In The Arena by TechArena - Castrol on Liquid Cooling: Additives, Filters, Failure Modes

Episode Date: January 9, 2026

From hyperscale direct-to-chip to micron-level realities: Darren Burgess (Castrol) explains dielectric fluids, additive packs, particle risks, and how OCP standards keep large deployments on track.

Welcome to Tech Arena, featuring authentic discussions between tech's leading innovators and our host, Alison Klein. Now, let's step into the arena. Welcome in the arena. My name's Allison Klein. This is a Data Insights episode coming to you from OCP in San Jose, California. And because it's Data Insights, that means I'm here with Janice Norowski from Solidai. Welcome, Janice. How are you doing? Oh, thank you, Allison. I'm doing well. It's great to be back today. So this is day two of us recording the OCP Summit, and you really brought the heat with this interview. Tell me to use with us today. I may have brought the heat, but we're also going to bring in the cooling. So we'll talk a little bit about that today.

We have with us Castrol, believe we're not. And when you think about Castrol, you think cars, what's in my car, what's that oil, right? No one ever thinks about data centers. But that's what we're going to talk about today. We're going to talk about how Castrol is revolutionizing AI infrastructure in the data center. So to help us with that, we've got Darren Virges, who is the business development director for Castrol. So welcome. Thanks so much.

Great to be here. Darren, it's so great to have you. I was so intrigued by seeing Castrol at OCP. Can you please share Castrol's scope of engagement in data centers and why you guys are at OCP this week? So, yeah, as you mentioned, castrol well-known as lubricants manufacturer around the world, actually a British company based in Pengmore,

outside London, interestingly enough, a lot of people think it's an American company. But in any case, hey, let's be honest with cars going in the rechargeable direction and that type of thing. That's a steady business, but it's not going to grow like that anytime soon. So Castro got engaged in data centers because it turns out immersion cooling,

so a full single phase immersion is a poly alpha olefin-based oil that you can put an entire server down in and actually cool it. And that was sort of the entry because obviously, oil and oil is a lubricant at its base. And so we moved on to what is now really popular known as direct-to-chip cooling, which is a different kind of liquid, and that's really where a big expansion is for us now. So the scope of engagement then is obviously bringing the cooling

liquid to the data centers. So, you know, that's a big step one for us. And we want to be full solution provider when it comes to those liquids. That scope of engagement is really bringing it in, right? The liquid has to be properly installed. There's a lot of purity issues. There's a lot potential contamination, the liquid has to be maintained, and then in the end, somebody's going to want you to take it away. Can't just dump this stuff down the drink. Yeah, sure. And so to be, you know, from start to finish, it's just kind of what we're after. Where do you see the adoption going and how do you see this expanding the overall market? Yeah, you really need liquid cooling for AI servers. The chipsets are just too hot to be cooled by air. Basically, if you want to try to do air cooling,

servers would need to be so separated. You'd probably only get a couple of servers per rack in order to rush enough air through to get it done. So if you're going to have densification, which everybody does land and everything is precious, you need to densify and put the servers near one another. In order to do that and get them cool enough, you're going to have to have liquid cooling. The directed chip, for folks who aren't familiar, essentially you have the chip set, which is the major source of heat in a server.

And basically you have what's called a cold plate, which is essentially a heat exchanger. right up against the chipset. Sure. And you're running the propylene glycol, PG-25, through that coal plate and essentially taking the heat away from the chip set. The market, obviously AI, is a lot right now. I won't say everything, but it's a lot.

And there's plenty of people, everybody's into AI. And so everybody needs liquid cooling who's into AI. And so in the United States, obviously where our power is considerably less expensive than a lot of places around the world, directed chip just adapts itself to what people are used to, which is a vertical rack. One thing that I was interested in is a dielectric company like you. How do you look at the value chain of how your dielectrics actually get deployed in data centers?

Who are you working with in terms of types of companies to get that done? Yeah. So there's a lot of people involved, obviously. The major things that you really need to think through are, okay, who's the server manufacturer? There is the, let's call it the delivery of the liquid cooling. So the coal plague manufacturer, which can often be separate from the CDU cooling distribution unit. So it's a pop and heat exchanger. And then there's obviously the people bringing the fluid to the table.

And the PG-25, to be honest with you, when I got involved, I'm a chemical engineer by background. I have a pretty decent understanding of chemistry. I'm thinking purple and glycol and water. Yeah. We're just going to race to the bottom here because what's, What's technical? The additive pack, basically the magic dust and PG-25 is really where the action is. And it's got to do with one of the biggest things is corrosion.

Because you're going to have, like if you were wandering the hall and seeing some different coal plate manufacturers, you would probably see a lot of copper around. Obviously, copper is pretty easy to corrode with water. Right. Talking about 75% water. You've got iron and other things involved. then the heat exchanger and the whole CDU up to the coal plate. So you can't have that created. You don't want, you're a data center to Patina, like the Statue of Liberty.

Yeah, exactly. You need to then, you know, basically get qualified to demonstrate that your liquid is, in fact, not going to corrode everything. Because otherwise it clogs up and you're done. You've got to shut the rack down, shut the servers down. There goes all the money while you try to correct things. So bringing the expertise of knowing what could possibly go wrong in the corrosion aspect and keeping the fluid healthy is extraordinarily. That's where it's all at, really, because think about it, you could have multiple Cdus,

so you could have pumps that back things up. You could have repeats in a lot of places, but you can't repeat the liquid. It is a single potential source of failure. It's really the blood, right? We've got two eyes, two hands, two feet. We can get away a real lot of repeat, but we only got one system of blood. Yeah. And that's kind of way you can think of the cooling fluid. I mean, you really got to take care of it or you got to shut the body down. You got to shut the data center now. It's fascinating. So obviously we're here at OCP, right? And OCP is deeply ingrained in this topic. How is Castro and OCP collaborating and how is this impactful for the overall market? OCP is incredible at helping to, I think, define and standardize things that are going on in the data center. So several of my colleagues, some of my colleagues, some of the data.

Kim, Kosmo Peckioli, Chris Fletcher, a couple of colleagues in Europe, they are really involved in OCP. You have these committees on sustainability, liquid cooling, different parts of the liquid cooling, and they participate in meetings all the time to help with those standards. So, you know, I'm kind of the sales guide, business development guy at the booth, talking to people. They're busy out actually giving talks and in the talks here to continue to help bring. this sort of standardization and definition to the industry, it's extremely important. So, you know, we are deeply engaged week in and week out. And then, of course, to come to the main conferences, you know, kind of goes with the territory.

Now, obviously, this is the third day of the summit. What did you take it away from the summit this year in terms of the liquid cooling industry's advancement and how people are talking to you? Does it feel like we're getting to the point where mass deployments, are really starting to take off? Or what other key takeaways can you share? Yeah, it is really taking off, really taking off. As a lot of people may know, as you would expect,

the hyperscalers. So let's think in terms of Amazon Web Services, Microsoft, Google, those folks, they are ahead of the curve always. They're preparing and preparing up front. And so they've all begun to deploy direct-to-chip cooling at scale. You know, I think, compared to a year ago where there's certainly far less deployments of the liquid.

Now people are like, wow, okay, does the liquid coal? As I was saying, does it cooperate with this? Does it cooperate with this? What sort of additive pack do you have? Really starting to get into the nitty gritty of, okay, we're starting to put this out there. How's it going to work? How's it going to fail? What are the details?

Whereas maybe a year ago, it was a little more, okay, are there liquids and equipment available so we can start to push this forward. So it's been a big advancement in a year for sure. So what would you say you would want to see more of in terms of cooling in this market space? Like where do you want to see it go and how do you want to see it advance? Yeah. So I would say the real advancement is again the devils and the details. As I said, with a chemical engineering background, understanding the complexity of the fluid and just how different things can go sideways in terms of potential corrosion. When you're thinking, again, in terms of this coal plate that's touching the chips,

in order to get really efficient cooling, you have very small channels in there. I don't have any nice product memorized. Is there someone down to 50 microns, 100 microns, et cetera? But that's relatively small. So if you start to have problems in you're generating particles, the ability to clog that up, or in any case, to be less efficient in your cooling, is really a potential big deal. Sure.

And as we deploy, and as you have large deploy, People being maybe a bit more open about what problems they're having so that there can be a wider discussion of, okay, yeah, every time we're having this problem, 80% of the time is this. So we can research that first rather than there's five things and we have no idea which of them can be at fault. So we need to research all five immediately. I think building a general knowledge, which points to OCP, because that's what it's kind of here for, is to have broader definitions and standardization. Also. Yeah. I think I've just been thinking all interview about magic dust, and I think that that's going to be my key takeaway from the interview.

I had no idea there was an additive pack that would differentiate dielectrics, and I need to go learn more about that. Thank you so much, Darren, for being here today. Sure. Sure. And some of this knowledge with us. This was great. And, of course, this wraps another episode of Data Insights. Janice, thanks so much for your partnership.

Thank you, Alison. Thank you, Darren. Thank you so much. Thanks for joining Tech Arena. Subscribe and engage at our website, Techorina.aI. All content is copyright by Tech Arena.