Vermont Legislature SmartTranscripts:

[Rep. Kathleen James (Chair)]: We're live. Great. Doctor Ren, hi. Are you here?

[Dr. Xiaoille Ren, Associate Professor, UC Riverside (ECE)]: Hey. Hi. Hi. Hi, everyone. Yeah. I'm here.

[Rep. Kathleen James (Chair)]: Great. I'm sorry that we're running a few minutes late. We wanted to just get you into the waiting room and thank you for joining us. I know you have written testimony, I believe, that you've submitted and we've posted on our website. So I'm going to give everybody a few minutes to pull that up. And then we'll go live. We'll let you know when we're live. Oh, we are live. We're live.

[Rep. R. Scott Campbell (Vice Chair)]: Ah. Okay. We

[Rep. Kathleen James (Chair)]: are live on YouTube. So why don't we give everybody just a minute to pull up your written testimony, and then we'll introduce ourselves. And we'll let you introduce yourself, and then you can maybe walk us through, if it's okay with you, walk us through your written testimony. Sure. And we'll take questions as we go along Mhmm. If that works for you.

[Dr. Xiaoille Ren, Associate Professor, UC Riverside (ECE)]: Absolutely. So it's just just a set of slides. And and I'm trying to make sure that everything that we say is in public. Right?

[Rep. Kathleen James (Chair)]: Correct. In fact, we are broadcasting we are live streaming even Sorry as we about that. No problem. Yep. Our committee hearings are public. We've got folks in the room who are interested in hearing the testimony. And then this is live streaming on our committee page. And it will be archived forevermore

[Dr. Xiaoille Ren, Associate Professor, UC Riverside (ECE)]: on

[Rep. Laura Sibilia (Ranking Member)]: our committee page as well.

[Rep. Kathleen James (Chair)]: So why don't we get started? I know you're joining us from afar, so we'll go around the room and introduce ourselves. We'll have anyone who's a guest in the room introduce themselves just so we know who's joined us. And then we'll turn it over to you and you can introduce yourself for the record and walk us through your testimony. So my name is representative Kathleen James and I'm from the town of Manchester.

[Rep. R. Scott Campbell (Vice Chair)]: And I'm representing Scott Campbell from St. Johnsbury. Chris Morrow, Windham, Windsor Bennington. Christopher Howland, Rutland four.

[Rep. Kathleen James (Chair)]: Dara Torre, Washington two.

[Rep. R. Scott Campbell (Vice Chair)]: Bram Kleppner, Chittenden 13, Burlington.

[Rep. Kathleen James (Chair)]: Great. And who in the room?

[Dana Lee Perry (Guest, The Clasen Group)]: Dana Lee Perry from the Classen Group.

[Rep. Kathleen James (Chair)]: Audrey Lida, interning for representative work. Super. Alright. For the record, doctor Ren, if you could introduce yourself, and we're eager to hear your testimony.

[Dr. Xiaoille Ren, Associate Professor, UC Riverside (ECE)]: Sure. So I'm Xiaoille Ren. I'm associate professor of electrical and computer engineering from the University of California Riverside.

[Rep. Kathleen James (Chair)]: Great. And I know you have written testimony. We all have it pulled up. So if you could walk us through, that'd be great.

[Dr. Xiaoille Ren, Associate Professor, UC Riverside (ECE)]: Oh, sure. So can I share my slides? Or

[Rep. Kathleen James (Chair)]: Yes.

[Dr. Xiaoille Ren, Associate Professor, UC Riverside (ECE)]: Yeah. It should be too. What am I just give me one second. It was here. But

[Rep. Kathleen James (Chair)]: No problem.

[Dr. Xiaoille Ren, Associate Professor, UC Riverside (ECE)]: Alright. So let me make it full screen, and then we can start.

[Rep. Kathleen James (Chair)]: Okay. Great. We can see your slides.

[Dr. Xiaoille Ren, Associate Professor, UC Riverside (ECE)]: Yeah. So I'm Alright. So, yeah, I'm very happy to present some our recent research on the power and water demand of data centers. It's basically a set of questions and some potential answers.

[Rep. Kathleen James (Chair)]: Right.

[Dr. Xiaoille Ren, Associate Professor, UC Riverside (ECE)]: So we're not trying to advocate for or against the construction of data center. Instead, we're just trying to provide some evidence based information. So, we all know that there is a growing demand for energy by the data center industry. There are different types of projections. And one of the most well cited projections from the Lawrence Berkeley National Lab report published in December 2024. So it shows that in 2028, The US data center center energy is projected to increase up to 6.7 to 12% of the national electricity demand, up from about 4.4% in 2023. So this is really a significant increase, especially if you consider that many of the data centers are located in certain communities, like those rural communities, which often don't have those infrastructure to support this type of demand. So one of the concerns people have about this build out of data center is the potential impact on their utility rates. So we did some recent studies on the correlation between large industry load such as data centers and the location of marginal prices in the wholesale market in Dominion, which is large electricity zone in PJM. So this is essentially showing that there's a correlation. So you said, if you have a large industry load in a certain area, there's going to be increasing the location of marginal price for the wholesale market, for the day ahead wholesale markets. And I should point out that this is not saying the residential utility rates will be increased because residential utility rates is determined by a lot of other factors. Wholesale market price is just one of the many factors.

[Rep. Kathleen James (Chair)]: So Doctor. Windsor, one sec. Your slides are jiggling, jumping around. I'm not sure why. I'm not sure if you're touching them on your end. They're jumping on our screen. Just it's not a big deal in case it's something that you're doing.

[Dr. Xiaoille Ren, Associate Professor, UC Riverside (ECE)]: It's still jumping because I'm-

[Rep. Kathleen James (Chair)]: You're not touching anything.

[Dr. Xiaoille Ren, Associate Professor, UC Riverside (ECE)]: I didn't touch anything.

[Rep. Kathleen James (Chair)]: Okay. It's it's not a big deal. And you're just to confirm what you said is that the wholesale markets may be rising, but that doesn't necessarily mean that utility rates will rise.

[Dr. Xiaoille Ren, Associate Professor, UC Riverside (ECE)]: Right, because utility have more complicated factors in the pricing. For example, the inflation, the operational cost, the labor, maybe some lawsuits. For example, in California, wildfire and some lawsuits against certain utilities actually increased their residential utilities rates in the end. Okay. So I move on to the next slide related to water, because there has also been some growing concerns around the data center water use. But if you look at those online discussions, we're often seeing two types of extremes. One is saying that the data center water is completely fake. The other side says data center is going to be taking away all of our waters. So let's see what is the reality. So here's one example of a cooling system. And this is just one type of cooling system. And this is from one of the large tech companies. So they're using cooling towers. And you can see the servers is using a lot of energy and that's generating heat. This heat will be moved to a heat exchanger. And this process is called server level cooling. And this is closed loop, or sometimes we just simply use air if the server if the server doesn't have a high power density. But once we move the heat from the server to the heat exchanger, we have to further dissipate the heat to the outside environment. And this is where the water usage could come in. And if in this case, the data center is using a cooling tower and cooling tower just evaporating water almost 20 fourseven to take away the heat. Another common usage of cooling system is this evaporative assistance. Essentially, we're using air to cool the heat exchanger to take away the heat. But during the summertime, when the outside air is really hot, that cannot take away the heat from the heat exchanger. We're going be using evaporator cooling to cool down the air first. And that really depends on the heat exchanger temperature set point. If they have a really high temperature set point for certain type of servers, then the evaporative cooling could be reduced to a very short, small amount of time of the year. And in some cases it could be even eliminated, but there have some limitations as well, which we'll be talking about later.

[Rep. Kathleen James (Chair)]: That assumes that you're using a public water supply. You're connected to your utility. And do you know how much water is lost to evaporation?

[Dr. Xiaoille Ren, Associate Professor, UC Riverside (ECE)]: Yes. So this is using, I would say over 99% of the time, these data centers are using municipal water, mostly potable water, but some cases they're also using reclaimed or recycled water. So as far as I see, there's no industry wide reliable statistics about how much water is from potable sources. But generally some company report over 90% and some company reports about 70% and other company typically don't really report how much percentage is the potable water, but they just say we're getting water from the municipal sources. So I would assume in those cases, their water is simply potable water.

[Rep. Kathleen James (Chair)]: Okay. Rick Campbell?

[Rep. R. Scott Campbell (Vice Chair)]: Yes, I had a question. I think you mentioned a second ago that some of the cooling is done with air.

[Dr. Xiaoille Ren, Associate Professor, UC Riverside (ECE)]: And

[Rep. R. Scott Campbell (Vice Chair)]: one of the things that we have heard is that some data centers use a lot of fans and the fans are so many fans that fan noise is an issue. I wonder if you have any information. I I know you're talking about water and power in your presentation, but I wonder if you have any information on the proportion of data centers that use that use air and and fans rather than water as a cooling medium?

[Dr. Xiaoille Ren, Associate Professor, UC Riverside (ECE)]: So roughly speaking, I would say maybe for the new construction about one third is using primarily air based cooling and they might be using evaporative cooling for certain days of the year, but they are primarily using air cooling. But I don't have a reliable statistic. I'm just referring to some of the industry reports and some news article as well. So it's not really just a rule of thumb.

[Rep. R. Scott Campbell (Vice Chair)]: That's close enough. You have a better idea than I do. Believe, thank you. So,

[Dr. Xiaoille Ren, Associate Professor, UC Riverside (ECE)]: and how much water do data center use? Well, according to the Lawrence Berkeley National Lab report, they show that in 2028, the water consumption could be roughly two to four times as much as the 2023 level. And it's about 35 to 70,000,000,000 gallons of water consumption. Consumption here is the evaporative portion of the water they take from the municipal sources. So in the technical literature, the water consumption is viewed as temporarily lost. But a more important metric for public water system planning is the water withdrawal. So usually the water withdrawal is about 1.3 to 1.5 times the water consumption because data center has a consumption ratio around 0.7 to 0.75. So essentially means that water withdrawal is a little bit higher than the water consumption. And this is the study that we recently did. We show that from 2024 up to 2030, The US water withdrawal could be generally is increasing, in 2030, assuming the baseline case where the water usage efficiencies remaining constant in the next few years, it could be the total water withdrawal in the high growth case could be rivaling the city of Los Angeles Department of Water and Power Supply. It's 150 to 160,000,000,000 gallons of water per year. And in the more optimistic case where the water efficiency is reduced by 10% each year, then we could reduce the water usage by about roughly half.

[Rep. Kathleen James (Chair)]: So Doctor. Ren, sorry. So these are kind of aggregated national statistics. Do you have information on how much water an individual data center would draw? Are we getting to that later? Is that something you know?

[Dr. Xiaoille Ren, Associate Professor, UC Riverside (ECE)]: We'll show some examples, but I can't really say just an arbitrary scenario, I don't know how much water they use because that really depends on their climate, their cooling system design. They could use zero water, which might be their own voluntary choice, or it could be a forced decision, which we'll talk about in the next few slides. Okay. So what are the benefits of water? Because using water sounds like a negative thing, but actually water is the most efficient means of cooling in many places according to a large technology company's sustainability report in 2025. And also more recently, just actually yesterday, an official press release from a leading technology company for their Louisiana data center campus, they showed that using evaporative cooling during the summer reduced the electricity demand by 25% to 35%. At the same time when the grid experiences the peak summer lows and regional power demand is at the highest. So the message is that there's a trade off between the water usage and the power reduction, especially during the summertime when the power grid is already at the highest stress level. So this is the benefits of using water. Using water is not necessarily bad. So just like we use water to wash our hands, it has some benefits which we shouldn't deny. Then the next question is given the huge benefits of water evaporator cooling to reduce the peak power demand, does The US have enough water for data centers? If you look at the total water withdrawal, know, it's just 160,000,000,000 gallons of water per year in aggregate and in the high growth case, doesn't sound that much. However, there's no national reservoir because these data centers are not getting water from the total national water budget. Instead, it's getting water from local municipal water systems. And this is where the bottleneck or constraint could be a binding factor. So here we are showing the immensely water usage of a hyperscale data center in Iowa. The city is serving about 70,000 residents plus a few tens of commercial and industry users. The first figure on the left shows the Mansley water usage of that hyperscale data center. We only have the Mansley data since water works in that city doesn't publish the daily usage. And the middle figure shows the percentage of the water usage compared to the largest 20 water users in the city published by that waterworks financial documents. So you can see that during the summertime, the water usage of the data center is much, much higher than during the wintertime. And this is due to the fact that this hyperscale data center is using evaporative cooling assistance only during the summertime. So they don't need water for the spring or for the winter. Mean, the minimum amount of water usage for those time is just for some humidity control and for their domestic water using in the data center campus.

[Rep. Kathleen James (Chair)]: Sorry, I don't really understand that. On the far right, monthly peaking factor. Can you?

[Dr. Xiaoille Ren, Associate Professor, UC Riverside (ECE)]: So the right most figure is a technical metric is called peaking factor. Peaking factors measures the maximum usage over the average usage. And here, since we only have the monthly data, we are showing the monthly peaking factor. Essentially, that's the highest monthly water usage of these water users compared to their average water, monthly water usage over the last four years. And for some users, which are not included in the financial reports for the entire four years, we do some adjustment as well. But generally you can see that the data center user has the highest peaking factor 4.3 among all the largest users disclosed by the Waterworks financial documents. So this shows that this data center has a very spiky water usage pattern. And this is only for the monthly usage. If you look at the daily peaking factor, according to some regulatory guidance, we need to multiply the monthly peaking factor by about 1.5 or higher. So we use the lowest value 1.5 that shows the peaking factor is about six. So this means if data center uses on average one unit water per day on average, in their peak days, their water usage will be 6.5 times, 6.5 unit of water. So this is a very spiky usage and this creates some problems or some challenges for the local public water systems. Since the public water system have to be designed based on the peak demand, based on the maximum demand. Similar to the power grid, we have to design the transmission lines and the generators, distribution network, everything based on the highs, the demand during the summer. So that's why we're, and also wintertime in the Northeast US. So that's why we are seeing power stresses in the summertime and also during the winter storms. But during the spring or fall, we rarely see any of the power grid strain. And similar to the water system, everything is designed based on the maximum demand. So if certain users has a really high demand during certain days of the year, the system has to be designed to accommodate those peak demand, even though the user doesn't use any water for the rest of the year. So this is a significant challenge and which is not reflected by the annual water usage. So some data center, for example, in this case, these days, by the way, these data centers is already using the highest amount of water among all the users in 2024 and 2025. But even though in this case, we look at the total annual water usage number is not super high. However, the impact is quite significant because of the peak demand.

[Rep. Kathleen James (Chair)]: Okay.

[Dr. Xiaoille Ren, Associate Professor, UC Riverside (ECE)]: Yeah, so this is another hyperscale data center in Wisconsin. And this is a primarily AI based data center, but they still have some in the public, they often say this is zero water. However, if this data center still uses quite a large amount of water for certain type of server, because not all the servers can be could be using this purely dry coating. So they are using, according to the planning document, the estimated peaking factor is about 30 or higher. So this is not the actual user because the data center has not been finished, has not been completed yet. So we we look at their planning documents released by the local, I think it's a county record or it's disclosed by the local media. So the peaking factor is about 30. And so because for this data center, they only use water for estimated around twenty days per year. And during the rest of the year, there's essentially no water use. So that's why we're seeing a really high peaking factor in this case. And in Vermont, because of this cold climate, I would imagine or expect that those large data center, if they are using evaporative water assistance, their peaking factor is also quite high, maybe similar to this Wisconsin case. So according to the Susquehanna River Basin Commission website, they have a FAQ for data center water. They say that most communities don't have the large surpluses of excess water supply readily available to provide to large hyperscale data center. It's not really about the annual water usage, it's mostly about the peaking factor. So during the summertime, when everybody's using water most, this water system may not be able to accommodate those high demand from the data center industry, even though the data center could be using a really low annual average water. So we interviewed about a water capacity request of a 6,000,000 gallons per day water request from new data center project. And some representative from Newton County in Georgia said, we just don't have the water since the infrastructure is aging, is already stressed, doesn't have the available capacity. And even though the county's total capacity could be enough more to supply this 6,000,000 gallons per day of water. The available capacity is less than six MGD. I'm not sure in Vermont, but in California is really, it could be illegal to kick out some of the existing customer to accommodate the new water use from another user. So it's what really matters is really the available capacity for the public water system. So that's why in many cases, these water capacities are binding and unrecognized constraint. And this is, they could be forcing data center to rely on waterless, less efficient dry cooling. And this can further increase the electricity demand during the summertime and add more stresses to the grid. So we are seeing zero water usage data centers, but this zero water choice may not be voluntary. It could be forced by the local water capacity constraint. Sorry.

[Rep. Kathleen James (Chair)]: Well, maybe you're going get to it. I was going to ask about in Vermont, one of the things that we've been talking about a little bit is we're a larger data center, you know, be interested in siting here. Mhmm. Wouldn't necessarily be using municipal water or public water, but would be pulling out of, like, the Connecticut River. Is that some have you looked at do you have data on or can you fill us in a little bit either at the end of your presentation or now about data centers that are pulling from other sources that aren't public water?

[Dr. Xiaoille Ren, Associate Professor, UC Riverside (ECE)]: Right, so according to my survey, we look at the 15 large data center, including five hyperscale data centers and 10, sorry, it's 11. So five hyperscale plus seven large collocations. Water usage is over 99% of their water is coming from municipal sources. Some are from their private groundwater sources like a well. Maybe a couple of cases is also getting water from rivers, but that's not very common. But it's possible. Actually, we're seeing some data centers in Europe just directly using seawater because of the unique location. So if the location is suitable, they could be getting water from the river, but then there's issue of thermal pollution since when they use the river water to cool their system, the water discharge could be a couple of degrees higher than the incoming water temperature. So that's considered as a water pollution, thermal pollution. So they need to somehow maybe retain the water for a while before it is discharged. So it can come back to the normal temperature. Yeah. So river water for cooling is possible. It's just not being used very often, very rarely in The US today.

[Rep. Laura Sibilia (Ranking Member)]: Yeah, R. Sibilia? Yeah. So Doctor. Wren, I think I might have mentioned this to you previously. There's one site that's been identified as a possible, which is a former nuclear power site, which is alongside the river and which did use cooling obviously there as well.

[Dr. Xiaoille Ren, Associate Professor, UC Riverside (ECE)]: Yeah, that's very, technically doable. And so that in that case it to reduce the pressure on the local public water system. So what is the challenge is really how we can accommodate this new industry peak water demand in those small communities. Some general background, are about 150,000 public water system in The US and around 50,000 community water systems supplying water constantly to, I think more than 25 users year round. So then more than 90% of them are small or medium. But this is compared to about 3,000 power utilities. So water systems are a lot more fragmented and under resource in most cases. And also there's about 1,300,000,000,000.0 funding need in the next round, roughly twenty years, according to the EPA estimates. And also water service affordability is rising as a national challenge. So around 10 to 15% of the households have affordable water services. But we're seeing this increasing, like more common corporate community partnership. Like yesterday, large technology company just announced that they will be investing up to $400,000,000 to upgrade the local water infrastructure to support its data center needs, hosting AI and cloud services according to their press release. So in this case, I think this clearly shows the benefits of using water because this is a huge investment and they want to use water to reduce the stress on the grid during the summertime. And also a lot of companies have been using this, have been doing this something called water positive. They try to replenish more water, but this is only looking at the total annual volume metric number. Doesn't reflect the peak water demand. So in our ongoing research, we are also recommending this pipe neutrality. So essentially, when data center take away this community community water system capacity, they need to add more capacity to sort of compensate for their capacity usage. So this similar to the water positive, but we call it pipe neutral. And also we need to coordinate water and power planning since the benefits of water should be recognized as using water is not just not always a bad thing. If you have water resources available, then using water reduce the power demand, that's usually a competitively viable choice. A life with no water, but that's a different story. With that, I also like to finally emphasize that according to the environment, environmental and the energy and environmental Economics Institute, the water availability is rising as a key challenge in the next few years. Yeah, so I'm done with my testimony.

[Rep. Kathleen James (Chair)]: Great. Yeah.

[Rep. R. Scott Campbell (Vice Chair)]: So I'm I'm looking at this at your last table here, and it appears to show you anticipate water availability to be to be less important in five years. I'm just is that am I reading that right?

[Dr. Xiaoille Ren, Associate Professor, UC Riverside (ECE)]: No. Actually, number two means the It's more important. Yeah. The better. The more important. Number one is the Oh, I was talking about that's the most important.

[Rep. R. Scott Campbell (Vice Chair)]: Alright. Got it. Thank you.

[Dr. Xiaoille Ren, Associate Professor, UC Riverside (ECE)]: So today, power availability and cost, that's the constraint and everybody's talking about it. But in the future, water is going to be even higher position in a higher position than power availability.

[Rep. Kathleen James (Chair)]: It's exactly

[Rep. R. Scott Campbell (Vice Chair)]: Okay.

[Rep. Kathleen James (Chair)]: Yeah. Okay.

[Rep. R. Scott Campbell (Vice Chair)]: Yes. Have you actually visited one of these facilities and what's your overview of the performance of the facility?

[Dr. Xiaoille Ren, Associate Professor, UC Riverside (ECE)]: So which I visited some large colocation data center many, many years ago when I was in Miami. But typically, those data centers are highly secured. So they don't like visitors to come to their place unless you're one of their potential customers.

[Rep. Kathleen James (Chair)]: All right.

[Dr. Xiaoille Ren, Associate Professor, UC Riverside (ECE)]: So at that time, my impression was the data center was because it was more than ten years ago. So at that time, the cooling system was set as a very low temperature. So essentially we have to wear a jacket if we visit data center. But today it's very different. So a lot of data centers are using a higher temperature set point. Instead of wearing a jacket, we have to wear shorts to visit those data centers because it's warm. Increasing this temperature can help reduce the power demand, reduce the water usage because we can use the cooler outside air for more time of the year. So the system is getting improved over the past few years. But we're kind of reaching a plateau in terms of the efficiency improvement. So unless so that's why still even the newest data center is wanting to use water vapor recording at least during the summertime.

[Rep. Laura Sibilia (Ranking Member)]: It's really great.

[Rep. Kathleen James (Chair)]: Thank you. Yeah. This is really helpful testimony. Thank you so much. Any further questions for Doctor. Renn?

[Rep. Laura Sibilia (Ranking Member)]: Thank you very

[Rep. Kathleen James (Chair)]: much Doctor. Renn. Thank you very much for making time and for joining us.

[Rep. R. Scott Campbell (Vice Chair)]: My pleasure.

[Rep. Kathleen James (Chair)]: All right, thank you. All right, we can go offline.