How Do You Power and Cool an AI Data Center?

The processors inside modern data centers are becoming more powerful, more dense and a whole lot hotter. Keeping them running requires an entirely new generation of power and cooling infrastructure.

Chris Massetti of Donwil joins Empire of Lights to explain what happens between the electrical grid and the silicon, and why Donwil’s mission-critical expertise has suddenly landed the company in the middle of one of the fastest-growing industries in the world.

Hit play to learn how closed-loop liquid cooling can dramatically reduce water consumption, how modern systems move heat directly away from powerful GPUs, and why Donwil’s power modules help smooth the wildly fluctuating electrical loads created by AI computing. Massetti also explores the rapid move toward prefabricated and modular infrastructure, the growing demand for skilled technicians and why having compute located closer to Pittsburgh could become increasingly important for healthcare, research and other AI-intensive industries.

It’s a surprisingly accessible trip inside the powertrain and cooling system of the modern AI factory.

The Pittsburgh Technology Council produces TechVibe to explore Pittsburgh's technology and innovation ecosystem.

The Empire of Lights is brought to you by Comcast and Babst Calland.

Transcript:

This is the Empire of Lights podcast series, where we are exploring all things around the data center ecosystem here, especially in southwestern Pennsylvania. There's more to data centers than you would ever freaking know.

Everything from concrete, to cooling, to real estate, and everything in between. And today, I'm hanging out with Chris Massetti of Dawnwell, who I would say is one of the leaders in just knowing what's happening within the industry, plus an expert around all things when it comes to keeping the data centers nice and cool.

And by the way, I am Jonathan Kersting with the Pittsburgh Technology Council. And Chris, I'm super happy to be hanging out with you today because, man, I tell you what, you are like the resource of all resources. Things that people just don't know 'cause you're so embedded in the industry.

You're just hearing things before they really become public knowledge. So we're relying on you today, man. You're the expert. Well, thanks, man. I appreciate it. I guess, you, you and BK from the Tech Council when the data center really became part of the lexicon for everybody couple years ago, [00:01:00] I didn't realize it at first because it's been part of my life for 30 years.

Right. Back in June was 30 years at Dawnwell. Wow. Which I now own, and the company's 60 years old, and the only thing Dawnwell's done since 1966 is power and cool data centers- Exactly ... around the region. And the way the business started, we were a a dealership, essentially, or a, a rep agency, I guess

uh, for a company called Liebert out of Columbus, Ohio. Heard of those guys. Now it's called Vertiv- ... which is a household name. And we're their local dealer for PA, Ohio, West Virginia. And we we service mostly Vertiv equipment and, and others. And we do have some other brands in our portfolio, but that's all we do is essentially mission critical, and it's all we've ever done.

So when two years ago, the the most popular word in finance was data centers- ... I didn't realize at first, but now I realize the full gravity of what's going on here. Yeah. It's clearly the hottest sector , in business, in development right now, and the, the capital's pouring into it, and the interest is focused on [00:02:00] it, and the media's focused on it, for better or for worse.

I did have a funny story I wanted to share with you. First off, a Chris Massetti story is always welcome here on The Empire Lights- ... because you always have some great insights. So what is your story, Chris? Let's jump into it. I'm very excited. I was at a little cross tournament out east n- in the Harrisburg area.

Okay. And, I see this guy and He's wearing a - stop data centers T-shirt, and I'm like, "Oh, we're not gonna get along at all," and I actually was hanging out with a buddy of mine who works with NVIDIA, and I, immediately pointed that out to him. Surely. And then I I kind of watched the guy, 'cause I'm gonna see what was up.

It was... And, he's checking his phone, he's checking Torney engine to see all his games are scheduled. He's in his team chat with his team and- Yes ... he's he's, he's texting, he's taking pictures, he's taking videos, he's sharing them. I'm like, "Stop data centers, but everything you've done for the last 20 minutes requires a data center."

Exactly. It's so pervasive in our lives, people don't even realize it. There's such a disconnect. It's like the food disconnect, or it's, it's like people [00:03:00] say, "You're not in traffic, you are traffic." Exactly. Thank you. That, Chris, see, that's why I love it. You bring these things up, and you bring it down to this very point that we can understand.

Exactly. It's like everybody is dependent in modern society of using data centers. A case, I have a similar story where we had our second annual PA Data Center Summit at the casino, and there was a bunch of protesters that decided that they're gonna protest the event, which is fantastic. You can do that.

However, all the flyers they made were made with generative AI. It's that's fantastic, and I'm sure the computer that they made them on, of course, going through a data center as they sent it to their friends via their phones and stuff like that. Once again, data center. So it becomes this thing where there's a lot of hate out there right now, and I understand people have fears.

And one of the big things we wanna do is address some of these fears, in that there's... I think if, if people can really know the facts behind what's going on, I think some of this could be alleviated, but it's so hard to have these conversations,... This is why we're doing The Empire Lights as a way to really try to have some honest conversation around what is going on.[00:04:00] 

And that's why I think hanging out with you today, Chris, is just so important, because you always do bring some of these insights, and you kind of bring that level of sobriety maybe to what the problem actually is. And you're very candid as well, too, 'cause you also see there are many obstacles to this, and there are things that must be addressed.

This isn't this, this candy land of "Hey, we're gonna make tons of money and just build stuff." It's like, it's just not that, yeah, data and tokens are the product, and data centers and AI factories are the modern factories in the modern industrial revolution. Exactly. We cannot get around this fact until we stop using these tools.

Now, you can't travel, you can't order Grubhub, you can't do anything without an expansion of data centers. And certainly in the region that we're talking about, Western Pennsylvania- Yes ... there's just no large scale data centers right now. And if we as a region wanna take advantage of the modern industrial revolution, we should be leveraging our natural resources, our intelligence as a [00:05:00] region, our geographic benefits that we have, such as low instance of natural disasters, mild climates.

We should be leveraging these things to participate in this in a responsible way. And when I say responsible- Yeah ... what I mean is, listen to the community, and if there are gonna be negative impacts of these developments to a community, that data center developer is very, very interested in working with the community to offset those.

I've heard a ton of uninformed opinions around water usage. That's a non-issue. I've heard electricity's gonna go up. Yeah, maybe. It might, but it's gonna continue to go up because we're charging electric vehicles, we're electrifying buildings entirely.

If we wanna eliminate our usage of petroleum products to power our world, and we wanna electrify everything, that petroleum usage is just gonna get shifted to the point of generation- Exactly. Right ... through natural gas-fired power plants. So while I do understand it, [00:06:00] I also understand it takes seven and a half kilowatts to charge an electric car in your home.

I do understand that, a data center, while it may use hundreds of megawatts, We are a net producer of electricity in Pennsylvania. We have a massive amount of electricity, and the data centers that come in to utilize that electricity will offset the cost of building more transmission so that the general public can use it.

They are contributing to building out this infrastructure. It's not like the public is building this infrastructure, and the data center is just exploiting it. They are the drivers to update our infrastructure for the 21st century. Because it does need to be updated and because it has been falling behind, and this creates an opportunity for it to be updated.

And I know that as I've been doing this series, I've been talking to a number of, of experts out there, and when it comes to the terms of electricity, of course, many of these data centers are actually bringing energy generation on site, on-prem when they can, when that's a way to do it.

And, and also the idea that, they're gonna be held responsible for making these upgrades so [00:07:00] their data centers can work. It's not something that they want the public to be paying for. So that is something that I think is really important, that these data centers could actually be modernizing our grid because of what they need to service their needs, which I think is just super interesting.

Yeah, I understand the concern of people that may live next to one of these hyperscale data centers. 100%. Yeah. There's valid concerns there, and I get it. However, and, and incidentally, I live about four miles from the Springdale data center site that Brian Regli's developing. Yes. I get it. I totally understand.

It's in a river valley. There's, there's concerns there, but that is a responsible group of people looking at all options to make sure that they are the best neighbors they can possibly be. These guys are really spending tens of millions of dollars to try to have a positive impact on their community, create jobs. Now, one of the complaints is data centers don't create direct jobs. That's kind of fair because it's really a building full of silicon. But the truth of the matter is, you still need [00:08:00] to maintain the grounds, you need to maintain the equipment, you need to maintain the roadways, you need to maintain the HVAC, you need to cater.

When you have clients coming in, your, the catering businesses do well. These are economic engines that far outweigh just the direct jobs on site. We're looking at what is basically stranded infrastructure. Right now, these sites are currently not producing much in tax revenue, zero in employment.

So the idea that you're bringing something that is useful for all society, 'cause we need these data centers so we can actually send pictures and texts and all the things that we do as human beings, as well as generate with, with AI and so forth. But the idea that, we have a, a service that's needed, something that's putting a lot of money now into the tax base, and employing some people.

To me it's I don't understand where the problem is. Employing a lot of people. I could hire four techs every quarter, four service techs every quarter if I could find them. Yeah, if you could find them, right? The funny point that, that I heard, somebody said, "Ah, construction jobs, that's a one-time thing."

And I said, "Wait a second. Talk to the union [00:09:00] halls. We have fantastic trade unions here in Western Pennsylvania." The best. They want this business. Yeah. They want this business more than any business they've wanted before, because it's incredibly labor-intensive to build these. And these are two three-year projects that require skilled, very skilled electrical work, very skilled mechanical work, very skilled earth work.

And these contractors, they've been living and dying on a project basis since the dawn of time, right? Since we had contractors. They are totally fine with a three-year project. That's a great project, when you're on something for three years. Oh. Because there's always the next project. No one works on one project for a lifetime.

You upskill your halls. The, these guys that work on these big liquid cooling projects that are coming to Western Pennsylvania, these these steamfitters and boilermakers that are gonna work on these large piping projects, their skill set is gonna be second to none on Planet Earth when they're done with these projects.

That's exactly it. And if we don't develop that skill set here, then our contractors can't travel to other places to build these things. Because they're [00:10:00] being built everywhere. This is not just Western Pennsylvania, but these are being built, every state has some projects going on. I think back to my buddy in the no data centers or stop- Yes ... data centers T-shirt, and I'm like, what does that say to the union guys that are looking to get their guys in the field?

I just, I can't fathom- Yeah ... the idea that somebody wouldn't want a- 10-figure piping job. These contractors, it's, it's the lifeblood of that community, and without good solid tradespeople, we're not gonna be able to upgrade our infrastructure to accommodate things like- electrification, and accommodate things like aging sewer systems and water systems. These are things that we as a complex society depend upon, and having skilled tradespeople that are making very, very good livings and have job security into the next decade, that's, that's important. And that, these are the projects that people want.

Data centers used to be like, if you looked at the builders exchange, they were like a single-digit percentage of what was on [00:11:00] the local builders exchange- Right ... bid sites. Now they're like 20-some percent of the projects or more. Go to the builders exchange- Interesting ... see how many $40 million projects are over or bidding in Western Pennsylvania.

It's not many, right? But the ones that are, are data centers. Absolutely. So I'm curious if we could switch gears just a little bit. Sure. I want to delve into your expertise around cooling and water use. I think that's something I'm seeing the most headlines around right now for some reason is, is water use.

And of course, with Dawnwell being in the cooling business, you have a depth of knowledge here when it comes to the actual amounts of water that are being used and the new technologies that are coming online that are really, eliminating the need for as much water as people think.

I've heard stories, it's, it's as, it's as little as an Olympic-sized pool full of water that's used for, on, on closed circuit things. Tell me what's going on with the cooling. Yeah. There, there are technologies that exist that have been used for decades. It's called adiabatic or evaporative cooling.

We don't use [00:12:00] that here, first of all. Okay. Because we don't have a climate for it. It's too, it's, humidity's ... W- water does not evaporate well in Western Pennsylvania. We basically live in a, a rainforest, for lack of a better term. But these things are almost net zero users of water as compared to a building its size.

Yeah, the sinks, the toilets, the dishwashers, they all consume water just like anywhere else would. Right. But the actual cooling systems for the next generation compute loads, be it AI or simple CPU loads, the actual cooling systems are all closed-loop cooling systems. Now, if there's humidification that needs to be done, that is water consumptive.

And if there's refresh cycles to, to, to fill the systems back up, that consumes water. But on a day-to-day basis, once you flush and fill the system and it ... Let, let me back up. Sure. When you put these systems in place, they are two gigantic loops. There's a primary loop that goes to the air-cooled free cooling chillers, and there's a secondary loop that [00:13:00] goes out to the IT equipment, and that runs through a heat exchanger.

Because the fluid that goes to the IT equipment is propylene glycol 25, and that's for it to be non-corrosive inside the system. That flows inside the building through a pump system, and then outside the building flows a primary system, which is some type of propylene glycol so the lines don't freeze in the winter and burst the pipes, right?

Right. So you have two fluids. They're heat exchanged inside the building. Both of those loops are closed loop systems, but when you install them, you have to flush and fill those systems, and that does take- A lot of water. There are technologies exist today, and the majority of data centers are using this, it's called near zero, where you actually take water from the municipal system or the aquifer, you purify it, you flush those systems, you fill them up, and then the water that you flush them with, you clean it and you put it back into the municipal systems.

Wow. Almost perfect. It's near zero water usage. Interesting. See, that's something I've [00:14:00] just learned, which is why I love having these conversations, because that's something I did not know, and I'm, I feel like I'm fairly well-versed in some of the key concerns within data centers. So that is something that to me is just really interesting.

And I keep looking at as the compute continues to change, I, I feel like things are gonna be running more efficiently and, I don't know, cooler. But I, I feel like these are concerns that are being addressed- Yeah ... by engineers as we speak- So- ... all the time ... e- efficiency of a cooling system's driven by density really, right?

Okay. It's driven by w- what's called delta T, or change in temperature across the media you're cooling, right? Yeah. So the warmer the water you put in and the warmer the water you can return, the more efficient the system is. If you have to, if you have to bring water down to 54 degrees year-round in order to cool these systems, well, then that's pretty inefficient, because on 95-degree days, bringing water down to 54 degrees in entry water, we don't have to do that.

Our entry water's high 70s, low 80s, right? So there's a chunk of time in Western [00:15:00] Pennsylvania where we're just running our primary loop cooling from the chillers through these free cooling chillers, where we're not even running compressors. We're literally taking 95-degree water outside of the data center, we're cooling it down to 85 or 83 degrees, and we're putting it right back in with no compressorization.

Interesting. That's where you pick up the efficiencies. These chips run very hot, and what we're trying to do is keep them from melting, for lack of a better term. These chips run at 35, 40 degrees C. What's that, 100 and some degrees? 104 degrees maybe? Yeah, that's, that's getting hot.

Yeah. These chips will run perfectly fine at that. That's what the cooling water is doing, is keeping them below that 40 degrees C threshold. And w- for, for safety's sake, we keep them lower than that. But my point being is we can take advantage of our mild climates 10 months a year to get virtual free cooling, lower the electrical consumption of the mechanical equipment- And keep the chips completely cool.

And the density of these chips is [00:16:00] such that we're not taking up gigantic floor space for the chips, and we're not taking up massive amounts of water to do this. The tubes that cool these chips are half the thickness of a human hair. We're not- Really? Yeah. Wow. And so- No way ... the cold plate that sits, I wish I could talk with my hands, 'cause I'm kinda doing the hand thing, but- So cool, man.

No this is mesmerizing to me ... you sandwich a cold plate onto a GPU, right? And then that has capillary tubes through the cold plate, and you pump filtered fluid through that. Now, we filter it down to as small as five microns, which is absurd. Yeah. And then we filter it through these capillary tubes, and the fear is that we, we block these capillary tubes with particulate.

So that's why it has to be flushed- That five micron thing, yeah ... then filtered, and then filtered. And those filters need to be maintained on a regular basis by a qualified technician that can- Yeah ... go in and replace the filters in the coolant distribution units. So the, the efficiencies are all gained by the fact [00:17:00] that we're pumping massive amounts of fluid through very small tubes, a lot of them in very close contact with the chip.

So we're not wasting fan horsepower blowing air through these cabinets. So when you close couple the cooling to the load, you don't waste fan horsepower. Wow ... the efficiencies we've gained in the last decade in terms of cooling IT equipment shouldn't be understated. This is a very, very good system.

Yeah. And it's very reliable, and it's, it's maintenance intensive, so that does create jobs. For companies like mine that maintain this equipment, we're certainly happy to do that, but union service companies, the union halls want to maintain this equipment. This is- Why would they not? Yeah, this is part of modern infrastructure at this point, as far as I'm concerned, so there's human interaction. There's an incredible amount of, of very skilled engineers that are involved in designing these systems. And we run systems very close to the limit of their performance. We don't put two N equipment that we need. We don't put twice the equipment that we need anymore.

Everything's run very tightly on N [00:18:00] plus one to very tight tolerances, and that eliminates sprawl. And the, and th- these, these campuses can be huge, but they are very efficient for what they do. The amount of compute power we now get out of a 40 or 50,000 square foot building on a 5 or 6 or 10 acre campus is incredible.

They are very dense, and density drives efficiency. That's why I love hanging out with you guys 'cause you know the details on this. That is amazing when I think about. And, and it's only getting better. That's what I think is even cooler, is that people are working on stuff right now that's g- gonna make it even more efficient, so that's exciting.

It's so funny, though, It changes so fast, and the silicon manufacturers and the tolerances and the revs of the software they're using move so fast that if I miss a three-day news cycle, I miss something very, very important. I believe it. Yeah. One thing that I would love to pick your brain on is what's happening on the political front right now.

I know at, at the Tech Council, we were trying to ask [00:19:00] for a moratorium on the moratorium that they were trying to put here in Pennsylvania- ... on building data centers, and so far there's no moratorium, so I think we've been successful for now on that. But I, I saw New York somewhere passed a moratorium.

And we're seeing some things happening that are, are a little bit, I, I think, frightening. What has you the most concerned right now? What's, what's your thoughts on the political landscape? I think the short-term success of the outraged public- Is hampering the long-term financial success of the region.

In terms of these things are going to get built. They may not be built in Pennsylvania. But they're getting built somewhere. Yeah. They're getting built, and our reliance as a society on next generation digital technologies, I, I hate to say AI 'cause it's such a buzzword, but- Right ... AI is, is a- Yeah ... is a big word for a lot of- Yeah

things that are happening. Proximity, as we get into more inference AI where [00:20:00] we're demanding faster answers, where, CEOs are using AI chiefs of staff and they, they can't stand the latency, it's gonna drive the need for local data centers to keep Pittsburgh companies competitive, and we're not gonna have them.

That's what concerns me on the political front- Yeah ... is we're looking at this from a public outrage standpoint, which I understand. I totally get it. If you don't know what a data center is, if you don't know what it does, how it operates, and why it's important, you can be very outraged the fact that they're coming to town, and I get that.

I'm to some degree sympathetic with people that live in proximity to them and, things that you always thought were gonna be one way forever are now going to be a different way. Truly. There, there's clearly something to that, but the truth of the matter is it's gonna hurt long-term the viability of the region.

It's gonna hurt long-term the viability of our medical centers to do groundbreaking research that require AI for, cancer screenings and for for, for different, G- genomics, the, the big [00:21:00] revolution on the forefront with AI is in medicine, and we are a medical hub. That is one of the five pillars that holds this region up, and if we deny them the ability to have local processing power, we're gonna have serious problems.

And then there's, clearly the federal government is doubling and tripling down on their AI investment with their Stargate projects, which, the horse has already left the barn on that one. Absolutely. And we were not successful in, in getting one of the sites, but that's not gonna stop.

And A- AI right now is still primarily in the training phase and figuring out where it fits into, into our world, but as it becomes more pervasive in what we do every single day, and it becomes more localized to the individual, latency becomes more of a problem. And not having local AI processing facilities is gonna, is gonna first hurt large business, but I think e- eventually it's gonna hurt users that wanna use [00:22:00] it for their, for their home businesses, for, graphic design or music production or any of the things that you can now use- these incredible technologies to do. If you have latency, you do not have power in AI. Because if you're trying to , trade commodities, you're trying to, you're trying to corner the market on a, a real estate deal, when you have a latency and you can't make decisions quickly, and somebody else who has less latency can make quicker decisions- Yeah

I think that's gonna be the next battleground in maybe not two years, but certainly- Yeah, in the near future ... in our working life. It's really interesting you bring up... I never thought about this before, which makes me think about, ooh, there's some great conversations around this. Yeah, the proximity of data centers, especially when it comes to healthcare and research, and how that is one of , our main pillars here in Pittsburgh.

And the fact that if we don't have the compute power nearby to avoid the latency as you're talking about, like that, that puts us at a disadvantage to where we've been at an advantage, I guess for the longest time, and now we could be losing grip on that [00:23:00] because we don't have proximity to where another region would.

So one more reason where it's like, wow, I didn't connect those dots yet. And once again, you've helped me connect some more dots here, Chris. Pretty cool. Yeah. I, I feel like it's, it's one of these things where, it's a, it's like it, it's been bifurcated, right? There's the AI. It's like this amorphous thing that people use to make funny videos, right?

And then there's AI that really changes human lives, right? Yeah. And there's a wide gulf in between that we don't even talk about, and that wide gulf in between is, like, how are you gonna get the best prices on airline tickets or hotels or, get your kid into a school where they're increasingly using AI to do these interviews.

They're using AI to figure out who this person is, right? Yeah, exactly. You wanna have technology that works for the user and seems like magic, and if we as a region don't keep pace with that, and by the way, the innovation coming out of Carnegie Mellon and the University of Pittsburgh is insane.

And- They're not putting their dev servers here in Pittsburgh as far [00:24:00] as I know. Now, if I, if I know, if I don't know about it, I can't speak to it, but- Sure ... as far as I know, there's no large scale compute facility for the number one AI institute in the world. - I guarantee they're doing their processing somewhere else.

Somewhere else, yeah. Wow. That would be Carnegie Mellon. I don't know. I... And again, I could be totally ignorant. They could have some project going on that I know nothing about, and somebody from Carnegie Mellon's gonna listen to this and be like, "That kid's crazy." Fair enough, right?

But I just don't see large scale compute facilities being built within a five-mile radius of the number one AI school in the world, and- Exactly ... that's no knock against CMU. That place is amazing. No, it's just where it is. Yeah, exactly. Yeah. Yeah, I just think there's a proximity issue that people are - ignoring.

Yeah, which is why this is our opportunity now for this region to actually lay the ground rules properly so these data centers can be built safely and responsibly, and make sure they're not overly disruptive to communities. I don't know, there's, it's, there's so much to unpack here, but what we've talked about, wow you, you've just brought a lot of new things [00:25:00] to the forefront here, Chris.

And this is why we're having these conversations. Not only do I get to learn, but our listeners get to learn as well too, from the folks that are actually in the trenches every day doing this work, which you've been doing for 30 years now, Chris, which is pretty amazing. You were doing cool data center stuff before it was cool data center stuff.

It wasn't cool, I assure you. I was shocked the first time I was in a room and somebody was, like, really excited about talking to me. I'm like, "Man, that's, that's crazy." I'm cool now. Give it a little time. Don't ask my 13-year-old, I'm still not cool. Yeah, well, give it time. That's great stuff. But seriously, what, what an amazing conversation.

There's just some key takeaways here. The proximity is absolutely huge to me. Looking at the new cooling technologies that are coming out, and just really looking, just reminding everybody that, it's not just AI that gets run in a data center, it's just how we live our lives. You pick up your phone, send a photo, send a text, open an email, it's going through a data center somewhere. And they're not going away, either. We're not gonna be building less of them, 'cause the world needs them, and there will be places they will be [00:26:00] built. We want to bring some of that action, of course, right here to Western Pennsylvania, where not only do we have some of the, the resources and the land, but we also have the technology and the manufacturing that back it up as well, too, and great companies like Dawnwell that are ready to keep all the centers nice and cool.

Yeah, I, I'm glad you brought that up. So you, you keep, you keep talking about the cooling side of it, which is great. That's been our main business since the '60s, but really what we do is we provide all the infrastructure from the chip to the grid, right? Thermal management systems because these, these data halls still do need some air to move around.

So we manage both the the air side and the water side of the thermal chain. And then the powertrain The challenge with these AI data centers, when they, when these chips first came out, the way that the chips process data is they run at 100% and then they drop down to 80%, then they run back to 100% really quickly, and it, it's a dynamic load.

So what actually our bigger business is, is power modules. Interesting. So the [00:27:00] power module essentially negotiates the power between the grid and the chip, and it does that by allowing the chip to do all these erratic things, and showing the front end of the power module to the grid as being very stable, so we don't show these dynamic loads to the grid, which cause all sorts of fluctuations, especially- ... in multi-hundred megawatt arrays. The front end of the UPS modules that we provide, UPS, uninterruptible power system- ... they have a battery backup system built in. It's a couple megawatts per module, plus or minus, maybe one and a half, maybe two and a half depending on what , the customer desires.

And they, they actually block the dynamic loads from reaching the grid, and it looks very stable to the grid. And then of course, if the grid does go out, the UPS systems power the loads and power the pumps on the cooling systems until the onsite generation - spins up. Gotcha. Now, the onsite generation is of course a political football for everyone, because if you ever stood next to a [00:28:00] four-megawatt generator when it's running- Yeah

it's, it's not the quietest thing in the world. So we're seeing really cool usage of onsite battery energy storage systems in a microgrid configuration. We're seeing some offset of power by use of solar to charge the batteries. But you look at great companies, like one of the most innovative battery companies in the world is right here in Pittsburgh.

It's your neighbors. Eos. Yes. Eos. Yes, yes. They're awesome. They're doing big grid-level storage batteries for medium voltage, essentially. And then the power systems we're providing are the lower voltage. We'll take

13,000 in or 22 kV in, 480 volt out to power the loads, and then the utility just has to worry about powering the microgrid on campus and providing the 22 kV or the medium voltage to the front end of these power modules. And what's interesting with the power module business is a lot of the time in building these data centers is in the electrical wiring and the piping side of it.

What we're doing, and by extension what [00:29:00] Vertiv is doing, the, the main manufacturer that we're working with on this, is, Is providing prefabricated piping solutions overhead for the data center that involve all the chilled water piping for the secondary loop, and involve all of the electrical distribution for the actual racks and servers themselves.

So as you're tilting up these concrete buildings, you can then roll in our smart runs that have- Mm ... all your piping and all your electrical connections. You can roll in the power modules we designed for you, and attach those to the building. You can roll in your CDUs and your power distribution units, and you can just connect them all up.

And while it is very labor-intensive to do that, it's not 20 or 30 months on site of bending pipe and wire. It's all factory-built assemblies that are designed specifically around the silicon you're using. So if you tell us you're u- what NVIDIA product you're using- Okay ... we can design a smart run for that.

So when the building's tilted up and the chillers have arrived and the power modules arrive, the smart [00:30:00] run arrives, and then the tradespeople are just connecting those prefabricated modules together, which is speeding deployment probably closer to around a year, maybe a little better than a year rather than trying to bend all that conduit so this is all pre-built in a factory. That's sort of, when you talk about like our business, we have basically three practices.

We have our small power and cooling practice that works with computer manufacturers to cool and power closets for network applications. Hospitals that have 1,000 buildings, we put power and cooling systems in those closets. We have another practice that supports- contractors and engineers for plan and spec stuff for small data rooms and buildings.

And then we have this AI hyperscale practice, which is designed around these prefabricated modules primarily. . And it's a full-time job, and, and these sites are in- incredibly large. It sounds like three full-time jobs to me, as far as I'm concerned. Well, yeah in, in the first two, the, the channel business that, that works within the [00:31:00] closets in the network that's a great busine- We love it.

That's my, actually one of my favorite businesses, 'cause that's how I got started in this industry. Our plan and spec business is great. We get to work with the contractors. We get to do some design build work with the local... we have local design build general contractors that are building these spaces out for enterprise customers every single day.

And they're fantastic projects to work on. And, they're using the same technology as these hyperscale data centers, but they're scaling it down to their internal GPUs. One cluster of GPUs can take the place of an entire room full of CPUs. Again, less real estate- ... more dense power, much more efficient.

And that practice is, is thriving, and it's, it's, it's great. We love that, but the AI hyperscale business has really moved decidedly towards modularity. Makes sense. No, this is, wow. I don't know first off, I think you sell yourself short. I think you get to have a lot of fun, and I think 60 years, I'd say you're cooler than ever, man.

Simple [00:32:00] as that. Because seriously, these are the technologies that are literally powering the future and this conversation's just been incredible. So many things we have covered today, Chris. These are the conversations we need to keep having so people understand the new technologies that are in play, the problems that need to be solved, and more importantly, how we're trying to solve them all.

So man, you are an expert, dude, without a doubt. Can't thank you enough- Thanks ... for hanging out with us today, man. Great stuff. And I'm always happy to engage. I'm always happy to have conversations about this. I don't think- Great ... it's a zero-sum game.

Thank you. I don't think you have to have one or the other. I think- Right ... there could be responsible development, and I, I would like to see that happen here in the local region- for the good of everyone. That's what it's all about, man. That's why we're doing the Empire Lights here, is in order to kinda get these conversations going and see who we can connect, and everything in between.

Chris, you are the best, man. We got, of course, links to Donwill in our notes for the show here as well, too. Check them out, and thanks for doing all the great work that you do, Chris. You are the best. Thanks, man. I appreciate it. No doubt. Hey, in case you forgot, [00:33:00] this is Jonathan Kersting with the Pittsburgh Technology Council and TechVibe bringing you the Empire Lights, and we can't do it without our friends at Comcast that are helping us make it happen.

I cannot wait to see you on the next one.