Infineon key visual in RGB

Liquid cooling of AI data centers

We’re bringing you the latest from the world of semiconductors – straight to your ears! From quick takes on trending applications to deep dives on product innovations, our experts give you their take on the tech behind the tech.

Podcast

Aug 25, 2026

Podcast

As AI data centers push server power to unprecedented levels - reaching 600 kilowatts by 2027 and potentially megawatt-scale by the 2030s - the challenge of keeping these systems cool has never been more critical. In this episode of "We Power AI" on Podcast4Engineers, host Kelsey Markl sits down with Muzaffer Albayrak, Application Marketing Manager at Infineon, to explore the evolving world of data center cooling technologies. Together, they break down the shift from traditional air cooling to advanced liquid cooling methods, including direct-to-chip cooling, rear door heat exchangers, and immersion cooling, while examining the crucial role of Power Usage Efficiency (PUE) in measuring data center performance. The conversation dives into the semiconductor solutions - from silicon to wide-bandgap materials like silicon carbide and GaN - that power the fans, pumps, and inverters keeping these massive systems from overheating. Whether you're curious about cooling distribution units, centralized cooling architectures, or the future of AI infrastructure, this episode offers essential insights into one of the most pressing engineering challenges in the AI revolution.

Listen on Spotify Watch on Youtube

Speakers

In this episode of the Podcast4Engineers, host Kelsey Markl speaks with Muzaffer Albayrak, a senior manager for Application Marketing at Infineon

Kelsey Markl

Host:

Kelsey Markl is Head of Strategic Marketing Communications and Content at Infineon with more than a decade of experience in communications, content strategy, and storytelling. Not being an engineer is part of what makes the journey interesting. Kelsey approaches technical topics from the perspective of a curious outsider, asking the questions that help make complex ideas more accessible and relevant to engineers and broader audiences alike. Passionate about innovation and the people behind it, Kelsey enjoys exploring the trends, challenges, and breakthroughs shaping the future of engineering.

Muzaffer Albayrak

Guest:

Muzaffer Albayrak is a Senior Manager for Application Marketing at Infineon Technologies, where he leads market strategy for Smart Drive Systems and industrial drive applications. With over 15 years of experience spanning the semiconductor, industrial, energy, and mobility sectors, he has held key business development and marketing roles at leading companies including Wolfspeed, ROHM Semiconductor, and Mitsubishi Electric. Muzaffer specializes in Go-to-Market strategies for power electronics solutions - including SiC and IGBT technologies - driving innovation in industrial automation, robotics, e-mobility, and renewable energy applications. His international experience spans Europe, China, South Korea, and India, where he has built strong technical and strategic partnerships across the global semiconductor ecosystem. He holds a degree in Mechatronics, Robotics, and Automation Engineering from FH Aachen University of Applied Sciences.

Additional information

More episodes

Guest: What is happening in the server room? Increasing the heat.

 

Host: Okay.

 

Guest: Because by increasing all the server capabilities, one server rack would reach around the 600 kW level in 2027. And they have the outlook to reach the megawatt level in the 2030s. It's very challenging.

 

Host: Sounds like it could get pretty hot.

 

Guest: Yeah, it will be very— on the one hand, it's fascinating to reach such kind of capability in the server rooms. On the other hand, the question is, how do you cool?

 

Host: This is the Podcast4Engineers, the podcast you just have to listen to if you're interested in what's going on in the semiconductor market. My name is Kelsey Markl, and I'm your host on another episode from our We Power AI series. Today, we're diving into cooling, cooling our AI data centers. And I'm joined by my colleague, Muzaffer Albayrak, who is an application marketing manager at Infineon specializing in industrial drives applications. He comes with more than 15 years of experience in semiconductors, and I'm super excited to have you here.

 

Guest: Thank you for having me.

 

Host: Okay. And maybe our listeners with a keen ear have noticed you are responsible for industrial drives applications, but we're talking about cooling. What do industrial drives have to do with cooling?

 

Guest: The cooling will be realized through these 2 applications, fan and pump, and that both these applications are counted inside the Infineon below the industrial drive. Therefore, I'm staying in front of you.

 

Host: Okay, good. So now that we've got that out of the way, why is cooling so important in AI data centers?

 

Guest: It's very easy. The energies which are incoming, electrical energies, converted in the server rack into the thermal energies. And thermal energies need to be efficient way removed out of the server. Therefore, cooling is mandatory segment for the data centers.

 

Host: I assume then with what you're saying, more power leads to more need for cooling off that thermal dissipation. Are data center operators investing a lot in cooling?

 

Guest: It is roughly, there are some levels based on the technology, selected technology for the cooling, but it's roughly 10 to 20% around is the investment needed overall, this CapEx cost for the cooling structures.

 

Host: Okay. Then how do we measure that? How do we know if we're creating efficiently cooled data centers?

 

Guest: There are some parameters which are called PUE, power usage efficiency, and this value reflects how efficient data center is structured. And especially for the AI data center, the target is to reach below 1.1. And the PU value is calculated by, in a relation, in one relation, incoming energy related to the IT server energies which have been consumed during the processing in the IT servers.

 

Host: Okay. So basically, everything you need to keep a data center up and running, including Does that include lighting and like everything?

 

Guest: It's correct. This is the PU values. Let's take one example. For example, 11 megawatts of energy are incoming, and 10 megawatts are used or consumed by server racks. And so, 1 megawatt, which is required for all the other auxiliary equipment, lighting, cooling, Okay.

 

Host: So, it seems like cooling does play a big role in the overall efficiency of data centers. How can we cool? What options are there?

 

Guest: Typically, there are 2 ways: air cooling and liquid cooling. Until today, air cooling was a dominant technology on the market for the data centers. But by increasing the energy demand of the AI data center required more cooling capabilities. And the air cooling, which CRAC and CRAH is computer room conditioning, and also the CRAH is a computer room air handling, it's a 2-way of the air cooling, is going to be already on the limit. So, market requires next solution. Therefore, liquid cooling is getting more and more share on the market, especially for the AI data centers.

 

Host: Okay. Is there, are there different types of liquid cooling? You mentioned 2 for air. What different types of liquid cooling are there?

 

Guest: There are 3 different kinds of solutions existing by liquid cooling. Rear door heat exchangers. This is a hybrid cooling, which the air and liquid cooling is some in combination working together. Direct-to-chip cooling is a cooling plate on the GPU and CPUs to cool the water circulation through this, the cooling plate. And immersion cooling looks like some waterbed that the old data server putting inside the water to cool the water which the server is included.

 

Host: Okay. So, are the server racks then really just standing in a pool of water? In water.

 

Guest: There's a pool of water. That is possible.

 

Host: Okay. Okay. It doesn't necessarily seem the most practical though. What are businesses actually choosing to go with?

 

Guest: Yeah. The immersion cooling is that there are some complexities, of course. Yeah, it looks, sounds very interesting, but they have different, definitely some market share on the market. But direct-to-chip cooling or rear door heat exchangers are getting more and more share on the market. Immersion cooling will stay definitely on the market, but they have some commercial and technical complexity at the moment.

 

Host: Okay. You mentioned direct-to-chip cooling. Can we take a little deep dive onto that one for a second? And can you tell me what exactly that structure looks like in a data center if you go that way?

 

Guest: Yeah, of course. Let's do this in that way. I would like to explain which kind of the subsegment that is integrated in the AI data centers. If you talk about the cooling structure, there are 3 different segments: cooling tower, chiller. Chiller, it's not demanded all the time based on the location where the IA data center is located. If you are in Alaska, maybe you don't need it because there's ambient temperature so cold that cooling tower will be enough.

 

Host: Plenty of snow. There you go.

 

Guest: Based on this, structure of the place, the chiller could be some variable component and also CDU, which is cooling distribution unit inside the server room. And exactly these 3 cooling structures which we just now counted down, rear door exchanger, immersion cooling, and direct-to-chip cooling happen inside the server room.

 

Host: Okay. So, you said most of the magic is happening in the CDU, in the server room.

 

Guest: Yeah.

 

Host: Right. Okay. What's happening in the server room?

 

Guest: What is happening in the server room? Yeah, it's a good question. What's happening in the server room? There's happening, it's increasing the heat.

 

Host: Okay.

 

Guest: Because by increasing all the server capabilities, which one of the big players already mentioned that one server rack would reach around the 600 kW level in 2027, and they we have the outlook to reach the megawatt level 2030s, it's very challenging.

 

Host: Sounds like it could get pretty hot.

 

Guest: Yeah, it will be very— I would say this is, on the one hand, fascinating to reach such kind of capability and the server rooms. On the other hand, the question, how do you cool? Yeah, this is exactly this, some play hand in hand. Yeah, without having this efficient cooling structure, it doesn't help to reach such kind of the server capabilities.

 

Host: Are there any kind of innovative solutions or innovative, I don't know, ideas of how we can cope with that increased power?

 

Guest: Just now we mentioned, so direct-to-chip cooling is currently going to be leading way. On the other hand, the other liquid cooling definitely has a very important share. I don't want to then oversee that. There's still— they are also interesting giga value or the market share increasing. The question is how we can cool such kinds of higher demanded energies, thermal energies out of the server rack. And there are several kinds of technologies. For example, we are talking about the CDU, cooling distribution unit. Cooling distribution unit and they are not only done one way, which is correct. This depends on the design of this AL data center. It can be a cooling distribution located inside the racks with some slide rails. They have some pump and water channel, which is allowed the server heat to cool.

Or there are some sidecars which cooling distribution units integrated in the sidecar and responsible for the rack, server rack, which we call this in-rack cooling. And also in the row cooling, which is going to be more in the future perspectives and centralized solution.

 

Host: Okay.

 

Guest: With high-demand cooling distribution units are responsible for several racks.

 

Host: Okay. Since we're a semiconductor company, can you talk a little bit about what factors go into the decision-making process when choosing the right semiconductors to support such a system?

 

Guest: Yeah, we already talked about this power usage efficiency value in the beginning. Target is to keep as small as possible. The IT server capability is clear, 600 or 1 megawatt or 300, whatever it is. So additionally, power which we are requiring for lighting, cooling, and so on. So each component needs to be designed in a very efficient way. One of them, for example, where I'm sitting here for that, the cooling. Cooling, we learned in the beginning, can happen through the fan and pump. Each end system requires inverters.

Each inverter has some power component inside. And by selections of the right power semiconductor components allowed to solve these inverter solutions, which is responsible for cooling in an efficient way.

 

Host: Are we talking about silicon devices then?

 

Guest: This is what I would like also to mention, and that if we talk about the power semiconductor and we talk about efficiency, a clear answer for wide bandgap is getting in their mind, definitely. I don't want to then say it's Everything's needed to be wide bandgap, depend on the solutions, depend on the requirement on this customer level. This is also the Infineon advantages that the right and wide product portfolio can focus from silicon IGBT, silicon carbide MOSFET, GaN HEMT. We have a really wide product portfolio which can be addressed for each subsegment.

 

Host: Okay. And you mentioned before that a centralized solution might offer the most opportunities in the future. What do you think we can expect from AI data centers in the next 5 years?

 

Guest: Also, the AI data center has been affected from this general trend of the power semiconductor or power or electrifications. For example, DC grid topics. Currently, it's not the topic for cooling today, but it's very similar. As like the conversations from AC to DC centralized somewhere in the features, also the cooling structure will be centralized in the, for example, the row cooling, which we mentioned about that. The one CDU will be responsible for several racks. This way of solutions, we are expecting this will be then dominating on this AI data center solution. But today there are several effective solutions in the rack and the row cooling.

 

Host: Okay. I think from what I get from this conversation, The evolution of data centers are going to be increasing in power and power.

 

Guest: Correct.

 

Host: But with that, we can't forget about cooling, that it goes hand in hand.

 

Guest: Exactly. What we are mentioning about is all the electrification, incoming electricity is converted into thermal energies. And thermal energy can damage also the components, which is very expensive. on the server rack, for example. And therefore, this needed to be covered by decreasing the temperature of each server rack. And this is the responsibility of the cooling structure in AI data centers.

 

Host: Well, I'm very excited about cooling in AI data centers and excited to see where the journey goes.

 

Guest: Thank you very much.

 

Host: Thanks so much. And to our listeners out there, be sure to stay tuned for the rest of our episodes in this We Power AI series on the Podcast4Engineers.