potomac and, other sources, but but with this increased capacity I think we just have to be mindful of the- Yeah ... Of using a diverse set of sources.
[00:20:45] John Ikeda: And to your point about diverse sources, I think typically data centers are using municipal water supply, right?
[00:20:51] They're hooking up to utility using the same potable water that that any customer would use. More and more data centers and utilities are looking at reuse, like you mentioned in Loudoun County. Are there, in addition to reuse, are there other sort of novel water sources that you think should be explored?
[00:21:11] I saw somebody in the chat mentioned mine water. People talk about putting data centers at the bottom of the ocean for cooling. What are some of the opportunities there that that you would see?
[00:21:27] Chris Carreiro: Yes I'll take this one. So we have actually one thing Park Place works on a lot is immersion cooling, where you're taking a server and putting it inside of a tank of oil.
[00:21:41] And very similar to that NVIDIA chip that can operate at over 100 degrees Fahrenheit, an immersion tank, you can actually run that around 90 to 100 degrees and cool it with that 90 or 100 degrees. So it's it's better at heat removal. The servers actually last longer. If the facility has a closed loop system, it's properly positioned, you could have almost zero water waste with that type of deployment.
[00:22:08] So things, like that, innovations like that are we're starting to see a lot more of it. It's a this is a big opportunity area because it's creating problems all over between water, power usage even noise ordinances with the amount of fans that are running on the top of a building next to the data center.
[00:22:29] But those types of innovations are, definitely helpful. Other r- refrigerant-based cooling is another sealed system that, that is deploying. Companies like Celsius, Xudicor are making these really high efficient solutions that can cool all the way up to 2,000 watts per chip.
[00:22:50] That's way well into the future of where we're going, probably past 2030. It would be able to cool those types of equipment without any water waste at all.
[00:23:03] John Ikeda: So what's the trade-off then? If these are much more efficient technologies, why aren't they being deployed at scale
[00:23:10] Chris Carreiro: right now?
[00:23:11] I would say from my perspective hydrophobia is a big thing. Water is never, has never been a s- something that people wanted inside of the data center for the either an IT admin or a data center operator. Generally, water's been outside of the data center, and just strictly air cooling has been the last 30 years in the data center.
[00:23:35] That's been the priority. Now that's ... It is starting to change, so some of the hyperscalers obviously have, had no choice but to cool with water to the chip, so usually d- that, those direct-to-chip solutions. Some of these that, that I spoke of the two-phase DTC with a S- a Celsius Zudacore in the immersion solutions, those are a little complicated to, to actually get implemented, so it's not a an easy lift for for data centers.
[00:24:06] And it's an emerging technology, so there's still some, people that are a little nervous to get those kinds of implementations. We're seeing a big shift in the past couple of years, but it's, it'll take a little while before some of this adoption starts making its way into the larger scale data centers than just some of the edge cases that we're seeing now.
[00:24:29] John Ikeda: Great. Landon, I wanna loop back a bit. We're talking a lot about the- this sort of water use around data centers. On the wastewater side what does that look like? How, are data centers and and wastewater treatment plants inter- interrelated?
[00:24:48] Landon Marston: Yeah. And maybe as I answer that I can touch back on a couple things that, that Kelsey and Chris had noted.
[00:24:55] A- and I'll start with that, then circle back at the end on the wastewater piece. So I think Chris had noted the rapid advancements in technology and what that means for water use, and Kelsey had noted if we assume business as usual, what future water demands will be.
[00:25:11] Kind of looking back at history, i- within the data center industry, and this is what got me interested in this topic eight years ago, nine years ago was this idea that these trends will continue on indefinitely. And what we've seen through the industry is adoption of new technologies that become more efficient.
[00:25:30] So going from 2010, if you were pr- to project out the energy demand, this is what the major concern was at the time, and even still today you project out the energy demand it was gonna be, like, 550% increase is what they were projecting. Maybe even more than that. They were... I, remember seeing some media articles, this was maybe a decade ago, saying that by 2025 that a third of the global energy demand was gonna be dedicated to data centers.
[00:25:58] And that seemed a little bit off to me, but I started digging a little bit more and I think what was happening is these projections were assuming, again, this kind of consistent efficiency moving forward. But what happened instead between 2010 and 2018, we saw near a, nearly a sixfold increase in compute- But only a 6% increase in energy, right?
[00:26:18] And so I think I'm optimistic that, the sector will in the same way they've done for energy, will likewise with water, become more efficient. Especially with, as you noted earlier, Ma- or John we're seeing this increased public pressure to do and I think that has implications not only on water availability and water use, which is a good thing, but from a utility perspective, that brings a lot of uncertainty, right?
[00:26:42] You could have these dramatic shifts in the amount of in the infrastructure capacity that's gonna be required to serve these data centers. Utilities are typically more accustomed to serving more consistent demands with clear assumptions of how that demand, those demands will growl- grow. And and now data centers are breaking that because, one, we don't know to the degree that these data centers will expand, and then we also have a very unclear understanding of how their efficiencies will change over time and how their water usage will change over time, which makes it very difficult to predict this in the future.
[00:27:14] Which, as most utilities know, you have to often make investment decisions over multiple decades, and you're asked to... And this could change within a couple years, right? The, there's a lot of speculation as well within data center industry where they're building these shells, but they don't... Whether they'll actually be filled or not, still a little bit uncertain.
[00:27:30] And so that really makes it challenging for water management more broadly. And so I just wanted to highlight that point. Now, to your question about- Wastewater, I think some of the same concerns happen on the wastewater side as well. What's that profile gonna look like? Not only the total volume of discharge, but also its composition.
[00:27:46] So when you have these type of evaporative systems that are that are not only evaporating water, but there's this blow down wastewater. And so this can typically be high salinity but it's... I want to highlight here these cooling systems aren't fundamentally different from cooling systems you might see in other industries.
[00:28:06] They might be much larger in scale, and there might be more water passing through them just because of the, drastic cooling demands. But evaporative towers have existed for decades, right? These things have been around for a long time. And so some of the same things that are associated with these these type of cooling systems, these traditional cooling systems that have been used in, other ind- industrial contexts are going to be applicable within the data center realm as well.
[00:28:31] And so I think one thing to note in ma- many cases with these evaporative cooling, this is really a concentrator. So whatever's in the water that's being delivered, that will be concentrated as some of that water evaporates, and then eventually the, whatever's left will eventually again, con- condense and make its way to the wastewater treatment plant.
[00:28:50] There's also other types of things like chlorine and bromine and things that get added to the water, biofilm growth in the warm water, corrosion and scale inhibitors, mi- phosphates and things like that might be added to, to help protect the piping. So these are all things that wastewater utilities will have to be cognizant of, but again, not fundamentally different from from other types of industries that are using these types of cooling system.
[00:29:18] I can elaborate a little bit, a bit more John, if you'd like, but I'll leave it there. Maybe Chris has more to say or Kelsey has more to say on that
[00:29:27] Chris Carreiro: Yeah. So on the blowdown side, we've actually we're actually partnered with a company that's helping reduce that, and it's actually a filtration system. So instead of... So it reduces the amount of blowdown water. To Landon's point about all the minerals and content in the water, once you have evaporation, you have a high concentration and that becomes caustic inside of the piping that actually finds its way back into the data center.
[00:29:54] There's a couple of innovative companies that are out there working on ways of filtering out that so you have much less blowdown, so a lot less exchanges of water. So that, that's a really big advancement. Another is it's called a, an adiabatic system. So that actually uses a closed loop, and it only uses a sprayer for evaporation on the coils outside when needed.
[00:30:21] So that's a much less that's, that wastes a lot less water as well, and it takes more advantage of kind of the region. Kelsey kind of mentioned that before like strategically placing data centers in climates that can sustain or, let's say, operate free air cooling for 90% of the year would be much more efficient and then maybe using chillers for the, that off time.
[00:30:46] You could actually completely reduce the any water waste at all. And you're right. These these systems in, in, from a data center's perspective, these systems, as Landon said, the, they've been around for 30, 40 years, probably even longer than that, these evaporative cooling towers.
[00:31:04] And there, there is some innovation in it. The challenge now is we're just seeing so much demand for all this compute that it's just, it's increasing the amount of of those systems. So it's just a lot more volume of evaporative towers or these core, even more cooling towers. So I think we, we need to take a, step back, look at exactly the placement for some of these data center builders and get these put in the right place where you can take advantage of of the physical or geographical location of it.
[00:31:41] John Ikeda: To that point then as you think about how much water data centers are using relative to other industrial users as you think about the geography, is this ultimately, is this a policy problem, or is this a public perception problem?
[00:32:04] Landon Marston: I can s-speak to that at least initially.
[00:32:06] So just to put this in context Kelsey kind of got, touched on this a little earlier, but to put some numbers behind it if you look at total water demand across the, US economy, data centers, and this is a high-end estimate, their direct water usage is around .04% of total water demand. So on on a national scale, energy, agriculture, so basically the ener- the what...
[00:32:31] Turn on our lights, Zoom and then the food that we eat, that's what really dictates how much water is used across the US economy. Most data centers, as you noted earlier, John, get their, water directly from a water utility, same utility that delivers water to our homes and our businesses. And one thing I want to note with respect to k-kind of managing water and policy, most, in many states across the United States, they require...
[00:32:58] Water's managed and kind of water use is managed at the state level. And so they typically, the state will require large water users, and what's defined as large varies from state to state, usually between 10,000 to 100,000 gallons per day or something like that, require large water users to report to the state how much water they're using.
[00:33:13] But because data centers are reliant upon utilities primarily, they don't have to report to the state, and this isn't necessarily a loophole they're exploiting. This is just the nature of, the policies that are that are in most states. Just that individual water usage that's, that I'm using at my house or, John, that you're using at your house isn't reported to the state.
[00:33:33] Likewise, they're connected to utilities, and they don't have to report that to the state either. And so there's a little bit of a it makes it a little bit challenging to get really firm estimates on many data centers' ultimate water usage, and that, this makes it really challenging to have that benchmark for utilities to use to try to gauge what is a reasonable amount of water.
[00:33:52] Oftentimes, we have to rely on models and coefficients and things like that, and that's a, useful proxy. But obviously, we would, to be able to better manage the system, we need better data, and that, that is coming down to a little bit of a policy. And there's s-several states and, different groups are pushing for different frameworks as far as transparency when it comes to water usage, as well as energy usage and other kind of r-resources used by data centers.
[00:34:19] John Ikeda: Okay. Kelsey, what's your thoughts on that?
[00:34:21] Kelsey Semrod: Sh- sure. I was also going to mention one of the, major again, from this r- resilience workshop we hosted last, spring in the energy water resilience in the context of data centers and demand growth just was the need for data transparency.
[00:34:38] And as Landon mentioned, certainly between the data center devel- developers and the utility providers, just as as that discrepancy maybe exists between the folks who are making the decisions in terms of approving, approving development versus the folks who, are required to provide that service to its constituents.
[00:35:01] A data center is like any constituent any any user. But also aside from the transparency between data centers and utilities, also between water utilities and, power utilities that came up frequently in some of our conversations. Of course, we want utility providers to be included early and often in the process of data centers and demand growth to understand if if we can provide that capacity.
[00:35:33] But also there are maybe power companies that are looking at they're tracking their electric load growth projections, and maybe they're not involving the water utility in their district necessarily as, as intimately as they they, should be necessarily. And just so ensuring that regional coordination in general is just a really powerful thing that isn't necessarily always occurring.
[00:36:00] As you would imagine a watershed is a different area than maybe a the area that a power utility provides in terms of its service area. But just that co-planning between a water and power utility is really powerful because those data center growth indicators are going to be the same in those areas, looking at population employment data, county site plans and those utilities have different industry relationships and maybe regional partnerships and different models that can be leveraged accordingly.
[00:36:34] So I would say that's, a challenge as well from maybe a policy and planning perspective that-
[00:36:40] That could be addressed and better managed.
[00:36:43] John Ikeda: Okay.
[00:36:44] Landon Marston: Yeah. Can I build off that? Please, I just think talking to different utilities it's a mixed bag, right? Kelsey's saying.
[00:36:51] Sometimes utilities are involved with the, regional or city planners, community planners from the get-go, and they're having discussions with these data center developers right from the very beginning. But in als- in many cases, as she noted, sometimes they're just told after the fact that the agreement's already been made "You need to service this much water."
[00:37:11] And they're like we don't know if we have the capacity. It'd been helpful if we were involved in this from the beginning." And so I think having those, coordinations from the be- the onset would be really helpful for utility and water management.
[00:37:23] John Ikeda: Definitely. I we've gotten a lot of great questions in the chat here.
[00:37:26] I want to make sure we're leaving enough time to answer these. So maybe just in the last few minutes of the discussion right now, looking ahead look at 5, 10 years down the road as we think about water and data centers and AI. What does it look like if it's done well?
[00:37:48] We've heard Chris in particular, some of these really interesting emerging technologies. I think Kelsey, we're looking at how do we define scope one and scope two more effectively? What is a successful decade of water use in data centers look like?
[00:38:13] Chris Carreiro: I'll, take this. My so my opinion first would be a data center that doesn't use any water at all with the technologies available. Even on the... It's a little more difficult on the indirect side, but at least from the direct data center usage, maybe reduce that entirely. I, I know TPU is another big thing that we're gonna be seeing a lot more of at least in the next couple of years.
[00:38:42] But basically these these models are gonna start leveraging TPU, ASICs but now have chips that are able to be cooled at a much higher temperature would be, one. And then I think ultimately shrinking data centers down to have more of a decentralized data center network 'cause that also solves the latency problem within data centers anyway.
[00:39:09] But basically smaller data centers spread out much less water waste, and then much higher temperature resiliency for that chip itself.
[00:39:20] John Ikeda: I-- And, maybe Kelsey, this is a question for you on the... If we were talking about zero water use at the scope two level or indirect level, that basically implies fully renewable power data centers, right?
[00:39:35] Photoelectrics or wind. Any other conventional source of electricity implies at least a fair bit of indirect water consumption, right?
[00:39:46] Kelsey Semrod: Yes, and I think a lot of the time we're, focused on those direct, that direct consumption and addressing net zero water use at the at the usage point itself, at the data center itself perhaps, and not necessarily in the indirect requirements associated with the power plants that are feeding that system.
[00:40:08] So yes, I think ad- addressing more of the source of where that power is coming from and and the the water requirements associated with those indirect uses would be really promising. And I- we mentioned Loudoun Water before. That's, a utility that's invested heavily in a reclaimed water system.
[00:40:33] 26% of their data centers in that county are are leveraging a reclaimed system directly though, right? So making sure that those indirect requirements are being addressed too. I think also just from a, We haven't talked about this, but investing in the water infrastructure and conveyance itself so much of our focus is on the usage and the end use, but our national water system is aging, right?
[00:41:01] And so if developers can invest in that infrastructure that's supporting its operations, then we can reduce leakage closer to the source. And then my final point would just, which we also haven't touched on too much, is just those heat recovery technologies. They're really promising. We talked about wastewater earlier, but leveraging the waste stream in the form of wastewater to power the data center itself, to power the indirect requirements.
[00:41:27] DC Water is doing this in, in my area. They're considering expanding it to supply thermal energy from the wastewater stream to others in the region, and they've said that applying that 10-degree Fahrenheit temperature change to a million gallons per day of wastewater would deliver one megawatt of thermal energy.
[00:41:48] So if we're looking at 200 mill- million gallons a day, that could supply 25 million square feet of cooling. So there are other cities and, places around the country and, the continent that are doing this, in Denver and Vancouver as well, using wastewater to to heat and to generate power.
[00:42:07] So I think that's a real opportunity as well to limit the, those energy requirements that are more water intensive
[00:42:16] John Ikeda: That's excellent. And that's for Water Environment Federation one of our big strategies is to advance the circular water economy. So any of those sorts of examples of where things that are traditionally considered part of the waste stream whether it's heat or wastewater itself, finding ways to use that and, recover it, I think is very important.
[00:42:34] I Landon as, you look forward next five, 10 years what are we doing to, effectively solve the water management question with data centers?
[00:42:47] Landon Marston: Yeah. Yeah, good question. I wanna kind of piggyback off what Kelsey was saying with respect to, I think, th- let me take a step back.
[00:42:55] I think there's been a lot of discussion in the media and elsewhere about all the challenges data centers bring, and I, don't-- I think some of those are legitimate. However I, think maybe me just being a little more optimistic, I think there's a lot of opportunities here as well. And so Kelsey had already mentioned we have aging infrastructure.
[00:43:14] We have this potential influx of investments that could come from these data center operators that can, might help with leaks and things like that. I would also note- You know, these data centers are increasingly moving from major urban hubs to more rural areas, and this can present challenges, but it also can present opportunities because you see in some of these places depopulation or a decrease in water demand over the last several decades, and that means excess capacity within their system.
[00:43:40] And so if you have a large user come in that's able to absorb some of that excess capacity, that could potentially or has the potential to lower the rates for, other rate payers. So I think there's some, opportunities there that haven't been fully investigated looking at, areas across the country that have the excess capacity that would be welcoming to data centers.
[00:43:59] And there's a lot of discussion about raising rates of electricity or water. I think there's also potential to lower rates i- if the right stru- financial structures are set in place and contracts are put in order. So I think that's one opportunity. I think another thing we touched on earlier that I would like to see in the next few years, and I made this point directly to the tech companies, is greater data transparency.
[00:44:17] This is something we discussed earlier. I think a lot of the, issues that we're seeing with kind of misinformation in the public is, mainly because a lot of this information wasn't shared, right? And, a lot of times these tech companies have non-disclosure agreements with utilities that let limit what they can share with the public.
[00:44:36] This kind of ties their hands, and I've talked to a lot of utilities that they said this is a constant struggle. Just any type of proposition that comes up with related to data centers, there's a lot of pushback, and they feel constrained on what they can say because of these NDAs. Now, some of it you're seeing.
[00:44:50] Microsoft, if you, a couple months ago, said we're doing away with non-disclosure agreements. I think that's one step forward. I don't know if other tech companies will follow suit. I think they're recognizing, and I've seen this kind of a public d- discourse in social media you're starting to see increasing levels of transparency within that industry, and they're starting to release information on their water usages and put that information into context, and I think information like that is helpful.
[00:45:13] I'm hoping it's not too late, but hopefully that will lead to more informed decisions such that we're not doing things that maybe to Kelsey's point okay, we're gonna use this system that's going to dramatically reduce on-site water usage, but it's gonna dramatically increase the overall water footprint- Yeah
[00:45:33] because it's moving everything off-site because electric- Yeah ... significantly higher in electricity demand. So I think just making providing these communities, providing these utilities with the best information available and then letting them make the decisions for their local context because in tying this back to the beginning of my, of the conversation each data center's unique, each community's unique.
[00:45:52] It very climi- climatically but also socially, economically but we need to provide sound data for them to make decisions.
[00:46:00] John Ikeda: Yeah. No, very well put. And I think our own experience with Water AI Nexus, I think a lot of the hyperscalers are realizing now that the more transparency is the right path forward.
[00:46:13] And it's one of the things we're trying to do, is to create a movement around a commonly agreed standard for measuring scope one and scope two so that we can compare this apples to apples and, promote the right sorts of policies that get the right kind of data centers in the right places.
[00:46:33] I wanna bring in some of these great questions we've got in the Q&A here. And to your point about utilities in some places maybe have excess capacity and data centers potentially being a, something that can help lower rates. If you look at the Great Lakes region with Cleveland Water Alliance obviously a massive freshwater resource.
[00:46:57] We have a lot of utilities throughout the Rust Belt that have probably excess capacity. And as a question was he- raised in the chat here is there also an opportunity to use deep water cooling technology? There, there's regardless of, the water resources that are available there for consumption.
[00:47:17] Just the thermal ability of the Great Lakes to cool, is that an opportunity that the region should be, taking advantage of?
[00:47:29] Chris Carreiro: Yeah I think they, should. It's Microsoft did have, Microsoft and Google both had oceanic data centers. There is a little bit of a challenge putting a data center at the, either at the bottom of the ocean or the
[00:47:43] bottom of the lake. It's the serviceability. It, takes a lot to get that data center back up to the surface when you have failures of that data center, which happen very frequently, obviously the larger the data center is. But there are challenges, and I do think that that should be e-explored a little more, k-kind of that thermal difference between the surface or even maybe some kind of heat exchange with either deep water or even geothermal is another example of, something like that, where you're taking something that's readily available in the earth for cooling instead of leveraging water.
[00:48:24] John Ikeda: Yeah. And that's, that was... So the question itself, there was the Enwave deep lake water cooling system mentioned. I'm personally not familiar with it, but I think it's certainly technology like that I think is worth exploring- and it's something where I would see there's a real regional competitive advantage there for cities like Cleveland.
[00:48:45] I've worked also with the the Water Hub in Mil- in Milwaukee with other Current in Chicago. These are different sort of water organizations around the Great Lakes. I think there's a real comparative advantage for, cities that are in water-rich areas to take advantage of that.
[00:49:08] Another question that came out of the chat here, and I think this is a good one what is the advice that you would give if you were meeting with the mayor of a small town, mayor of a s- or the manager of a small utility what should they be asking data centers when they come in?
[00:49:31] Landon Marston: So I can add a couple things, and I'm gonna post something in the, chat here in a bit. We've got a survey, and we're... That's one of the questions that we're asking utilities directly. So if you represent a utility and you would like to give your input we will not only collect that, but we will send it back to you.
[00:49:46] And so we're gonna anonymize and aggregate all the r- all the, responses that we get. And that's gonna be a resource that will be useful to you to understand what other utilities are doing because this is very much ad hoc, like each utility's operating independently, so trying to get some bearing on what others in your sector are doing or not doing.
[00:50:05] We're hoping that will be a useful resource for you guys. So if you wouldn't mind, I'll post the survey. I think John's gonna post it and Max is gonna post it again later. If you complete that survey, that would be really helpful, not only for us, but hopefully for you all as well. Back, but back to your question, John, with respect to what they should ask, right?
[00:50:21] So I think there's, a long litany of questions that one of which would be not only the standard questions, which is like what is your average daily demand, but what's that peak? That's gonna be really important, this was talked about earlier, because that can be very dif- it differs a lot from other sectors that they're accustomed to serving.
[00:50:38] And they could have really sharp peaks compared to other sectors. I think trying to understand not only that maximum peak, but also the build-out because you have many data centers that are coming out in phases. And so if you plan capacity and for a fully built-out phase, and it's they only end up building half of that, you need to have contract clauses and things in place to be able to recoup those potential costs.
[00:51:01] So things like, Things like the availability fees, right? Making sure that, they have enough w- water available when they need it, and they're paying for that kind of take or pay kind of contract clauses. So if, they say they need a million gallons a day you- they're gonna have to pay for 80% of that whether they use it or not.
[00:51:22] So this could help protect, again utilities from this potential build-out that may not, might not be used. Maybe because the data center doesn't develop in the way that they're expecting. Sometimes you have these kind of speculative builds. And then we also have this possibility that we're like we saw 20 some years ago with a, with the dot-com bubble.
[00:51:40] Maybe we're in another bubble as well, who knows? And so if that, progression slows down tremendously, what's that mean for the water utility and their operations? They maybe have already invested heavily in new infrastructure with a 20 or 30-year payback period, and now the people that they thought were gonna be paying it back are no longer in business, right?
[00:51:57] So I think things like that, just making sure all those financial and contractual pieces are taken care of are really important. This goes back to Kelsey's earlier thing about bringing in all the parties early on so there could be coordination among them and sharing of information. So this data transparency idea again.
[00:52:15] John Ikeda: Excellent.
[00:52:16] Kelsey Semrod: And a quick- Please. A quick add, John, if you don't mind. I'll, post something in the chat as well. Just I- a few days ago I was, sent a, data center primer for water service providers. And it I've only seen the, PowerPoint, a brief version of it, but folks can also request the full PDF.
[00:52:37] And my understanding is that it, it does have some, really fantastic information on a- as Lyndon was mentioning, identifying the representatives from the data center side to the infrastructure regulatory side and to external folks, what kind of questions to ask, what relevant information you should be communicating to the developer as well, and having them be aware of as they're as they're engaging in these conversations that have to do with capacity, but also project phases and design assumptions.
[00:53:10] I just posted that in the chat.
[00:53:12] John Ikeda: Excellent. No, thank you for sharing that. And I also put a link to our Water AI Nexus which is also working to compile different resources for people who are interested in this topic as well as some various insight reports that we're putting out on water use in data centers.
[00:53:28] And also on the other side of it how utilities are using AI, for example. I know we're almost at time. I do want a couple more interesting questions to raise very quickly if anybody has any last thoughts on this. One is the flip side of what we're talking about geography and if the Great Lakes has a sort of comparative advantage in putting in data centers.
[00:53:50] For more water-stressed regions, should they be going in at all, right? Somebody asked about building a data center in a, desert, in, in an agricultural valley sort of north of the Great Salt Lake. Should there be data centers there at all?
[00:54:06] Chris Carreiro: Yeah. That's actually a, massive region with data centers currently, and I answered a little bit in the question there.
[00:54:14] But a lot of data centers chase where power availability is, and there was a lot of f- very cheap power in the desert for a while. The challenger trade-off is is basically the heat. Is, that the right region for a data center where you can leverage free air cooling, where these newer style data centers need a lot of a lot of cooling?
[00:54:38] So that's basically so that's the trade-off. And then ultimately the, other reason why there's not more data centers going up in the desert is is power availability. There, there isn't any more power available there so that's why data centers aren't expanding. So they're looking for other regions as well.
[00:54:56] But yeah, definitely a good question.
[00:54:59] Landon Marston: Yeah. I would say this idea of should it's an academic sitting on a college campus, maybe not my position to say that. I think what I would say is that these communities need information to make informed decisions themselves, and they can balance these trade-offs that Chris was mentioning and try and decide what is best for their little community.
[00:55:17] In many cases, we're talking the A- the American Southwest, very water-scarce region. Any new water user, whether it's a data center, a subdivision development, an expansion of irrigated agriculture, that's going to put additional stress on an already stressed system. And so these communities, these states need to decide how do they want to allocate this scarce resource.
[00:55:36] Is it for data centers? Is it for expansion of agriculture? Is it building new neighborhoods? This is a decision that they'll need to make collectively and and, decide that themselves. Yeah.
[00:55:46] John Ikeda: No, ultimately my sense is our role in the water sector is to provide clear, transparent information so that communities can make those decisions and understand those trade-offs.
[00:55:58] But also to, to help advance technologies that are gonna really solve this. 'Cause at the end of the day I think it's important to remember we have to make sure there's water resources available for all uses. But at the same time, it is an incredible resource, and we have so many complex water challenges from managing aging infrastructure to just understanding the water resources and their availability, building resilience against climate change.
[00:56:26] And AI is increasingly a powerful tool to help solve some of those challenges. So I, think there's an important role for the water sector as the, hinge that connects both of those. Really want to thank our panelists today, all the great questions, all the participants, and I'll I'll turn it over to, Max to close us out.
[00:56:47] Max Herzog: Thanks so much, John, and definitely echo your thanks to our panelists. Really excellent insights shared here today, so thank you for bringing your perspectives. We will share in the follow-up email, we're short on time so I'm not gonna throw this slide up right now the survey that Landon mentioned that Virginia Tech and University of Virginia are distributing asking for perspectives from water and wastewater utilities about data centers.
[00:57:14] So really encourage folks to consider completing that if you're coming from that sector. Also please do stay tuned. We'll be announcing the next Water Data Forum session. It'll be taking place in November. Here in the next month or two I think we're gonna be focusing on workforce in the water sector, which also does obviously have some integration overlap with this topic in AI.
[00:57:37] So with that, I'll thank our panelists and moderator once more, and thank, thanks to all of you for taking time over your lunch hour, at least here on the East Coast to talk data centers with us. And yeah, we'll be following up with the recording and additional resources. So hope everyone has a good rest of your day
.svg)
