[00:00:00] Speaker A: Hello and welcome to the Consulting Specifying Engineer podcast. I'm your host, Amara Rosgas and today we're talking about data centers with Jeff Staub. This CSE podcast is now two years old, so I'm excited to be here. Thanks so much for joining me today, Jeff.
[00:00:19] Speaker B: Thanks, Amara. I appreciate you having me on to talk a little bit about data centers. It's an exciting time to to talk about it, so I hope that your listeners from Consulting Specifying Engineering magazine get something out of this so it's. Hopefully we can have a good dialogue and share some great information.
[00:00:39] Speaker A: Yeah, yeah. And there is a lot of discussion about heat recovery in data centers, so I thought I'd talk to an expert on this topic. So let me officially introduce you to Jeff Staub. He is the director of OEM sales at Danfoss Climate Solutions.
He has nearly 30 years of experience in engineering and H Vac applications.
Let's start at the very basics. Jeff. What are some of the current trends for heat recovery in the global data center market?
[00:01:12] Speaker B: Sure, absolutely.
Again, super exciting times. And I think the trend holistically is growth.
As everything that we're seeing within data centers, I think the use of AI is tremendously growing the efforts within data centers. And of course as the chip density continues to increase, there's the increased need for heat rejection or heat removal or cooling those chips. And traditionally in many H vac systems you're just rejecting that heat to the outside air and it's complete waste, waste heat. So there's a tremendous opportunity for to recover that heat, to use that heat for other areas within the building or off site building. So we're seeing an increased usage of heat recovery in data centers because of the heat density that's being used.
For example, I think if you look at some of the future trends, I think the global trends are projecting increased usages of 12, 14% of growth of data centers.
I believe the numbers around 135 facilities currently use heat reuse in their facility. That's projected to grow to above 1700 facilities by 2030. So tremendous growth coming in the next couple years. Now that growth is also a little bit diverse. When we look around the world, we know that Europe will have a much, much larger growth than what we see in the United States.
But still with the data centers, whether it's hyperscalers or co locators, we're going to continue to see this increase in heat and chip density. So the more opportunities for using that, the reason really for increased usage in Europe is because there's a lot of district energy heating grids already in place and that's a very practical application for heat use application. So you'll see this very easy adaption within Europe at a very high rate, but we will continue to see good growth within the US that is double digit growth along with the growth of data centers.
[00:03:38] Speaker A: Okay, okay. And you're talking about waste heat here and you mentioned this, but where is the heat being used in a data center or what's happening with that?
[00:03:48] Speaker B: Yeah, so if you look at a data center really when you look at the fundamentally it's, it's about the utilization of a chip and when you're using the chips into a server that, that's built into a rack just by utilization of the microchips, you have a heat coming off of those chips and to keep the operation of the chips in a, a good operating level. So there, there's no downtime within that you need to remove that heat and how you remove that heat, there's different ways that you can remove that heat and whether you're blowing air onto it. More and more we can see a very rapid adoption of liquid cooled technologies where you're pushing a liquid over the chipset and as that liquid cools the chip, it picks up the heat and then that circulates into a cdu, a cooling distribution unit where you now have hot water. And then you have to do something with that hot water. Maybe you're moving that to a chiller, maybe you're using that into a water heater and then that chiller is rejecting the heat to the outside air, if it's an air cool chiller, or rejecting it to another water, water flow loop into a cooling tower where you then cool the water. So with all of the racks, the chip density, you have huge amounts of heat. And then the question is, what are you going to do with that heat? Are you just going to reject that into the air or are you going to integrate it into other systems to really, really implement a full system that's very efficient to use Other examples and whether that's water heating, multi use family heating, there's a lot of different areas where you can use that heat recovery.
[00:05:40] Speaker A: So this sounds really complex and it sounds like there are a lot of challenges.
Can you tell me a little bit more, what are the integration challenges or even the overall challenges that come up when incorporating a heat recovery system into a data center? And what do engineers do to plan for or, or work around these challenges?
[00:06:01] Speaker B: Yeah, you're absolutely right. There's multiple challenges when we look at this. I think one main challenge of course, from an engineer standpoint when you're looking at and incorporating heat recovery into the system, is really the speed of adoption overall, the growth of AI and the growth overall of data centers. Data centers are being built at a very rapid pace right now. And to add one more step in the process of designing heat recovery systems, the data centers are coming on board very quickly. And even at a rate that you can see when you're looking at hyperscalers or CO locators, they're deciding where to build these data centers in proximity to the grid. There's concerns that the grid will even be able to sustain the growth of data centers. We see data center companies that are pairing with utility companies, with nuclear substations. So it's, you have these guys just looking at the challenges of how do I build a data center and how do I power that data center to add on heat reuse. And what I'm going to do with that is another step in that design process. And many times the speed at which we're implementing data centers isn't going to allow for it. The other area is just what is that design of the data center? I think with the adoption and moving towards liquid core cool data centers, we're allowing to get to slightly higher temperatures. And when we talk about challenges and it's really when you talk about what to do with heat, you're talking about is that air heat that I'm using, is it water heat and how do I transfer that from one facility to another?
And many times an easy way to transfer heat is through a fluid such as water air, you have slightly more challenges. You can use air, air to air recovery systems where within the same building, if you have a data center in the winter, you can use that heat rejection and transfer that to heat the offices in that building.
But it's not a really good media to be transferring heat to remote locations where there you're typically going to be transferring heat into water. And when we talk we actually, when we're looking at heating water, there's different uses of heating water. When we use hot water heaters in our house, typically that's at 125 or 130 degrees.
If you're getting into sanitary water, you need water at 150, 160 degrees.
If you're looking at industrial applications, you need water up to 200 degrees. And potentially if you're getting into power generation or steam, you need high pressure steam that's coming across at 250, 350 degrees steam. So when we talk about overall heat rejection or heat reuse, sometimes it's just well, what media am I using and what temperature? So if I use, let's say air to water heat exchangers coming off of chips, maybe I can get that water up to 100 degrees. So it's really low temperature water.
So what are you going to do with that? With liquid cooled applications coming off of a CDU or a chiller, I can get that water, water up to 130, 140 degrees. So I can use sanitary water applications coming off that and pumping the water throughout. Let's say multi purpose, multi use housing sites.
So if for certain, how do I get from the heat off of the chip to not just warm water but to a usable hot water, do I need to add boost heat pumps to really heat up the water to additional levels? You of course, I think the challenge of integration of who's going to pay for the capex when. When you look at most data center companies, they're in the business of course of looking at chip capacity, but they're in the business of providing, let's say digital services, computing power, cloud services. So they're really interested in the operational of the chips.
Is it in their interest to create and pay capital expenditures for all of this heat reuse energy recovery systems? Maybe they use hot water in their building. But if it's going off site, where you're going to be heating and providing higher temperature water or steam into district heating networks, why would they want to pay that? So you of course have your integration of who pays for this, what are the CapEx versus the operating costs? And if you're sending this heat into, let's say the community, do you need to have service level agreements with utility companies, with the city or whoever's going to be using that heat? And then of course you have, in general you have conflicts between that cooling and the recovery systems. Because when you're talking to many data center companies, you're going to be talking about chip uptime. What's the resiliency of the chip set? I need to have operational 100% of the time, but there's also a varying usage on these chips. What's that load that's going into the chip and that's creating a variable load of heat that's coming off of the chip.
So when you have varying loads, but the data center companies, they're concerned with uptime. I need to ensure my chips are being cooled.
If you have a variable load or demand on those chips. That means the amount of heat that's coming off of that is also variable. But now if you're integrating that heat recovery into the community, into a vertical farming application, a medical system, they're going to need a specific amount of heat to provide a very stable demand of heat reuse. Well, the chip guys, they're worried about cooling their chips, not providing a stable heat output. So maybe you need to have a thermal buffer in between the two because the chip guys want to make sure they have absolute 100 uptime for the chips, for all of the telecommunications, the cloud based systems, for really their users and consumers to have appropriate data processing from the chips.
So I think those are, those are some of the four main areas of concerns and challenges that engineers are going to have with implementing heat recovery into data centers.
[00:12:53] Speaker A: So in other words, insanely complex. We have to think about literally everything.
[00:12:58] Speaker B: Yeah, yeah, absolutely, but it's complex. But I would also say okay, if you're engineering in a manufacturing location, certainly you have to look at CapEx versus OpEx. So it's what, how much you spending on a piece of equipment versus what's your revenue. So I think that's pretty common. I think when you look at, you know, the speed of implementation, I think the speed overall of technology adoption is coming at a much faster rate. So this is just the rate of change in, in the technology sector. So I think it's very different from very stable manufacturing. We need new design cycles, we have to change. So I think the speed of implementation, it's becoming more of a commonplace. I think evaluating the financials and it's, and I think really in mechanical engineers we're used to, to designing systems for what's my low density, you know, how is my seasonality and how do I create efficiency across the seasonality of the heat load for summer performance in the northern part of the US in the southern parts of the US So it's not anything that's impossible. I think you just now have to take that cooling system, couple that with an advanced heat recovery system and, and then really look at who's going to be paying for use and how can you integrate that use with partners such as utilities or municipalities.
[00:14:29] Speaker A: Okay, yeah, that's a really, really good breakdown.
I'm going to shift gears here a little bit on you, Jeff.
Are there any industry initiatives around implementing a specific technology or architecture for these data centers?
[00:14:46] Speaker B: Yeah, I think overall, I think some of the things that we look at, I think from data centers, I think They've looked at what's called PUE power usage effectiveness and it used to be very high numbers. And really that's just a ratio of the total facility power to the IT equipment. And you really want to drive that to a lower number, getting down below one and a half, down towards 1.3.
So we went from a power usage effectiveness and then many. When you looked at how do you cool these chips, there was the implementation of water pumps, cooling towers, and very soon recognized that, well, there's a water usage effectiveness as well. And how do I minimize water usage effectiveness?
So we can see the adoption of, let's say going from just air side cooling, where you're blowing air over the chips, to a liquid cooled system. So I think when we talk about an industry adoption of technologies, we can see a higher adoption rate of liquid cooled technologies. And that's interesting, especially from a heat recovery standpoint where if you are heating water as a byproduct, you can pump water around, you can do this very effectively with variable frequency drives or on the pumps. And now you can pump this to an area where you can use this.
I think other areas that I would say are going to be more and more standard are, let's say, heat recovery skids, heat recovery racks. We can see that it was originally thought that data centers, it's variable usage, it's how many megawatts of chip power am I having versus cloud storage versus AI computing power?
And it was really thought, I think in the early cycles of designing data centers that everyone had to be a very custom data center, that this one was going to handle so much power. But what we can see, and I believe to meet some of the speeds of implementation, is that you're seeing more and more just modules. And whether that's down at the server level, whether that's the cdu, if you need to increase your capacity, you add another CDU, but they're the same CDUs, you just add two, you add four, you add six of those. And then when the CDUs, that cooling distribution unit is taking that heat away from the chips, you're then pumping that into, let's say the chiller. And now you have modularized chillers, same chillers on a data center, but you might have 20, 30, 40 chillers. The same with heat recovery units. Whereas you can create a plate and frame heat exchanger or braze plate heat exchanger along with a pump and a variable frequency drive. And rather than making 100 different units, you'll have modularized skids where if the Capacity increases, just drop in these skids and you can increase the capacity of that heat usage. So I think one of the things where is early on in the designs with the airflow, how do you do an airflow design to cool chips? Is it a plenum? You know, how are your air walls, how are your computer room air handling units and, and your fan set up? I think they were very, very customized. And I think moving to liquid cooled applications, we're getting into much more of a modularized approach. And I think that's going to allow for a bit of ease of application for engineers to work with equipment that they're familiar with with liquid cooled ch getting into these modular, modularized skids of heat recovery equipment.
[00:18:38] Speaker A: Okay, okay, then tell me what types of heat recovery systems are available and how can mechanical engineers gauge which is best for their particular project?
[00:18:50] Speaker B: Yeah, I think if you're looking at, going into, if there's interest from a data center company that exists today, I think really you'll look at a retrofit from an air to air heat recovery system.
Many of the systems that exist today have cray units in them, computer room air handling units. So there, there's only really air to air side if you're going to use heat recovery and that's going to allow you to heat other rooms in the building. So if you have a data center that's in, in the northern climates or in the winter, you can use air to air, you can use air to water recovery units. So in air to watery, you're going to pull that air across a coil that has water being pumped through it. And that water that's typically a low temperature water that you're going to have typically 95 degrees, maybe you can use that in the building for sanitary water, for general water use.
But really when we talk about the use and the implementation of heat recovery, it's more on the liquid to liquid heat recovery. And this is where when you really look at driving down your pue to the lowest level, most new data centers will all be liquid to liquid heat recovery. And that liquid to liquid, you're going to be doing chip direct to a liquid and that liquid is then going to either heat other fluids or be cooled by other fluids. And it could be cooled by a direct vapor compression chiller. Or you're going to pump water direct out for higher water temperatures to the sanitary water in the building.
Or you can pump that further into the grid so you can actually pump it outside of the building. So if you look at, let's Say a multi use residential building per se, you could have radiators that are pumped up through the building so you can directly take the water from the hot water coming off of your chipset and pumping that further into a building to provide hot air. The radiator is then hot water transferring to a hot air to heat a building.
I think you could use that in a vertical farming application where you take this hot water and place a data center next to an indoor grow room or grow facility where you need climate controlled greenhouses where you can heat this greenhouse in the winter months, essentially getting this heating year round. So in a liquid system this is really where, where we see the future of going. And then the other type of heat recovery system is using that liquid to liquid heat recovery, but also adding a secondary heat pump to it.
So if you add secondary heat pumps to it and not now, you're coming off the chipset with a warmer water of maybe 130 degrees, 140 degrees, you can actually add a heat pump to it and boost that temperature to higher temperatures. And I think one of the interesting areas, because there's such a need for the power consumption, how are we getting power to these data centers?
And when we look at the integration where data centers being built very, very close to power generation facilities, you can now put in a heat recovery system with the heat pump and create steam that goes back to this power generation facility and create a co generation of power and you can use some of the steam back into your power generation. So from your power generation cycle and whether you're, you're burning gas, whether it's a nuclear stat, you're creating this high pressure steam to turn turbines and if you can refeed some of that steam off of a heat pump driven heat pump, that's heat from the data center, you really have this 360 degree circularity of reusing some of that energy to generate some of that energy. So it's, I think heat pump assisted is, is really going to be used much more in the future. But here again this is where you're going to have to weigh out some of your capital expenditures. And is it a hyperscaler that's generating, that's buying let's say power to generate, you could say data computing for their customers, but then also reusing the heat themselves.
I think it's a little bit more of a challenge. If you're looking at co location who are really generating that chip computing for a wide range of customers, how are they going to reuse that heat or are they going to sell that heat to the local municipality and who's going to pay for that? So of course you're still going to have to weigh your operating in your capital expenditures into it.
[00:24:00] Speaker A: Yeah, yeah. And thank you for mentioning Colo facilities because that's always a big topic.
[00:24:05] Speaker B: So that, yeah, I think that's when you look at some of the challenges. I think looking at the future, we expect the hyperscalers really to integrate heat recovery 75% of the time. So that's going to be at a much, much higher rate than a colocation facility. There's going to be a lot of colocation facilities but really it's because they have such a broader base of customers than a hyperscaler. Just using the data themselves, using the heat generate, pulling the let's say power themselves.
It's all within a single entity. Where I think it becomes a bit more of a challenge with a co location facility as to based on their client base, how many customers are paying for their chip computing, where does the heat go, how do they integrate this heat into a local facility, whether that's a college or adjacent commercial building?
I think it becomes a little more challenging just based on, let's say how they gain their revenue from their customers.
[00:25:11] Speaker A: Well, it sounds tough like you visit a lot of these different customers and like I said right before we started recording, I am a travel junkie myself.
So tell me what was a fun trip you took recently?
[00:25:26] Speaker B: I just had a fantastic one here recently. So it's, I'm avid outdoor, I like to, I like to hike, I like to be in nature and it's, I went about 15 years ago, did first hiking through the Rockies and went out to see Red Rocks. And I also like, I'm a big fan of music and venues and 15 years ago my wife and I, we went and we had hiked all through Red Rocks out in Morrison, Colorado just outside of Denver. Denver. And we just had an opportunity. One of our favorite bands was playing out there. So just last week we spent three days out hiking Rocky Mountain National Park. But we had had a great, great evening. It actually cooled off. So it's, we were worried about the hundred degree Denver weather but it uh, it cooled off to uh, a very nice 80 degrees at uh, night in Red Rocks and got to see our first concert in Red Rock. So it was a fantastic. Hopefully it's not once in a lifetime trip that more of those will come but that's, that's the excitement for the year. Able to do hiking and, and incorporate a music concert into. It was, uh, was a fantastic opportunity.
[00:26:44] Speaker A: Yeah. Yeah. Sounds like a double whammy. That sounds amazing.
[00:26:48] Speaker B: Yeah, it was a great trip. Great trip. So. And I got to spend it with my wife, so that's also a good one as well. So maybe not for some, but for me it's.
That was a good one. So.
Cool. Cool.
[00:27:01] Speaker A: Well, thank you, Jeff. It's been an absolute pleasure to chat today.
[00:27:05] Speaker B: Yeah, absolutely. I appreciate the, the time, Amara. So it's. Hopefully your, your listeners will get something out of this and yeah, hopefully as we move through this data center journey that we can adopt a higher acceleration of heat recovery because really, when we look at data centers overall, we can see the further development of energy reuse effectiveness. And really the ultimate goal for these data centers are, is to determine whether waste heat recovery really adds value and it's without compromising the overall cooling reliability. So I'm really exciting for the journey over the next five to 10 years and seeing how the evolution of the data centers is coming about, but also how heat recovery in data centers is coming. So thanks, Amara.
Yeah.
[00:27:56] Speaker A: Yeah, and thank you.
This is where we need to wrap things up and there have been a lot of questions about heat recovery.
So for more information on this topic, visit the Consulting Specifying engineer
[email protected] and don't forget to check back here regularly for new episodes.
Thanks so much for joining us and we'll be back again soon. Bye. Bye.
