“We're living under a water stressed world. Maybe not today, but a certain time of the year.” This new reality stated by Dr Shaolei Ren questions both the resiliency and the water footprint of the data center industry. 

In this episode we bring an academic perspective on these questions with two of the most renowned experts in the field, Dr Shaolei Ren and D’ Mohammad Islam. They conducted an intensive survey in the US and shared many insights about the water consumption of data centers with their host Gaël Duez such as: 

 And much more!

❤️ Subscribe, follow, like, ... stay connected the way you want to never miss an episode, twice a month, on Tuesday!

📧 Once a month, you get carefully curated news on digital sustainability packed with exclusive Green IO contents, subscribe to the Green IO newsletter here. 

📣 Green IO next Conference is in Paris on December 9 - 11. Every Green IO listener can get a free ticket using the voucher GREENIOVIP. A small gift for your huge support. 🎁 

 

📧 You can also send us an email at [email protected] to share your feedback and suggest future guests or topics.   

Transcript (auto-generated)
Shaolei Ren (00:01)

Data centers generally will evaporate 50 to 80 % of the water they withdraw. And on the other hand, if you look at the urban household setting, typically the water consumption rate is around 10%. So the way that we use water and how we discharge water is very different.

Gaël Duez (00:22)

Hello everyone. Welcome to Green IO. I'm Gael Duez and in this podcast, we empower responsible technologists to build a greener digital world one byte at a time. Twice a month on a Tuesday, our guests from across the globe share insights, tools and alternative approaches, enabling people within the tech sector and beyond to boost digital sustainability.

A month ago, I released the episode, Thirsty Data Centers in the heart of Silicon Valley with Mashaika Allgood, and it became instantly popular to say the least. A week later, Google released its report on the environmental impact of a median Gemini query. This report ignited several heated discussions, both on the energy and water consumption figures.

I decided to go back to the force of knowledge or actually to those who made their living in producing scientific knowledge, academics. And when I asked around about the most prominent researchers in the field of the environmental impact of data centers, especially their water consumption, two names popped up at all the time. Dr. Mohammad Atiqul Islam, an assistant professor in the Department of Computer Science and Engineering at the University of Texas and Dr. Shaolei Ren, an associate professor of electrical and computer engineering at the University of California. Both have written several papers on the topic. They have not shied away from entering the public debate with clever vulgarization. So welcome Mohammad and Shaolei. It's a pleasure to have you on the show today and thanks a lot for making it possible despite our huge time difference.

Shaolei Ren (01:59)

Thank you for having us.

Mohammad Atiqul Islam (02:02)

Thank you and you are most welcome.

Gaël Duez (02:03)

Thank you. So being lucky enough to have two academics with me tonight, or actually this morning for you, we can start by doing the things in the right order. And before jumping into the debate about the scale and the trend, let's define the scope. And actually defining the right scope was part of the alert about the Gemini report that Shaolei shared on LinkedIn. Some key figures were misinterpreted. So, Shaolei, could you tell us how do data centers consume water? What do we talk when we water consumption of data centers?

Shaolei Ren (02:43)

Sure. Data center uses water both directly and indirectly. So the direct water consumption is primarily for cooling. Data center don't have to use water, but water is a very efficient approach to cool down the data center facilities. So essentially we are using water evaporation either throughout the whole year, 24-7 or during the summertime to take away the heat to the outside environment. And this is the direct water consumption. Additionally, when you generate the electricity, especially using thermoelectric power plants and sometimes even using hydropower, there's going to be some water loss into the atmosphere. And this is also considered water consumption or indirect water consumption by data centers. So if somebody wants to take a life cycle view of the water usage, they also look at the supply chain. But ⁓ unfortunately, we don't really have concrete data to have quantitative analysis for the supply chain part. So in most of the studies that I have already seen in the past, they focus on direct and indirect for electricity, not for the supply chain.

In our paper, we call it scope one and scope two, following the greenhouse gas protocol definition. So essentially, we have two scope of water, scope direct for cooling and indirect for electricity generation. And sometimes people show the total number by default. But in our paper, we actually show both the separated water consumption across each scope and also the total number.

Mohammad Atiqul Islam (04:18)

That's basically the two different kinds of water consumption that you have. And again, I want to ⁓ emphasize on the fact that you do not have to use water-based cooling in your data centers, but that's the best approach that you can take. If you do not use water-based cooling, it's called dry cooling, and people are actually paying a lot of attention across different industries, not just data centers, to build better dry cooling systems. But again, the problem with those dry cooling system is that they are less energy efficient. So basically to run the data centers, not only the servers that run those workloads and run those computation only take power, but other components in the data center also takes power. So you want the other components to be also energy efficient. A great example would be your own homes, a typical residential cooling system, which doesn't use any water rather use electricity to your home. So basically you can use a similar approach for data centers, but they typically twice as much energy compared to the data centers that use water in the cooling.

Gaël Duez (05:28)

So means that there is a sort of a trade-off between energy consumption and water consumption, if I understand you right.

Mohammad Atiqul Islam (05:33)

Yes, and later when we are going to discuss what we can do moving forward, this will come up and I will also want to like discuss a little bit more how this idea of dry cooling can be actually utilized to make data center more sustainable.

Shaolei Ren (05:48)

So maybe I can add some more context for the water consumption because that's actually a technical term. It's defined as the difference between water withdrawal and water discharge. So let's say when we take a shower, we are using a lot of water and that water is considered water withdrawal. It's not by definition, not water consumption because most of the water that we use for showering will just go straight into the sewage system and that's considered water discharge. If you look at the difference, the difference is really minimum. Data centers generally will evaporate 50 to 80 % of the water they withdraw. And on the other hand, if you look at the urban household setting, typically the water consumption rate is around 10%. So the way that we use water and how we discharge water is very different.

Gaël Duez (06:38)

This is the moment I'm going to ask the dummy question. The water on planet Earth is a closed loop, a closed circuit. Why does it matter that water is either evaporated or goes down the drain?

Mohammad Atiqul Islam (06:55)

So I can take this question. So basically the problem with when you have evaporation based consumption, so you are releasing the water into the air and it actually floats away. And for let's say for example, I'm in Texas and the water that flies away from Texas because of evaporation, it will take hundreds of years actually to come back. Even though we think of the water cycle, the water cycle is actually pretty slow. So the basic idea is it's not like the water is getting away from Texas is going to come back to Texas. Actually it's going to go back somewhere else where they already have a lot of rain, probably going to increase the rain there. So this is actually a water displacement problem that is happening everywhere. That is why we have these drought problems. So if you keep this going on, that evaporation is going on, more consumption is going on, the dry places are gonna get more dry, and wet places are gonna get probably more rain. So that is why to look at the consumption as well, even though the water is never leaving the earth, right? But still, the displacements of water are actually the problem that we want to address. That is the environmental impact of water, basically, for data center, hopefully.

Gaël Duez (08:10)

And for the other scope, indirect scope, is it the same issue? Is it mostly water withdrawal or water consumption?

Mohammad Atiqul Islam (08:20)

In both cases, that's water consumption that we're talking about. So when we calculate the water evaporation, we are also thinking of water evaporation from the power plants. It's not like the water withdrawal of the power plants. There are actually, different ways of cooling those power plants. One of the ways is they do not use evaporation, but rather let's say they get water from

Gaël Duez (08:24)

Okay.

Mohammad Atiqul Islam (08:44)

A river and then they run it through their system, the water temperature goes up. This is how they are releasing the heat. Then they're releasing the warm water into the river back. So you have very minimal water consumption. But one of the major problems with those kinds of power plants are basically they need to withdraw a lot more water than a power plant which is evaporated. So that's why even though another big problem with those kinds of power plants are the geographical location you need to be close to a water source which can give you a lot of water plus the temperature difference that you can have ⁓ the discharge water versus the water that you take in it cannot be too high because otherwise it will drastically affect the water condition of the of the water so it's actually heat pollution into the water so that's why right so that's why

Gaël Duez (09:34)

Heat pollution for the biosphere. Okay, got it.

Mohammad Atiqul Islam (09:38)

Evaporative cooling is the dominant cooling system across the industry.

Shaolei Ren (09:43)

Yeah, I think besides water consumption, water withdrawal is also a very important metric because that somehow reflects the competition among different water users. Let's say when we plan the public water systems, we have to make sure everybody is using the water, we are able to supply the demand with enough capacity. And in that case, water withdrawal is also a very relevant metric. And also if you look at different users competing for the water resources, not really competing, but they're sharing the water resources. There's some time there's a complex concept called water rise. And that water rise is defined based on the water withdrawal, not necessarily based on water consumption. So in the environmental sciences field, I think those water withdrawal and water consumption are important metrics by default, when we look at the water footprint of certain things like ⁓ hamburgers or data centers, we look at water consumption.

Gaël Duez (10:49)

Okay, got it. And before we go into the scale and we enter the repetition between the scopes and what is at stake, I have one last question. And I'm sorry for my listener if it sounds a bit Eurocentric, but this is something that I hear a lot among European experts. And it goes like this. Yeah, you know, the water consumption of data centers is really a US thing because in Europe, we don't have water cooling and the way we produce our energies is maybe a bit less water consuming. So I will skip the second part because I don't have the ability to judge, but it seems that yes, the fact that we don't use that much evaporation tower seems to be positive from a sustainability angle. And yet one month ago, Masheika Allgood kind of shaked my beliefs on this ground saying that actually not that much and having or not having an evaporation tower doesn't prevent you to a significant amount of water consumption. So could you some lights on this discussion also?

Shaolei Ren (12:02)

Yeah, data center have different options for cooling system. can either use dry coolers and water evaporated cooling or ⁓ natural air cooling with water evaporated systems. So just as Mahammad pointed out, this type of system usually have some trade off between the energy consumption and water. ⁓ It's a regional thing. And in Europe, I would say probably uses less water on average than the US data centers due to perhaps some stricter environmental laws. On the other hand, I think I've seen some documents saying that European data center do use some water and not zero. And also there's more research on waste heat reuse from data centers in Europe than in the US. I think in the US, waste heat recovery or weight heat reuse from data centers is essentially zero. But in Europe, there are quite a few cases, even during the Paris Olympic Games, there was some data center using waste heat from the servers to warm up the swimming pool. But there's no such thing in the US. So ⁓ it's not surprising that the European data center uses way less water or less water, as I would say, than the US data centers. So perhaps in Europe, the water issue is less severe compared to in the US. But I would say, we need more data to understand the problem better.

Gaël Duez (13:36)

And right now, Shaolei, I feel terribly stupid because actually the heat waste and the heat we use is actually a topic kind of high in my mind, especially because of the European energy directive, which now makes it compulsory for data centers above a certain electricity consumption to report on these metrics. But… I've always connected this metric with energy consumption, like don't waste good energy, decarbonized energy. And now that you said it makes absolute total sense that actually if you don't have to cool down your data center, because actually you used the heat somewhere else, actually it should also improve the water efficiency of your data center. So that's my ⁓ humbling moment, I would say. But it makes total sense.

Shaolei Ren (14:24)

There are some challenges with the waste heat reuse. Technical challenges include that the heat from the data centers are typically low grade heat, which means the temperature directly from data centers heat is not high enough to be directly used. But this is a challenge that we can address partially, I would say, at least. ⁓ On the other hand, the challenge is the data center needs to be close enough to the heat users. Let's say office buildings or residential communities. In Europe, this is probably easier to achieve because people tend to live closer and more concentrated living. But in the US, it's more of a real challenge than elsewhere.

Gaël Duez (15:13)

Plus you've got other pollutions like noise pollution, you've got air pollution from backup generators and so Mohammed said right from the start, it's a trade-off all the time. And one last question to define well the scope. What about...

Shaolei Ren (15:18)

Yeah.

Gaël Duez (15:30)

The so-called water cooling

Mohammad Atiqul Islam (15:33)

Can take this question. So when we talk about water cooling, basically for the server end, when you are cooling the server, you cannot use water because it conducts electricity. So you are using liquid cooling. I believe you wanted to say liquid cooling, not water cooling. But anyway, people do.

Gaël Duez (15:52)

Yeah, I wanted to say it but I wanted to sound dumb enough so that you will actually lecture me and then you never know if it can be useful for everyone listening.

Mohammad Atiqul Islam (16:01)

So, yeah, so when we say we are using water in data center cooling, people sometimes misunderstand that they think, okay, this is actually the liquid cooling that people talk about when we talk about GPU cooling. So liquid cooling, first of all, again, it cannot be water when you are touching the electronic components. And there are different ways of cooling. One is called immersion cooling where you are putting the whole… Electronic components the GPUs and the CPUs everything is immersed inside the liquid which is taking the heat away from those components But there are also other kinds of coolings called our cool plate cooling. So basically instead of immersing the system into this liquid what they have is they have mechanical components which carry the heat using liquid from those chip, GPU chip, CPU chip away from the system and they are typically much more efficient and most importantly they can carry more heat than air. In the past we will use air flow to take the heat away from this component but now if you use any kind of liquid cooling the amount of heat that you can carry away from the system is higher so that eventually the result is you can put more servers or more GPUs in a smaller area. So basically these days the power footprint of server rack is now 30 to 40 kilowatts it used to be like 7 to 10 kilowatts and there are actually some limits how much you can get away like take away using air it's about 15 kilowatt for a rack because you need to have enough airflow and then air has a limited capability of like how much heat it can carry.

So you are carrying the heat from the servers away. Now this is where you can use water when you are taking it away. So you have a liquid that doesn't carry electricity. And then that hot liquid can be now cooled using water while it's not touching the water. So basically there is a heat exchanger which transfers the heat that those liquids is getting from the microchip and the GPUs and then it releases to the water and then the water actually carries it away from the server room to outside. Typically there is actually another loop outside that this loop that carries the water from the server room to outside. This is also closed loop. The same water runs through the system. You do not expose this water. And then there is another heat exchanger outside for evaporative cooling, which now transfers the hot water heat to the cold water heat and that outside loop sends the water into the cooling towers and that outside loop is actually losing water. And this inside loop for getting the heat from inside the data center to outside, this is actually very commonly used. It's not just data center. Even my school, we have a centralized system where we are generating the heat and all those different buildings, they get hot water from the central system. And then they are transferring the hot water heat into, let's say, our heating system.

Gaël Duez (19:18)

Okay, so three different circuits. The first one will be the liquid cooling one, which is a closed loop. The second one, which is another closed loop, which is the heat exchanger between this liquid cooling and the evaporation tower where actually the water consumption, because it's evaporated, happens. Am I getting it right? Okay.

Mohammad Atiqul Islam (19:41)

Yes. And in the third loop, you have the most, you know, the control or the option. So basically, instead of use, so again, the second loop transfers the heat to outside the third loop, right? Now it doesn't have to transfer the heat to outside water loop, it can release the heat into the air. So then it becomes air cooling. You can use mechanical chillers to take, extract the heat from those, like the closed water loop, and then it becomes mechanical. So this is where you have the control and typically ⁓ evaporation based cooling has strong tie with the weather condition that you have. So basically outside temperature, the humidity that plays a big role. Sometimes the weather condition is such that you cannot use even even use evaporative cooling. And typically in those kind of system, you have to use hybrid cooling. So certain months of the year, you will have to use a different way of cooling.

Gaël Duez (20:09)

Okay got it.

Mohammad Atiqul Islam (20:38)

And typically that happens when actually it's outside this two code.

Gaël Duez (20:41)

Too cold, not too humid.

Mohammad Atiqul Islam (20:42)

Yes, if it's too cold and because when it's colder, your air's water carrying capacity goes down. So the same amount of water in the air will actually cause humidity to go up when your temperature goes down. That means when you want to use evaporative cooling, you cannot evaporate as much when the outside actually is colder.

Gaël Duez (21:03)

Interesting, a bit counter-intuitive, but it makes total sense now that you explained it. Okay.

Mohammad Atiqul Islam (21:04)

So there is like a evaporative. But the good thing is when you have, like you cannot use the evaporator cooling because it's too cold outside, free air cooling would be the best option because now the air is cold, right? So you don't have to. So this is actually a typical combination that when it gets too cold, you wanna use free air cooling and then when it's like a regular temperature, then you use evaporator.

Gaël Duez (21:30)

Interesting this mix of technology. And now let's move maybe to the second big topic that we need to discuss and how bad is it doctors because we discuss techniques here and having a better understanding of what is at stake, the different scope, the different components that consume or withdraw the water but is it bad? And maybe my first question would be between the direct and indirect water consumption, what has the most weight? Because I used to remember, it was, I think, three years ago, one of the very first episodes of Green IO, that I was shocked to learn that, once again, in Europe, so I'm not saying that it's case everywhere around the world, that actually the water consumption from data centers was mostly coming from energy production rather than direct water consumption. But it seems when I read some articles that A, it might be not the case anymore and B, it's definitely not the case everywhere around the world. So maybe can one of you explain first this difference and then what numbers are we talking about? Because is it such a big deal or not?

Shaolei Ren (22:51)

Yeah, I think in general you're right that the indirect water consumption is a more significant part compared to the direct water consumption. So if you look at the US, we're looking at one tech company, US average data center, and the ratio is about 87 of the total water is coming from the electricity generation part. And this ratio varies case by case. In certain places, the ⁓ direct water consumption ratio could be higher, but I would say in general the indirect water is more significant.

Gaël Duez (23:29)

Do you have any examples where this ratio is reversed and that is actually more the direct water consumption that prevails?

Shaolei Ren (23:38)

It's not necessarily reversed, but I think if I remember correctly from the 18th location that we studied, Arizona has a higher portion for direct water consumption due to the higher temperature. And I think that data center is also using evaporative cooling during much of the year.

Gaël Duez (24:02)

Got it. And so my

Mohammad Atiqul Islam (24:03)

And sometimes, you know, it's not only on the direct cooling side, but rather if you have more renewable generation with zero water footprint like California, they tend to have a little bit less indirect water consumption. So they could be more comparable. Again, not reversed, but lower than many other places.

Gaël Duez (24:24)

Okay, got it.

Shaolei Ren (24:25)

Yeah, probably we'll be talking about what is the problem or what is the issue with water usage? when we discuss that part, I think we will point out that ⁓ water is never a single number. So if you look at the overall number, it kind of misses a lot of information. we also need to look at when the water is used. So for some data center, are using, if they are using natural air, evaporated cooling with the water evaporation system. They don't use water 24 7, but they are using water during the moments when the other users are also using water most. For example, during the really hottest week of the year. look at the overall numbers, minimum doesn't seem to be any trouble at all. But if you look at the window data center is using That's probably the worst time of the year. And that's going to be pressuring public water infrastructure. in certain cases, been reported that certain data centers are even requiring more water than the entire county's usage during certain days.

Mohammad Atiqul Islam (25:28)

I want to add another thing because Shaolei is talking about when. So when in my opinion has two components. One is the rate of water consumption that is happening. That is one when basically for the same cooling depending on the time of the day or depending on the season you may consume more water for again cooling the same amount of heat or getting taking out. So this is one of the when. Another one would be if you look at the availability of the water. So in different time of the season, you will have different availability. So when you have a lot of rain, maybe you may care a little bit less about water than when it's like a dry season. So these are the two kinds of when that plays important role. And then I believe you can already connect this when also relates to where, right? Because at different places you have different kinds of stress. And the dynamics of when you are more water efficient versus when you are less water efficient also varies with location. So these are a lot of different things that are happening that we can look into and also we can actually utilize in our advantage.

Gaël Duez (26:39)

And just to get some proportions, so you studied 18 locations for data sensors in your paper, am I right?

Shaolei Ren (26:47)

Yes, including the US average by one tech company.

Gaël Duez (26:53)

Including the US average, thanks. And for these different places, do the data centers create significant water stress or is it something neglectable? Are we talking about, I don't know, 5%, 10%, 15 % and on top of it, everything that you just explained, which is it depends on the what, the where and the when. So I'm trying to understand. All these figures and with all these components and this complex explanation that you provide, and thanks a lot because there is no shortcut, I guess, with this question of the water stress. But yeah, is it serious? Because if we think in Olympic swimming pools, sometimes it sounds massive, sometimes it doesn't sound that big. So when is it an issue and what can we say to assess when it's an issue and when it's not?

Shaolei Ren (27:45)

I think overall at the state level or at the national level, the water usage by data centers is really modest, probably less than 1%. But for individual communities, unfortunately, we don't have the concrete numbers in our paper. And I still don't have the concrete numbers yet, because it's really hard to get the total public water supply for each individual communities. My guess is that ⁓ it still varies a lot depending on how large a community is. For certain smaller ones, the data center water usage could be really substantial. We've seen some reports, I mean, it's not verified yet, but some reports says certain data center could use more than one third of the local public water supply. And in this case, we're talking about direct water withdrawal by the data centers, and also the public water system. Public water system is supplying water to residents, to homes, and to businesses.

Gaël Duez (28:51)

And that one third might not occur during the entire year, but actually might occur at the moment where everyone needs water the most. Especially we're talking here about residential, but also a lot of about agriculture, I guess, no.

Shaolei Ren (29:07)

No, actually the public water supply is not for agriculture. because it's the water from the faucet, from tap water that we can drink directly. So agriculture uses a different type of water and that actually comes to a more complex issue because water not only has different scopes but also has different colors. The water that we're talking about here is...

Gaël Duez (29:08)

No? Okay.

Shaolei Ren (29:33)

By data center directly, direct water consumption that's mostly drinking water. for data center cooling, for generating electricity, generally that's water. It's the water from the rivers, from the lakes that humans can use. And for agriculture, they also use a lot of green water. That's the water in the soils to be used by plants, by crops. And that's… not usable by directly by humans. it's a yeah, that's more complicated. I guess we probably don't want to go to go into that part. But the agriculture is separated from the one third number that I mentioned. electricity water is not considered in that case. And we're just talking about the water which are directly by data center and also the overall communities, drinking water resource.

Gaël Duez (30:26)

But that's an important point because I heard quite a lot the tension, the local tension, because as you rightfully explained, it's a local community matters. It might be completely neglectable having a data center in your neighborhood in some areas where water is plentiful. that's not a topic to debate. in some communities that can create massive tension. And quite a lot what I hear are these two things. Because of the data centers, we don't have enough water to drink. Because of the data centers or agriculture is at stake. And actually what you're telling me is that maybe the second part of the narrative is not really relevant because agriculture and data centers, don't consume the same water. The first one would consume green water, the second one blue water, and they don't tape into the same aquifer or am I missing something here?

Shaolei Ren (31:23)

I think when we talk about droughts, the agricultural water, regardless of the water usage, it all matters. And of course, agriculture is really probably the largest water user in many places. But if you try to look at the pressure on your community's water infrastructure, agricultural water usage is probably not relevant at all. regarding your first comment that you've heard, some people say they have no drinking water if there's a data center. I think this is probably too exacerbated. I think we should be objective, the one third number that I said from my ⁓ report is I wouldn't say that's representative of many places. First, most places don't have data centers. Secondly, if they do, I think 130 is still on the higher end in most cases. If you're talking about ⁓ using up all the water, I can't see any such cases in reality There was a report earlier this year saying that in Newton County, Georgia in the US, ⁓ one data center is asking for the amount of water that is more than the county's entire usage. But that's just a request. It doesn't mean the data center is already being built and they are taking away the water. So we still have time to build more infrastructure to plan for the supplies. So that people are having no water just because of data center. That's probably not something that we want to deliver to the public. And Mohammad, maybe you can chime in.

Mohammad Atiqul Islam (33:09)

Yeah, basically, yeah, I think I totally agree with Shaolei that we absolutely should care about water consumption of data centers, but yet we do not see that data centers are taking away water from people's faucet yet. There is a stress. That's why the utilities are very careful when they're approving new data centers. I know many data center projects in Dallas that they're waiting for utility approval both for their energy and for their water supply. That means actually the local communities are aware of the problem and right now the infrastructure everywhere is not adequate to actually sustain the growth. So at this moment, the data center water consumption is not causing the problem, but ⁓ if we do not look into it, we are not careful about it, it might create problem down the road.

One of the problems that I know in my local ⁓ North Texas area that we do not have any natural lake around us. So we built several lakes using dams and they had to now build another one because again, it's not just for data center actually, North Texas is growing really fast in population as well. But this now kind of like puts ⁓ data center in the mix as well that now also Dallas is actually one of the hotspot for new data centers as well. So that's why it now becomes ⁓ a competition between a public utility and then the data center that, okay, who should we prioritize more, right? And this has become a topic of discussion. But again, we do not want to say that they are taking away water from people's.

Gaël Duez (34:54)

That's why I love talking with academics because you tend to have a much more balanced approach and also providing multiple aspects and multiple parameters for a single equation, which is very satisfying from an intellectual perspective, not always necessarily very easy to leverage for decision making, whether you're someone working in a data center or a politician for public service for instance, and just rephrasing all the components because I want to make sure that I got it right before we move to what we should do and we already started to tape into this questions, but let's try to rephrase it. But before that, the multiple angles of this discussion and I'm… I'm parroting you here. Water consumption of data centers is not the number one issue when it comes to water consumption at the moment in the US. might not be the case in some places. So where matters a lot? When matters a lot? Either from a seasonal or daily perspective, but also waters stress is happening. And all these components plus the trade-off with energy versus water, plus the different technical choices, you blend all this and then you've got a good problem to solve, I would say, and to optimize. Am I right to putting these different parameters in the equation?

Shaolei Ren (36:29)

Yeah, I think you're absolutely right. So we need to ⁓ understand the problem, but we're not fearful of the problem.

Gaël Duez (36:29)

Okay, that's an interesting one. And so what do we need to understand the problem and what should we do? And that maybe would be the last questions of the show and how should we act? And I know maybe this is a moment you want to discuss about the lack of transparency, at least the perceived lack of transparency is what I've heard. I don't live in the U.S. from main operators and also the technical solution and basically also what us as software practitioner could do. So there are multiple approaches to a very simple answer, which is what are the solutions? And of course, I'm sure that you've got plenty of things to

Shaolei Ren (37:14)

Yeah, think we need to understand the problem and there are multiple scopes of water usage. They have different impacts. They also require different solutions. But if you don't really have a clear understanding of different scopes and different type of water, then you might not even address the right problem. For example, think the direct water consumption and even water withdrawal is more relevant to the local public water infrastructure planning and the indirect water consumption is more relevant to the policy making because when we try to allocate the water resources among different sectors, the indirect water consumption will play into a significant role. But at this point, there's no standardized way of discussing the water usage. Some people prefer to talk about the direct water usage only, which is fine. But we also see different or sometimes even misleading comparisons of different numbers across different scopes. And that could mislead the discussion and give not accurate information.

Gaël Duez (38:27)

And Shaolei this is sort of the elephant in the room, because what you mentioned is you raised the alarm about the Gemini report from Google is that they didn't use the right number and that the result was the opposite of what they claimed. But maybe I would love to hear it directly from you because I want to be sure I didn't misunderstand what you wrote.

Shaolei Ren (38:51)

Yeah, I mean, we were first we were not planning to talk too much about that particular case because we don't want to make them look bad. That's not our intention. But we if we talk about the facts, the fact is one paper released in August this year, use their direct water consumption, average direct water consumption to compare with our highest total water consumption across 18 locations. And if you read our paper, we actually present both direct and indirect, and also the total separately. So we have three numbers for each location. And if you look at the US average, look at the onsite water only versus their presented onsite number. It's not orders of magnitude difference. Plus, our estimate was done in 2023, but the paper's number is for May 2025. They also acknowledged that their number was 33 times higher back in May 2024. So if you look at their May 2024 number, that number was even higher than our onsite average number back in 2023. So I'm not arguing that their system has improved the efficiency during the past 12 months or one year. That's definitely a remarkable improvement. However, doing this direct comparison is not really meaningful because it sounds like the estimates, including our number were wrong. Well, I think their number doesn't support that claim or that suggestion. the number that we presented doesn't support the claim that their system is more efficient than other systems that we study because we're looking at different things, different models. We're looking at GPT-3 175 billion parameter model, but they are looking at very different system and they are not even talking about the actual output lens. So I don't think the direct comparison is really meaningful give some constructive messages.

Gaël Duez (41:19)

And if I may, know, the main takeaway that I take from the discussion you had is peer reviewed papers. I mean, I don't know. I'm not in anyone's mind at Google. I don't know what happened, but I'm pretty sure that usually when you enter into a peer-reviewed process. This is exactly the sort of discussion you have. And even if you stick with your analysis, this is the moment usually as far as I remember and as far as I've experienced at home, because my wife is a researcher, this is a moment where you clarify your hypothesis. you say, well, maybe we picked this number for this reason and that reason. And I think the main issue at the moment here is that we've got part of the main reports, one of the most important sources of information that are going away from the scientific method of publishing with open data, peer reviewed, etc. And for me, this is really an alert. I don't question at all the fact that most of the people writing this paper, they want to do good things in the world and at least in the tech industry. But it's just that 500 years ago, we built the scientific method to avoid this kind of mess.

And that should be a good thing to stick to it until we've proven that there is a better method. And as far as I know, and please correct me if I'm wrong, because after all, you're the scientists, I'm not. There is not. And this is really the issue with not having enough open data, enough transparency and enough peer reviewed paper, even coming from the best intentioned people. The facts are that this sort of debate should happen before it comes public because then the numbers are strung away and it's like running after a while. You cannot really get back the number and say, by the way, they were a mistake, you know, and actually that should have not been analyzed that way. It's too late. It's in the public debate and it will mislead a lot of people, including decision-makers. But that's, I would say, the transparency angle and it's me rambling. That's not the point of this podcast.

So actually, Mohammad, I know that you also wanted to talk a lot about the technical approach to solve this water stress, both from withdrawal and consumption and all the trade-offs. We've listed the trade-off between energy and water. But actually, there might also be even a trade-off between which kind of energy do we put into data centers, because obviously the energy coming from solar or wind might need much less water consumption than the one coming from ⁓ fossil fuel or nuclear or even hydro. So could you tell us what we can do, what we should do to reduce the water stress coming from data center?

Mohammad Atiqul Islam (44:08)

So there are different ways we can look into what kind of solutions we have. One of the things that we can do is how we can make data center water efficient. So basically we want to run the data center. At the same time, we want to consume as less water as possible. If you look at direct water consumption, now then you can go into water cooling systems that use less water. basically things like dry cooling, based, outside air based cooling, things like that. But again, as I mentioned before, they are more energy intensive. So what may end up happening is that you are reducing your direct water consumption, but because you are consuming more electricity from the grid. Now, if the grid has worse water efficiency, your overall water consumption where you both considered indirect actually may go up. So that means if you go for a solution where the direct water consumption is going down but you are trading it off for more energy consumption from the grid, you need to really look at the grid's water footprint. So that now ties your decision to which location you are at, what kind of energy sources that you have in the power grid.

So this is the way that you can improve the direct water cooling and overall water consumption. Other things that we can look into is water harvesting. Basically, there are challenges in water harvesting because of the intermittent nature of rain and other sources that you are going to use. But if you tie this water consumption or the water harvesting with some techniques that

the way that you control your data center. So basically you can control how much energy you consume and that directly reflects on how much water you consume in the data center. So now you can look at software site solution as well. If we are careful about what kind of work that we're doing in the data center, then we can actually now complement our water harvesting, which is intermittent. So the source may go up and down. So that means we can do the smart scheduling of a workload or smart planning of the workload that we're doing to improve the water efficiency of the data center. And another thing actually, this is something is completely my own opinion. I talk about this whenever I get a chance. I see a future where we can have zero water data center and the future would be you have a data center and you have ⁓ renewable energy source right next to it. So maybe, maybe a wind turbine or a combination of wind turbine and solar panels so that you are generating the energy that you need right next to you. You are not getting anything from the grid. So zero water from the grid. Solar is considered almost zero water. You do need some water to clean the solar panels, but that's very minimal. And now you can also combine it with dry cooling because you are getting the energy from water free source. Now you have dry cooling and the solar panel right next to each other, right next to the data center, you now will have a zero water data center. And not only zero water, it actually will be also zero carbon data center. So this is something we can absolutely do.

Gaël Duez (47:42)

During the use phase, because if we put a full life cycle analysis with all the minerals needed both for the IT equipment and the solar panels, there will definitely be a few tons of water per kilo of a solar panel. But you mean during the use phase, a zero water data center should be something okay.

Mohammad Atiqul Islam (48:06)

And I'd

Shaolei Ren (48:06)

I fully agree.

Mohammad Atiqul Islam (48:08)

Like to add one thing here as well. So basically ⁓ there is a interest of utilizing nuclear power plants because when you are putting data centers in many different places, one of the constant is not that we do not have enough energy to power those data centers. The US power grid infrastructure is not strong enough to carry the… energy or electricity generated at different locations of those power plants to the data center. So the infrastructure limitation is also actually also stopping new data center growth. So that's why people are thinking what if we put a nuclear power plant next to a data center. Now the data center gets all its power from the nuclear power plant. doesn't run through the weak or not weak. Constraint power grid infrastructure anymore and now we can have more data center where we utilize the data center for ⁓ larger computation. Now this works in favor of solving the energy crisis that we do not have enough energy to build new data center, but nuclear problems typically consume a lot of water. So this solution doesn't work if you look at both water and energy crisis. So this is something I want to bring it up because there is a lot of interest. Among many companies, we are now having new contracts with existing power companies to pair a data center with a nuclear power plant. In some places it will absolutely work, especially if there is no water stress in that particular area and you have a nuclear power plant. Now, if you build a data center next to it and the data center is just run by all the nuclear power, then it should be great solution because nuclear also have ⁓ zero carbon emissions.

Gaël Duez (49:57)

This is something that in France we experienced first hand that because of climate change, that were supposed to be water stress free are not anymore and this is why two years ago we experienced the very first time being net importer of electricity and France tends to be a rather big exporter of electricity for the entire Europe because the drought was so crazy that a lot of nuclear plants had to scale down activity because of not enough water being available. So thanks a lot for bringing the topic that nuclear plants might be a solution for decarbonization, debatable under the full life cycle, but yet definitely better than power plant, but definitely also an issue regarding water consumption from the nuclear plant.

Mohammad Atiqul Islam (50:52)

Is the interest in nuclear power plant in the US especially avoid carbon emission, but actually it's rather that we do not have enough energy that we can supply to the new data centers that the IT companies want to build. Again, there is actually another dimension to it. It's not like we do not have enough generation capacity. In most cases, we even have enough generation capacity. The problem is the infrastructure that you're going to use to carry that energy from those distant power plants to the data center that infrastructure is actually not capable in most places in the US. So that's why is an interest of utilizing nuclear power plant next to

Gaël Duez (51:36)

Yes, thanks Mohammed. Shoalei you wanted to say something.

Shaolei Ren (51:39)

Yeah, I think I just want to add that building these transmission lines to carry the power from the power plants to the end users like data centers is really an expensive investment. And that's why we're seeing a growing interest in co-locating data centers with power plants like nuclear power plants, and even natural gas turbines in certain cases. Additionally, would like to bring up the point that ⁓ drought is more like a slow earthquake. comes and goes. It can hit any places now. I think in my point of view, no matter how much water you use, you should view this water supply as a supply chain resilience. So let's say we have a small data center using water evaporative cooling in an area, but the water pipe could be leaked or could be due to drought. There is less water pressure. Then essentially the operation will be interrupted by those events. So if you view this as a resilience problem, think essentially we're living under a water stressed world. Maybe not today, but certain time of the year.

Gaël Duez (52:53)

Next of the sense. Okay, that was an in-depth discussion on all the aspects. I'm pretty sure we actually barely scratched the surface on everything that you know, but at least for the professional audience listening to the Green IO podcast, I think it will be a pretty packed episode regarding all the dimensions of water consumption, water withdrawal, and all the different colors of water. So maybe my last question, and Mohammad you already shared a very optimistic view that maybe a future of zero operational water data center is possible, but would you care to share a positive piece of news regarding sustainability and maybe even IT sustainability?

Mohammad Atiqul Islam (53:38)

I'm really excited to see these days is that data center, whenever we talk about data center, energy consumption or the energy need comes up. But in the last two years or so, I'm seeing every article, they're also mentioning water usage. So that means this consumption of data center has become issue that people care about and this is where I actually want to share one of my own opinion and you are free to get rid of it from the anyway... I believe the responsibility falls more on us to be environmentally sustainable and to pressure or put pressure on the lawmakers to

Gaël Duez (54:11)

I won't.

Mohammad Atiqul Islam (54:27)

create laws that will put pressure back on the companies who run data centers. Now the data center companies or these IT companies who have running these data centers, they are doing everything within the law and their job is to maximize their profit. If they care also about environment, that is good and many companies actually do, but we should not just expect that they will be the one who lead this effort, but rather we should be the one that leads the effort. And we ask our lawmakers to create laws that will preserve the rights or that will preserve the environment for us. So we should not only ask the companies, IT companies who are already actually have to make big decisions, like they're spending hundreds of millions of dollar investment they're talking about, right? And this large amount of investment, they need to be getting more about the benefit or the profit they can get out of it, right? So I believe we should ask our lawmakers, we should encourage our lawmakers through asking them or letting them know what are the problems that we are facing. Also as a researcher, I believe we are doing our part, but the lawmakers can ask for more transparency from those companies so that we have a better picture and we can look into what are the different ways we can improve the system. And that is something we also want to inform our lawmakers so that they also know that these are the ways actually available to the IT companies. One of the research areas I'm looking into is how can we, it's not about transparency to the public. Like after a year, I publish a report of my yearly water consumption. What if we were, we are letting the people know your hourly water consumption? And then, as people, now have information about our water footprint. Now we have the means to actually make some changes in our behavior. This is something that is not available right now. I understand that it's a very challenging problem to let someone know that if you are using a certain product of a certain company, you are using this. Actually, this is a very technically hard problem to solve as well. But I see if our lawmakers make the effort to push the IT companies, they will the resources to build those systems and let us see what is happening so that we can also contribute to this solution.

Gaël Duez (56:53)

Thanks a lot Mohammad for this. Shaolei, you want to share something? Yeah.

Shaolei Ren (56:55)

Excellent point. I mean, I agree most parts, but whether we should use the word pressure the lawmakers, I don't really want to pressure them. I just want to do my part. My part is to present the numbers, present the facts. And I want to understand the problem.

Gaël Duez (57:13)

That's funny because this is one of the big discussions happening within labs at the moment, whether scientists should stick to facts and just present neutral facts to policymakers and let them do what they have to do. And another proportion of the scientific world saying, actually, we need to go out from the labs and apply some pressure and be part of the general public discussion because there are so many lobbyists and so many public relations budget overwhelming everything that we say that we need to sort of lose our neutral stance that used to be the way scientists do and I don't have any opinion on it I just can see that the debate is happening pretty much everywhere.

Shaolei Ren (57:57)

Well, yeah, that's it. Yeah, that's a hard decision. I think I'm okay with going out of the lab and doing this outreach or generally public engagement activities. I'm fine with that. But at the same time, it's really hard to stay absolutely neutral, like 50%, 50%. But if you're staying around the… a reasonable range, I think around the neutral position. Yeah, I mean, it's hard to say. We can reach out. We can do this public engagement. We can do the outreach. At the same time, my bottom line is all these activities should be based on facts.

Gaël Duez (58:51)

Hmm. I think that's a beautiful closing word. Or I think that's a beautiful closing sentence. Like all this activity should be based on facts. Thank you so much, both of you, for joining, for taking the time to explain, to enter in great details all the different components of this water stress issue with data centers and remaining, as far as I can see, absolutely neutral. So a big thanks, making your time available for all the podcast listeners. And I hope that we will continue this discussion. I hope that maybe… one or two of you will have a chance to attend Green IO New York and explain a bit of this on stage But once again, thanks a lot for joining and taking the time.

Shaolei Ren (59:38)

Thanks.

Mohammad Atiqul Islam (59:39)

Thank you, Gael. It's nice talking to you.

Gaël Duez (59:42)

Thank you for listening to this Green IO episode. Because accessible and transparent information is in the DNA of Green IO, all the references mentioned in this episode, as well as the full transcript, are in the show notes. You can find these notes on your favorite podcast platform and, of course, on the website greenio.tech. If you enjoyed this interview, please take 30 seconds to give us five stars on Apple Podcasts or Spotify or a like on YouTube.

Sharing this episode on social media or directly with colleagues working in the data center industry seems also a good idea to make sure we'll have enough water to drink and water our crops. ⁓ now you know that one of those two assumptions is not true. In our next episode, we will welcome Iona Tsyrulneva, a respected Singaporean expert on AI who will try to answer the billion tons of GHG question, can we be optimistic about AI sustainability? Stay tuned.

By the way, Green IO is a podcast and much more. So visit greenio.tech to subscribe to our free monthly newsletter and check the conferences we organize across the globe. London was a blast last week and the next one is in Paris on December 9th to 11th. As a Green IO listener, you can get a free ticket to any Green IO conferences using the voucher GREENIOVIP. Just make sure to have one before the 30 free tickets per conference are all gone. I'm looking forward to meeting you there to help you, fellow responsible technologists, build a greener digital world.