Liquid Cooling Technology Revolutionizing Data Centers with LiquidStack’s Joe Capes

Uh, this is Alan Hummel. Welcome back here to Textron tv. My next guest is a first time, uh, first time on Textron tv.

I'm not sure we've had even the company on Textron TV before, but his name is Joe Capes and, uh, Joe welcome. Welcome to Textron tv. It's great to have you on here.

Thank you, Alan. Good to talk to you, Joe. We're gonna jump into Liquid Stack and, and everything about the company or what's happening, but before we do, I always like to give people a sense of who they're talking to.

Why don't you just give them, if you wouldn't mind, share a little bit of your journey. Sure. Uh, so I've been in the IT industry since 1992.

I can't believe I am actually saying that, but, um, I'm, I'm right there with you brother, so don't worry about it. Um, started out my career at a company called a PC American Power Conversion, uh, who were, who were the darlings of the Nasdaq in the nineties. I think, uh, the only stock that outperformed a PC was Dell.

And, uh, my first half of my career was really focused on, on power systems, backup power, specifically in, um, year 2000, made an acquisition, uh, of an air cooling company around the data center space. So I had to teach myself the refrigeration cycle and psych, psych psychometric charts, and, uh, basically taught myself the mechanical side of, um, of our industry. So a little bit of a unicorn in that way.

And, um, I'm a serial entrepreneur. Uh, this is actually my, my third startup scale up, uh, previous to, uh, one was an advanced battery technology company called Premium Power. And then I also founded a company called Centric, which was the first company to commercialize rear door heat exchangers.

Um, been here at Liquid Stack since 2019 and, uh, would be happy to tell you a little bit more about our company and, and our journey. Absolutely. Why, why do we do that?

I think that is quite a journey and I love to hear it. Let's hear about liquid stock. Sure.

Kind of a, I think it's a cool story, no pun intended, but, uh, when I joined in 2019, we were, um, mining Bitcoin and the company was actually founded in 2012 to mine Bitcoin when it was trading for about $5 USD. So, uh, super, super early days. Um, the company pioneered something known as two-phase liquid immersion cooling, mainly because, um, Bitcoin is, is, um, highly power intensive, requires a lot of, of heat rejection.

And, um, basically the more efficiently you can operate a Bitcoin mine, the more money you make. Um, we, we, um, we looked at the opportunity to advance our technology for other applications and ultimately pivoted out of crypto mining into data centers and edge computing. And, um, during the first half of liquid stocks journey, we, we really focused primarily on two phase immersion cooling, but in the last few years, we've augmented our product portfolio into single phase immersion as well as direct to chip liquid cooling.

And, um, I think really in the last 18 months, you know, the, the industry has really started to, to scale, um, namely because of the, the, um, advent of artificial intelligence and the, the scale up of AI right now, uh, AI systems that are being deployed by companies like, like Nvidia and, and other chip manufacturers are approximately 80% liquid cooled. But, um, over the next few years, we'll approach a hundred percent liquid cooled. So I, I would say our, our time has finally arrived.

Excellent. You know, like in real estate, location, location, location in technology, it's timing, timing, timing, right. You be in the right place at the right time.

And that's really, think about it a journey, right? Starting out Bitcoin mining, uh, at $5 Bitcoin, right? All the way through to the today and, and moving into the, really the look what's driving this is obviously AI data centers, right?

77% is vacant and 75% of all the data center space under construction right now is actually pre pre tenanted, pre pre taken, right? And, and of course, with all of these new AI data centers, really the only way to, you can't just air cool 'em, it doesn't work. The liquid cooling becomes a necessity.

So you have, you know, I think it's a trillion and a half dollars pledged towards AI data center, the data center space over the next couple years. It's a lot of liquid cooling, man, a lot of liquid cooling. Um, you know, Joe, I think some people have a view of liquid cooling, though.

Like I live down here in Florida, right? We have, our boats are liquid cool. The engines on our boats are liquid cool.

They suck in seawater, run it through a manifold that runs around the engine, not inside, but, and cools the engine down and then sends the water back out, right? It's a little warmer for the wear. Of course, that's not really how data center liquid cooling works.

Why don't you, if you don't mind, talk about, you know, at a base level for those people maybe aren't familiar with liquid, uh, cooling for data centers? Sure. Well, uh, the predominant technology that's being scaled right now is known as direct to chip Liquid Cooling, and it's actually been around for decades, um, has been deployed pretty heavily in the gaming industry where, you know, gamers are looking for the highest possible performance out of their compute systems.

Um, the, the principle is pretty simple. You, you have a, a cold plate, what's known as a cold plate attached to this, the surface of the, uh, semiconductor chip, and then you are, um, pumping water or a refrigerant, uh, through, through that cold plate and then removing the, the majority of the heat from the system via return liquid loop. Mm-hmm.

What a lot of people don't know is that the, the temperature of the fluid, uh, being, being used on the, the surface of the chip can actually be quite warm. So, for example, um, you know, fluid temperatures can be in excess of 40 degrees C which is, you know, north, north of a hundred degrees Fahrenheit. Um, that provides a lot of benefits and efficiency, um, but also in performance, because obviously chips are really limited by, by their ability to, to reject heat.

And so about 70 to 80% of the heat can be removed from, from a server in this way. The, um, the other types of, of a, of liquid cooling that are also gaining traction involve immersion, where you're basically using a dielectric fluid, which does not, does not, um, actually carry electricity or conduct electricity rather. And this allows you to actually immerse the hardware into this, um, bath of either single phase or two phase fluid.

And when we talk about phase change, it's really important to note that, um, two phase heat exchange is considerably more effective and efficient than single phase heat exchange, mainly because you're using the principle of, of latent, uh, heat removal or heat rejection, um, where you're actually, you know, taking, let's say a, a liquid and converting it to a gas and, and then back to a liquid again. So in layman's terms, um, I, I have a lot of analogies I like to use, but I mean, if you're a dog lover, uh, you know that dogs really, they don't have sweat glands the way that human beings do. They, they reject their heat through the surface of their tongue, which is a relatively small surface area, um, as a proportion of their, their total mass.

And, um, the way that they do do reject that heat is through phase change. They basically have saliva on the surface of their tongue that saliva converts from a liquid to a gas and, and is about, um, three or four thou thousand times more effective than air to hair heat exchange. So, um, you know, I think that the, the main point of liquid cooling and AI is that AI factories are, are revenue generating, um, sites.

They're, they're not cost centers like data centers. They're generating immense amounts of information and an intelligence in the form of tokens. And those tokens have value.

So the, the less energy that you're using for cooling, the more energy you can use to generate, um, tokens. Absolutely. At the end of the day, that's what it's about.

Um, Joe, you know, before we jump in, I, we're gonna talk about a new product GI guys have coming up, but before we do that, just for people who want to get more information, what's the website for Liquid Stack? Super easy. com or in, in present tense.

You don't need the, the, the three Ws. com. Excellent.

All right. Let's pivot if we can. You guys recently made a product announcement.

Why don't you, uh, share with the audience a little please? Sure. So, um, right now we're concentrating on the infrastructure that supports direct to ship Liquid Cooling, uh, products are called CDUs, which is an acronym that stands for Coolant Distribution Unit.

And it's essentially the heart of the liquid cooling system, and I mean in a very literal sense. So, um, the CDUs actually control very, very precise temperature, pressure, and flow. And, um, actually are, are an intermediate intermediary intermediary heat exchanger between the chiller plant and, and the rack in the data center.

Now. 35 megawatt CDU. And as we've been looking at the market, we've realized that CDU capacities are going to continue to increase in line with the, the, the rapid increase of power densities at the rack level.

So we, we know already that rack densities are, are reaching 600 kilowatts and are likely to go north of a megawatt in power density. And to put that into perspective, only two years ago, the, the average rack density was about eight kilowatts. So we're, we're talking about an in order of magnitude in power in heat density compared to where the industry was only only 18 to 24 months ago.

Um, what we've announced is a new platform called Giga Modular. And Giga Modular is the world's first modular, scalable, uh, 10 megawatt CDU. So it actually allows you to deploy the, the platform and it in a pay as you grow, uh, type of approach, but it also allows you to select the, the amount of heat rejection capacity that you need, um, according to your, your AI deployment.

And not all deployments are the same. Uh, we have some operators that are starting out with, with a couple megawatts of, of ai, uh, workloads, and then we have others that are deploying, uh, hundreds of megawatts. So, uh, the, the platform called Giga Modular is, is an approach that allows you to scale, um, very, very efficiently and also to help manage your, your cash flow in.

So, doing Excellent. Now, this, this is available now or just kind of being pre announced. We, we just announced the, uh, platform, uh, in, um, at Data Center World Congress in France, and we will be releasing it to sales in Q4 of this year, and it will be released into manufacturing in Q1 of next year.

We currently manufacture in the dallas Fort Worth metroplex and are in the process of, uh, expanding our manufacturing capacity outside of the US as well. Excellent. Joe, all we hear about is how much money is being pledged to build new data centers, these AI data centers, as they're called AI factories and, uh, and, but even traditional data centers are, you know, they're, uh, reading a, an old car plant in Ohio, Lordstown, or whatever it was called originally, uh, Foxcon bought it, and they, first they would, there were stores of making phones there and doing some other things, and now it's being just made it to a huge data center out near Reno Lake Tahoe.

They're making a, a data center, building a data center complex that's basically bigger. And I'm trying to remember what they said, like some outrageous thing. It was, it was huge.

Um, are you guys seeing that at Liquid Stack? Just they we can't, we can't get these things online quick enough. A thousand percent Alan, um, power is by far, by far the biggest challenge right now to the scale up of ai and some of the sites that you just described, we, we term adaptive reuse and, um, you know, there's many, many examples of projects that, that we have deployed where, where the site is, is, uh, rich in power capacity, but ultimately is being adapted from another use case into an AI factory.

And I, I think the, the speed at which these sites are being, um, constructed and deployed is, is, um, it's, it's really akin to what we saw in crypto, you know, maybe 10 years ago, because every day that you're not generating, um, AI tokens, you're not generating revenue. So the operators that are, are deploying AI workloads are, are really almost in a race against time, and that's putting a lot of pressure on the supply chain and the ecosystem that manufactures all of the, the infrastructure that supports these workloads. So it's, you know, it's part of the reason why we just added a second factory in Dallas and why we're also expanding our, our manufacturing overseas.

You know, as I said before, location, location, right? It's everything. So good for you guys.

Congratulations. It's a really interesting story, Joe, when you think about it, right? It's about being opportunistic, nimble, agile, enough to take advantage of, of where the market is.

Um, I wish you luck with Liquid Stack. Thanks for coming on Textron TV and making us smart about this a little bit today. Thank you, Alan.

My pleasure. I think All righty.