Latest / Technology Now / Power, water and place: what makes the optimal data center location? | Andrew DesRochers
Transcript
- MICHAEL BIRD Sam, Sam, Sam. as you know, we talk about the impact that AI has had on our organizations and our infrastructure quite a lot on this show. This is a technology podcast after all. Do you remember dial-up internet? SAM JARRELL Oh yeah, and the noise. Ugh MICHAEL BIRD We talked about it on the show before. my dial-up internet sounded like somebody doing a sick electric guitar solo, but yours sounded less exciting SAM JARRELL No, it sounded like a, the screams of a robot. MICHAEL BIRD Oh my goodness. The screams of a robot is the perfect Ah. Yeah, you're right. anyway, do you remember in the days of dial-up internet, we expected things to load slowly, and I think we had more patience.but as the technology improved and things got faster, We expect immediate answers, However, have you noticed that actually the way that we wait for answers with AI chatbots, They take time to think and answer the question SAM JARRELL So in a way, like we've come full circle and we're reintroducing patients back into the system MICHAEL BIRD We have. and this expectation of latency has given designers a bit more leeway when it comes to being able to design new data centers and infrastructure sustainably. Because if people are willing to wait a few seconds for an answer, we can move the processing to somewhere which may be less of a drain on local power or water supplies SAM JARRELL So we’re talking sustainability today? MICHAEL BIRD Yes, we are talking about sustainability today I’m Michael Bird SAM JARRELL I'm Sam Jarrell MICHAEL BIRD And welcome to Technology Now from HPE. MICHAEL BIRD Now, for a long time when we've talked about the impacts of AI, we focused on just how much energy they used. However, in the past few years or so, the conversation has begun to shift a bit SAM JARRELL Right, because it's not just about how much power we need to run these models, is it? The data centers that also house our AI need to be kept cool, and that means we also must talk about water MICHAEL BIRD Yeah, exactly. It's a bit of a hot topic at the moment, isn't it? And the way that we cool our data centers can vary dramatically based on location, for example. So it's generally much easier to keep everything at a nice cold temperature here in the UK than it is for you in parts of sunny California SAM JARRELL Yeah, that's true. But also, this can be a slightly prickly topic to talk about as we mentioned, and for a very good reason. People are understandably worried about the impact that a new data center might have on them if one were to open up nearby MICHAEL BIRD Yeah, very true. And it can also be difficult to actually separate fact from fiction when it comes to working out what's actually going on. So, to get a bit more of an insight into how data centers can be run and scaled up sustainablyI met with Andrew DesRoches, a Principle Technologist for Sustainable Transformation at HPE and I started off by asking him just how much water is actually used in the deployment and day to day running of an data centre. ANDREW DESROCHES that's really the difficult part is we don't have an exact number because not only do we estimate that every data center is different, we can't say that, a large AI training data center can be even comparable to maybe a smaller enterprise grade data center, vastly different qualities and not all of them are necessarily consuming water as a resource. There's also not a lot of good transparency into that data, right? Operators aren't necessarily tracking their water consumption, if they are consuming water, and they're not being transparent about that data. So We don't have those numbers today, and I think that lack of transparency is where a lot of that scrutiny that we're hearing today is coming from MICHAEL BIRD What I don't understand is why data centers consume water. I understand you would initially fill up a loop, You put, some, coolant in it, some, I don't know, anti-biological growth, That feels like that should be a closed loop. So where is the water usage actually happening? ANDREW DESROCHES Yeah, so that's the interesting part, right? For folks that are not fully immersed in the space and aren't- That's the fun part. MICHAEL BIRD Excuse the pun. ANDREW DESROCHES and aren’t n Necessarily thinking about liquid cooling or enterprise technology cooling in general every day, they may hear liquid in the name liquid cooling and think, "Oh, well, that's where my water consumption is going. That's why in my local community now, I have reduced water pressure, the quality of my water has drastically been impacted." And despite the fact that it says liquid on the tin, you're absolutely right. Direct liquid cooling for the most, for the majority of cases, that's a contained loop. Yes, we may see some level of loss with purging that loop for cleaning or, some level of disposal to make sure that there's no particulates in that water. The vast majority of water consumption in data centers is because of the facilities level cooling loop, and that's because we've historically seen operators tor-turn to evaporative cooling as the solution of choice because of its inherent energy efficiency, right? And that consumes, the vast scales and vast quantities of water that we're hearing scrutiny about today. And so that's our first recommendation to operators that are looking to build a new facility is assess your cooling options at the facilities level. We're starting to see that dry air cooling for the facilities cooled systems is actually showing price and energy efficiency parity with evaporative cooling, and that it may not be the most economical or even energy efficient option anymore just to grab, evaporative cooling off the shelf every single time MICHAEL BIRD And so evaporative cooling is where you would what? Run water over a radiator and use the effect of water evaporation to sort of evaporate the heat with it, for want of a better phrase. And why does that use so much water? Maybe it feels like an obvious question, ANDREW DESROCHES Correct. Well, it's, it's obviously not destroying the water. It has to go somewhere right? It's evaporating that water away and sending it back up into the atmosphere. And again, that was seen for years as really energy efficient because it's taking advantage of the natural properties of water to suck that heat away and evaporate that heat off, right? And now what we're seeing is that water is traditionally being pulled from local water utilities. And as AI workloads are coming online, and as our data centers are becoming more and more dense and more and more, heat intensive to accommodate these AI workloads, it's consuming more and more water to cool these systems down. And so that's where we're seeing the, major levels of consumption. And on a more positive note, because of the innovations in the space the facilities level cooling systems are refreshed multiple times over the life cycle of the data center building, right? The building is gonna outlast the cooling system. Data center operators have the opportunity to look at those options and assess the most economical and resource efficient option for them, right? It's not just solve for a single variable of energy anymore, it's solve for the multi-variable equation that it is running your data center operations as efficiently as you can MICHAEL BIRD So f- for evaporative cooling, the issue is taking clean water that people can drink. Why can't you just take, I don't know, water from the sea or, or gray water from I don't know, people's showers, whatever? Yeah, ANDREW DESROCHES Yeah, that's a great question, and I can't speak to that specifically, right? I wouldn't say that I'm an expert in the corrosive properties of seawater or what have you. But what we are seeing is that operators are looking at these out-of-the-box solutions to cooling their workloads. They're asking themselves the questions of, "Can I move this workload to a potentially colder climate to take advantage of the, the naturally colder atmosphere or the naturally colder climate to, to cool this workload for free? And I don't need to use energy. I can just, tap into the local cool climate to do so." And that poses a number of challenges for our operators around performance of their workloads, latency of their workloads, serviceability. Can they get people there to take care of the equipment as it needs maintenance and refresh? So it's, it's not just a energy equation. It's not just a cooling equation. We can't make sacrifices for performance at the same time. And so balancing all of these things while driving efficiency as part of the strategy is critical for our customers to solve. That's why, unfortunately, there isn't a one-size-fits-all solution. And so our team is actively working with our customers today to, to develop the most intentional and efficient strategies for their specific use case, for their specific workloads, and to meet their business needs while driving efficiency at every turn MICHAEL BIRD I mean, do you think we'll start to see organizations considering, thinking more about the geography of their data centers? I mean, I guess typically today it'd be like, "Oh, we'll put a data center next to our head office," But could you see a, a market opening up in, northern Canada and the Nordics and, places like that where there maybe is some infrastructure available, but actually the ambient temperature's much lower? ANDREW DESROCHES 100%. I could see it happening, but it's not gonna happen overnight. It's gonna be a slow and gradual change. But the positive momentum that we're seeing is actually coming from a really interesting place. We're hearing from our customers more and more that their users are, more accepting of latency because of their experiences with generative AI, right? Think about your own experience with gen AI, right? You see that answer built out in front of your very eyes. You see it writing its thoughts out And oftentimes that's happening faster than you can read it yourself. And so operators are starting to ask themselves, "Are our users more tolerant to latency? Do we need millisecond latency or is half a second okay? And with that latency, can I site a data center or can I site my workloads in a colder climate to take advantage of those free cooling days and cut down on my overall energy and resource consumption?" MICHAEL BIRD That's a really good point. We've gone from typing a webpage in a web browser and expecting it to immediately appear, fully load, to now the use of LLMs where you see it sort of working and then it kicking out word by word. it's not gonna be right for every single workload, right? If you need a answer fast to, solve a, critical challenge, a health risk, a safety risk, that can't be tolerant to latency. We need to, make sure that that is robust, accurate, and fast. ANDREW DESROCHES You know, if we have, customers in the banking industry that are supporting high-frequency trading or, you know, workloads that require almost instantaneous response times, that's another variable we have to throw in this equation to understand the intention of the workload and then design efficiency around that. MICHAEL BIRD I suppose there's maybe a sovereignty element there as well. Like, some organizations will say, "Well, regardless, it has to be within this country or within our organization. It can't be in another country because the data requirements has to be within the EU or has to be within a particular country or set of countries. So in that sense, if our country's in the desert, we have to put it in the desert, or if our head office is in this town or this city, it has to be, close by or whatever ANDREW DESROCHES For sure. For sure. And I think it's important for customers to consider the different geographies, the different climates across their country, and understand how they can be efficient within their means, right? Some countries may have d- drastically different climates across their different regions, right? Especially here in the US we can look to site a data center in an area that is more acceptable to dry air cooling, and we can tap into local renewable energy resources, then we're designing the most efficient solution possible, right? It's not just, "Hey, we're gonna site our data center here because this is where it's always been." We're encouraging our customers to take that zoom out and really assess what the best option is for them and their business realities today as they're looking to meet the needs of their business. MICHAEL BIRD do you think governments are gonna start regulating...I know, in the UK for example, the energy companies are saying, "You have to let us know if you wanna use more than this amount of energy," and, you know, "You have to consider how or where that energy's gonna come from." Do you think that the same is gonna be the case for water? ANDREW DESROCHES I think so. I think that we're gonna continue to see scrutiny on energy as a resource, and then water is that next horizon. And we're not necessarily seeing it at least here in the US, from government organizations. We're actually seeing it from communities. We're seeing it from community leaders, from activist organizations, asking the question of, "Hey, if this data center's gonna come to my town, are my water resources going to be impacted? And I think that we as data center operators, we as technologists, need to be doing a much better job collaborating with those communities and showing them the levels of efficiency that we're striving for, being very transparent about our operations in order to build that relationship, quell those fears, while still being able to execute on our goals, right? And so I think that as technologists and as data center operators, we need to be collaborating with our community to increase that transparency and reduce those fears in order to be successful in this new age, MICHAEL BIRD and of course w- we're seeing more and more AI workloads, so that's presumably changing the conversation again because, a, there is a, a desire for these data centers to be built, presumably quite quickly ANDREW DESROCHES And as much excitement as there is, we're also seeing an equal level of failure rates. We saw that in 2025, 50% of the proposed data center projects here in the US actually never came to fruition. They didn't break ground. They didn't start their build. And there's so many reasons why that could happen, but a big reason for that is pushback from communities. Now, I do want to go back to a point that I made earlier about a data center being a data center being a data center. I think that because of the excitement around AI and because of the acceleration that we've seen over the last couple of years, the average non-technical person that hears a data center is coming to town is probably thinking of these large scale AI training data centers. Now, those operate at a completely different scale than your average enterprise is going to operate at, right? Your average enterprise is not going to be training a net new model. Very rarely are they going to do so. They're generally going to be looking to bring a model, uh, into their organization, do some level of tuning and deploy it for inference. And so I think, again, going back to that point about community collaboration, being really transparent about what the work that you're going to be doing, what scale you're going to be operating at and what level of resource impact your operation is going to have, is critical for us to gain that access and gain the right to operate in the areas that we need to operate in MICHAEL BIRD Andrew, thank you so much for joining us on Technology Now, really appreciate your time ANDREW DESROCHES of course, michael.. Absolutely. Happy to be here- SAM JARRELL That was an interesting conversation. obviously you would wanna be able to take advantage of like cooler climates. But when having like the sovereignty discussion along with that, not every country necessarily has access to the same climates within their geography, right? In the US we have varied climates across the entire country, so it's easier to maybe consider putting like a data center in somewhere that's very cool like Alaska and keeping it maybe out of a very, very hot place in the south if that's better to take advantage of that climate. And it's all still underneath the US's control. But if you have a smaller country or you just don't have that varied geography, then you really don't have that opportunity. MICHAEL BIRD Yeah. and I think my big takeaway from the interview was just, how important location is for so many reasons. there is the location element as it relates to the climate and how cold it is or how hot it is, and therefore the impact that might have on, how you cool, the equipment. Access to, power infrastructure. You know, is there renewable sources close by that would make the sustainability element of it and, possibly the costs lower from a power perspective. is it easy to put, fast data connections? Are you close to actually where the data needs to go or the, the compute needs to be accessed from? and, there's the, sovereignty element of it. so location is quite a crucial part of it and actually latency being perhaps in some instances much less of an issue, perhaps also means that, that, that location conversation, like the net can be widened. But I suppose the other location element is the local community as Andrew pointed out SAM JARRELL Yeah. you're right, and the other piece of this being, the local communities and whether or not people are comfortable with having a data center popping up in their neighborhood. A few years ago, I feel like this conversation didn't even really exist, but now there is a lot of confusing information out there, and, much of it quite negative, but people are right to ask questions. And I appreciate, that Andrew emphasized the needs to be transparent with communities and be clear with them about what data centers do or do not do and also how, the particular data center going up in their neighborhood, like, how does it use power and water? Because it's not a straightforward conversation. as, you might believe from what you hear online. MICHAEL BIRD There's more nuance to the conversation, which I think needs to be communicated correctly. and I think that impacts organizations because, whether you'll be building a data center yourself or utilizing a data center, I think like organizations probably need to be thinking about how that can be communicated within the workforce or within the wider customer base. Because there probably will be questions as to, where does this data live, not just from a sovereignty perspective, but also how energy efficient or how sustainable, how much water is it using, all that sort of stuff. SAM JARRELL Yeah, I even f-found it interesting to just, learn a little bit about, dry air cooling as well. I'm wondering, though, if there will be questions around, how much power does that use versus other things in the future as new and new, newer and newer solutions start to come online, will that make communities more comfortable with data centers? MICHAEL BIRD There could be a weird scenario where energy efficiency isn't necessarily the be-all and end-all. Like for example, in, in the UK, particularly the north of England and Scotland, we've got quite a lot of wind turbines, A lot of the energy gets used further down south in the UK, but the energy generation is much further north. Would it make sense to put those data centers closer to those wind farms where you could maybe have like a direct connection and actually energy efficiency isn't the big thing there because there's so much abundant energy that the grid simply can't use, and actually the climate's much cooler. We're very fortunate in the UK. We have the North Sea, which is quite shallow and quite windy, so it's perfect for wind turbines. But, I suppose if you're in a much hotter climate, you maybe get more sunlight, so maybe solar works Maybe we can get Andrew back, and we can dive into some of those topics in a bit more detail. MICHAEL BIRD now, Sam, our knowledge of how to be more efficient and more sustainable with our technology is evolving as fast as the technology itself, as we said. So the final thing I asked Andrew was very simple. If we had this conversation again in 2030, which is just about three and a half years away. Three and a half years away? No, I'm not ready for that news. It's much further away, Sam. how did he think it would change? ANDREW DESROCHES Well, I'm an optimist at heart, so I'll put it this way. I hope that we are not having this conversation. I hope that we have reached a point where efficiency by design is just a way that our IT practitioners and our data center operate- operators operate. I hope that we are, driving efficiency at every turn, that we're conscious of our energy consumption, conscious of our water consumption, conscious of our impact on our local communities, and that we're not necessarily having this conversation around, communities pushing back in droves and increased scrutiny on AI energy consumption. I really do hope that by embedding efficiency by design and collaborating with our communities, that we're not having this conversation in 20 years. SAM JARRELL Okay that brings us to the end of Technology Now for this week. Thank you to our guest, Andrew DesRoches And of course, to our listeners. Thank you so much for joining us. MICHAEL BIRD If you’ve enjoyed this episode, please do let us know – rate and review us wherever you listen to episodes and if you want to get in contact with us, send us an email to technology now AT hpe.com. Sam, what should, what should the subject line be? SAM JARRELL Location, location, location MICHAEL BIRD and don’t forget to subscribe so you can listen first every week. Technology Now is hosted by Sam Jarrell and myself, Michael Bird This episode was produced by Harry Lampert and Eva Higginbotham with production support from Alysha Kempson-Taylor, Nik Damarell Beckie Bird, Alissa Mitry, and Jenessa Ayache. Our theme music was composed by Greg Hooper. SAM JARRELL Our social editorial team is Rebecca Wissinger, Judy-Anne Goldman and Jacqueline Green and our social media designers are Alejandra Garcia, and Ambar Maldonado. MICHAEL BIRD Technology Now is a Fresh Air Production for Hewlett Packard Enterprise. (and) we’ll see you next week. Cheers! SAM JARRELL Bye y’all