Why Are AI Data Centers Moving to Liquid Cooling?

Simon Jesenko says most of the water figures in the papers describe air-cooled data centres. Once they're fully liquid cooled, that water consumption is gonna drop immensely.

Clips and outtakes on Instagram ›

In this episode · 16 chapters

  1. Water use: data centres, almonds and golf courses
  2. Open and closed loop cooling
  3. Meet Simon Jesenko
  4. What people get wrong about data centres
  5. From direct-to-chip to system-level cooling
  6. Where the water actually goes
  7. Hot aisles and cold aisles
  8. Cold climates and space
  9. Selling liquid cooling to a risk-averse industry
  10. Immersion cooling and crypto mining
  11. What dielectric liquid is
  12. The oil companies behind the fluid
  13. Power supplies and the edge
  14. Is there a ceiling?
  15. Why power equipment is easier to cool
  16. The handover question

Simon Jesenko is CEO of Iceotope, which cools servers with no fans and no air. Between 15 and 20 litres of dielectric liquid, a highly refined oil, circulates inside each server and reaches every component, from the GPU to memory and storage.

He explains why direct-to-chip cooling captures 80 percent of the problem, why early installs leaked, how immersion cooling started in crypto mining, and why Iceotope is focusing on the power supply units inside the rack.

Simon Jesenko, CEO of Iceotope, tells the Thinking On Paper podcast that AI data centres are moving from air cooling to liquid cooling, and that most of the data centre water use reported in the press comes from air-cooled sites. Iceotope cools whole servers with a dielectric liquid in a sealed loop. “Once they're fully liquid cooled, that water consumption is gonna drop immensely.”

Why are AI data centres moving to liquid cooling?

Simon Jesenko says most of the data centre water use reported in the press comes from air-cooled sites. Iceotope cools whole servers with a dielectric liquid in a sealed loop, and he expects water use to fall sharply once sites are fully liquid cooled.

Why does air cooling use so much water in data centres?

Air cooling drives most data centre water use, Jesenko says, because chilled water is needed to blow cold air over the chips and much of it evaporates. He says most of today's coverage of data centre water consumption describes these air-based systems, and that most sites are still part air, part liquid cooled.

How does direct-to-chip cooling compare with Iceotope's approach?

Direct-to-chip cooling, a cold plate fed with liquid that sits on top of the processor, leaves every other component air cooled. Jesenko says direct-to-chip “captures eighty percent of the problem”, while Iceotope's system-level approach removes all fans and cools memory, storage and every other part with liquid.

What liquid does Iceotope use to cool servers?

Iceotope cools servers with a dielectric liquid, a fluid that does not conduct electricity and can touch electronics, which Jesenko describes as a highly refined oil made by the major oil and gas companies. He says each server uses roughly 15 to 20 litres that circulate in a sealed loop inside the chassis, with years of operation and no degradation.

Why did tank immersion cooling never scale in data centres?

Tank immersion, where whole servers are dropped into vats of fluid, first succeeded in crypto mining but never scaled in data centres because it changes how sites are built and maintained, Jesenko says, down to needing a crane to lift equipment. Iceotope uses a small amount of liquid inside each chassis so data centres stay rack-based.

Why is Iceotope targeting power supplies and edge sites?

Iceotope is targeting in-rack power supply units and edge deployments, Jesenko says, because power components are hard to cool with cold plates and edge sites lack data centre infrastructure. Removing fans from power supplies frees energy for compute, and sealed liquid cooling runs “completely silent” in dusty or humid locations.

Where Jeremy and Jesenko disagree

When Jesenko suggests data centres in space would not have an overheating problem, Jeremy calls that “another myth”, pointing out that orbit still needs heat exchange and large radiators. Jesenko agrees the heat would still have to be dealt with somehow.

Questions

Why are AI data centres moving to liquid cooling?

Simon Jesenko says most of the data centre water use reported in the press comes from air-cooled sites. Iceotope cools whole servers with a dielectric liquid in a sealed loop, and he expects water use to fall sharply once sites are fully liquid cooled.

Who is Simon Jesenko?

Simon Jesenko, CEO of Iceotope, tells the Thinking On Paper podcast that AI data centres are moving from air cooling to liquid cooling, and that most of the data centre water use reported in the press comes from air-cooled sites.

Will liquid cooling cut data centre water use?

“Once they're fully liquid cooled, that water consumption is gonna drop immensely.”

What does Iceotope do?

Simon Jesenko is CEO of Iceotope, which cools servers with no fans and no air. Iceotope cools whole servers with a dielectric liquid in a sealed loop.

Transcript

Water use: data centres, almonds and golf courses

Jeremy Gilbertson Disruptors and curious minds. Welcome to another episode of Thinking on Paper. My name is Jeremy. This is Mark. We are back in the realm of data centers today. Very, very controversial topic in a lot of ways, but it's also the infrastructure that powers a lot of the stuff that we do and how we move through the world, not just AI, but all the other stuff. we've got an interesting guest today. We're gonna dive deep into some alternative solutions that provide cooling to IT infrastructure in data centers that kind of make the whole equation of data center cooling a little bit more interesting. So I want to start, Mark, if I could, just from

Mark Fielding Go for it.

Jeremy Gilbertson the high level and kind of say, all right, what does cooling in data centers look like? Right. So water usage in data centers and that sort of thing. I did did a little bit of research right now data centers are kind of the bad guy. you know, could data centers do things a little bit better? Sure, in a number of ways. They are listed as a high water user, right? So I did some did some research and did some comparatives and actually found a few more things that use water as much or more than data centers. And it isn't what you would expect, Mark. Do you have any guesses?

Mark Fielding Well how'd you agriculture.

Jeremy Gilbertson Probably not far off there. I think I think you're on the right track. So let's let's level set. So Lawrence Berkeley Labs, Lawrence Berkeley National Labs, pretty pretty reputable data source. right now has data center water usage at 17.4 billion gallons per year. That was 2023. That's projected to rise to anywhere from 38 to 73 billion gallons by 2028, Lawrence Berkeley National Labs. Let's talk about some other usage. Mark, you said agriculture. That

Mark Fielding Cows.

Jeremy Gilbertson it that is correct. So cows were brought up too, but let's let's talk about agriculture.

Mark Fielding Were they?

Jeremy Gilbertson So California almonds, California almonds, 1.5 to 1.8 trillion gallons per year. And this is according to California Water Impact Network, study done in September 2024. C_WIN December 2024. alf alfalfa. Alfalfa. Alfalfa. Alfalfa. Yeah, even even bigger. or actually right around the same, it looks like the the water usage there, but according to the Pacific Institute, shows this water usage for alfalfa as more than almonds. Beef is up there as well, golf courses up there as well. So before we before Do not because I dive

Mark Fielding we do you have avocados? Because I was all so we gotta remind the listener that this is we in a week of very doom-ridden, horrible news about AI, we're bringing the good news today. Because we've been talking about water sources and water resources, but we actually have a man and a company that are their aim is to limit and change how water is used in cooling data centers. And you know, it's some good news. But I like how in today's show we're gonna kind of be positive and great about data centres, but now we've annoyed all the farmers and the golfers.

Open and closed loop cooling

Jeremy Gilbertson That's right. Well, someone's always got to be mad. right before we bring Simon on, wanna talk about just the systems, how cooling systems technically work in a data center. And you have, let's just keep it really simple. You have open loop and closed loop systems. open loop system is basically think of like a shower. When you get in a shower to cool off, cold water comes down, hits your body, goes right out the drain, right? So you're losing the water, you're losing nearly all the water that's going in the system. You have closed loop systems, which is more like How a radiator in car works, how a refrigerator works.

Cold water comes in, goes around the hot stuff, gets hot, comes back out, and goes in to get re-cooled. those are the two kind of systems. We're gonna actually gonna talk about cooling on the chip level. We're getting down into the IT equipment and not just into the rooms that are housing the IT equipment. So long diatribe, Mark. Who are we talking to today? Let's get into the weeds and figure this thing out.

Meet Simon Jesenko

Mark Fielding The our guest today is Simon Jesenko, CEO of Iceotope, and Iceotope are well, I'll quote their website, precision liquid cooling that solves thermal bottlenecks in AI infrastructure. And we are going to learn more about that today. So, Simon, thank you for thinking on paper with us today. And Jeremy's shower analogy. Does it work?

Simon Jesenko Yes, definitely it does. Well, first of all, thank you for inviting me to the show. Jeremy Mark. pleasure to be here and talk about cooling and talk about the issues the data centers have today, especially when it comes to water consumption.

What people get wrong about data centres

Jeremy Gilbertson S Simon, can we can we just what do most people get wrong? what what do the general public mostly get wrong about the data center story?

Simon Jesenko I would say, you know, the the perception of what data centers were before, when they were air cooled and the air cooling technology was used is what everyone thinks is gonna happen as this, you know, data center crazy rollout is is happening, is gonna continue. But I can tell you there's a lot of technologies that are being introduced into those data centers today that solve a lot of those issues. You know, the two main issues with data centers is energy consumption and water consumption. And if you think about those two issues, you know, there's different technologies being introduced into those data centers to combat those two, I think, very important things when it comes to how people perceive data centers and how they operate.

So the new data centers of tomorrow are gonna be way more efficient and built differently than the old data centers have been.

From direct-to-chip to system-level cooling

Jeremy Gilbertson we started with casting just super basic high level the types of cooling in general. I mean, I'm not gonna go into the difference between air handling units and computer room air conditioners and all of those things, but in general close

Mark Fielding Stop showing off.

Jeremy Gilbertson closed loop, open loop systems. and now we're talking about the IT equipment itself that lives in these racks, that lives in these air-conditioned rooms. We're now able to get precision cooling. down to the potentially the chip level. Some people are out there doing it. You're doing it a little bit differently. Tell us, give us a little history about precision cooling and then what your new take on it is.

Simon Jesenko Sure. And I can give you a little bit of history of what sort of liquid cooling is, if you will, because it's important to capture that. So traditionally, data centers have always been air cooled, which means a lot of air being blown across and over the equipment to take the heat away from all the hot spots, all the components. And that technology has reached its limits with the you know the AI, if you will, evolution. And all the new chips that are being introduced now into the into the space. So what's happened there as a result is that industry had to move to a different alternative way of cooling that compute.

And liquid cooling was the natural shift. Now, where you look at the let's say the the components that get hottest today, the focus of the data center world is on the chip itself. That's the hottest component that is. That is in the system today. So the technology that's been introduced into those data centers is something that's called a direct-to-chip technology, which is basically a plate that's put on top of the chip. There's copper pipes attached to the plate, and there's water or dielectric liquid floating into the plate and extracts the heat from the chip. But that means the rest of the components inside of that system are still air cooled.

So there's still an element of air cooling, which is this highly inefficient way of cooling Now, our technology basically is a system-level technology. So we use a system level approach. So we cool all the components inside of the system. And we do it a little bit differently to the technology that I just described, which is the direct-to-chip technology, if you will. We basically remove all the fans, all the air components. There's no air components at all in the system anymore. And we use dielectric liquid, which is this highly refined oil that can touch IT. And is very efficient in terms of how it captures heat.

And that that basically liquid gets pushed around with pumps that we design and control around the system and all the components, not just the CPU and the GPU, the hotspots, but also all the other components, such as memory, storage, etc., and cools the system as a whole. Now, this is highly, highly efficient when it comes to heat capture, which the current liquid cooling technology direct to chip is not. So if you think about it in simple terms, Direct to chip, which is the technology that's being rolled out today, captures eighty percent of the problem. Our technology captures hundred percent of the problem when it comes to cooling.

Where the water actually goes

Mark Fielding So that direct to chip cooling that you spoke about.

Simon Jesenko Yes.

Mark Fielding So the water goes onto the plates directed to the chips to cool it down. Is when I open a newspaper and I read about the extravagant amount of water that data centers are using, does that refer to that? direct to chip cooling or is there something else that that water usage is referring to? So it when GROK or OpenAI or Anthropic are building and have built these huge data centers, is that where that water consumption is pointed?

Simon Jesenko depends what the what what the references are there. But I would say most of the what's written in the papers today about water consumption really reflects to your point earlier. I think Jeremy when you mo mentioned the closed and open loop, right? It's that open loop system with air, which you have you have to have cold water inside of the data center to basically be able to blow cold air over the chip. So it's still air based technology that's being used. Where this is talked about, because that technology really uses huge amounts of water. When you introduce direct-to-chip technology, and you still have a bit of water there, still have to have to have a bit of a air cooling there in that sort of that technology, but it's it's much, much more reduced versus the traditional air-cooled data center, where you know this usage of water is astronomical.

So when you talk about when you when you when you're reading about this today. The transition is just happening. So a lot of data centers are still air-cooled and they're a combination of liquid cooled and air-cooled. So it's not fully liquid cooled. Right? Once they're fully liquid cooled, that water consumption is gonna drop immensely. It's gonna drop by a huge amount.

Mark Fielding as a percentage of the hyperscalers at the moment where they are in that transition?

Simon Jesenko Oof, this is yeah, it's early days still, I would say, and the data center rollout. I wouldn't have exact numbers to for you. I wouldn't wanna sort of know speak at the top of my mind on that really. But I think it's very early days.

Mark Fielding But it's a transition, it's a it's a change, It's a transition. okay.

Simon Jesenko And it's going from you you think about it also, it's going from changing some parts of the data center to liquid cooling while still the re some other parts remain air cooled. So, this is not like a, you know, you you don't have a lot of data centers that are built from the ground up completely liquid-cooled, right? So it's introducing liquid cooling gradually into those data centers. There's still some parts of compute that are that is able to run with air cooling. And you know, you have to understand that data center ecosystem is very risk averse. So they would like to use air cooling for as long as they could because they trust the technology.

They it's been around for for many, many years. Introducing water into a data center or you know liquid into a data center is counterintuitive in a lot of ways because it's sort of okay, you have IT and liquid that doesn't normally mix, right? So everyone is thinking I need to do this gradually, slowly, right? So before pushing the button on everything, risking my job while doing so, etc. Right. So I I want to be very, very careful as I introduce liquid cooling into my data center. So it's going sort of gradually is being built up. But there will be data centers completely liquid cooled in the in the future.

Jeremy Gilbertson So, Mark, just

Mark Fielding Got it.

Hot aisles and cold aisles

Jeremy Gilbertson to just to give you a little bit more of a a look into this stuff, kind of from the from the from the basic version of it. So think of it as like room-based cooling. So you have like a giant room that you're trying to cool and you set up your racks to where the front of the servers or the front of the IT equipment are facing in one direction and the rear of the IT equipment are facing the other direction. So you have what's called a hot aisle and a cold aisle. So a hot aisle is where the cool air So you the cool aisle is where the cool air gets pushed into the server, sucked in through the fans that Simon mentioned,

Mark Fielding Right.

Jeremy Gilbertson cools the equipment down and exits on the backside in the hot aisle to where the the the hot air kind of is. And and managing the difference between hot aisle and cold aisle is how you get efficiency. But now we're talking about actually bringing that down to the to the device level, to the chip level that will and there's a cup there's like four different ways, I think, under my research that you can do this. direct to chip or or precision liquid cooling. and and we can kind of dive into that a little bit. Does that kind of cast how the airflow works? And you're basically putting all this cool air into this giant room, designing it, hopefully it gets it in the right spot.

Mark Fielding Works, thank you.

Simon Jesenko Yeah. You have to have a lot of co cold air to be able to extract all the heat from these air cooled components, right? And that's where the water consumption comes in because you need cold water to basically push through the pipes so you can have that cold air blowing, blowing through. And there's a lot of water that's being evaporated in that process.

Cold climates and space

Mark Fielding So you couldn't build them out in Antarctica 'cause then you've got the latency problem or is that just it's not cold enough because even that air's not cold enough? Or is it a latency problem? What like why or just an environmental one?

Simon Jesenko Well there there are da yeah, there are data centers that are that are efficient, that are being built in colder environments, for sure. That is a that is a a benefit. You can use those environments, you know. Even Elon Musk has been talking about building data centers in space. You wouldn't have the problem of of heating of of overheating in space, right? So there is an element of external temperatures that impacts what kind of a data center you need, for sure. So that would be a positive in those in those sort of climate in those climates,

Mark Fielding I'm not I'm not saying we should build data centres in the Antarctic, Jeremy.

Jeremy Gilbertson Yeah, we're Phillips already got going in in Leo, so we're we're in good shape.

Simon Jesenko Mm-hmm.

Jeremy Gilbertson The that's another myth though, too, when you think about putting data centers in space, like, space is cold, you still have the heat exchange. You still gotta get the heat out. And so you gotta put these giant radiators up there too.

Simon Jesenko Still still

Jeremy Gilbertson And it's like

Simon Jesenko would have to deal with that somehow. I don't know how we would deal with latency, but I guess that's you know, I know that they've been sort of thinking about how to do that as well. some companies have some interesting ideas that I've read about, but yeah, that's a different problem altogether. I think we are we'll stick to Earth for now. Yeah.

Mark Fielding Let's stay on earth. Stay stay on terra firma for a minute.

Jeremy Gilbertson Totally. Let's stay yeah, let's stay terrestrial.

Simon Jesenko Yeah, correct.

Selling liquid cooling to a risk-averse industry

Jeremy Gilbertson So all right, well let's let's talk about you mentioned risk aversion, and I can definitely echo that. You know, in my time in the space, there it's definitely, hey, this is tried and true. It keeps things up, downtime is is death in data centers and the whole nine yards. if we

Mark Fielding That's not unique to data center world, that's a h the human condition.

Jeremy Gilbertson Potentially. I don't know. You that that's a whole we could go philosophical on this point. We, you know, people need to create space for themselves, but that's a whole nother Simon. We're going, we're going hard left. Let me bring it back here. Ha ha ha. Bring it back. we're looking at the risk aversion piece. the way I understand how your stuff works is you don't make servers, you don't make network appliances, you don't make any of that stuff. You have a technology to bring cooling to more precise levels on that equipment. So herein lies the challenge. You have to go to the Cisco's, the HPs, the NVIDIAs of the world and say, Hey, I'm gonna weave more infrastructure in on your box to make it better.

But it's got some liquid in it, it's got some stuff in it. that's pretty tough. How do you approach that? And how is that going?

Simon Jesenko It is tough. It is tough because it's a shift, right? A shift in the mindset, a shift in the way also this sort of I would say this OEMs, ODMs operate that we are that we are working with, right? And rightfully said what what you said, Jeremy, is correct. We are basically a liquid cooling technology provider, if you will. We're experts at liquid cooling. We have a special way of how we cool any form of IT, not just servers, but other other forms of IT as well. So we need to work with the IT hardware manufacturers designing systems from the ground up with our technology embedded in.

That's the whole model that we are pursuing. So it requires a lot of proof, I guess, in the first instance, right? So you have to prove that your technology is superior to any other technology that exists. And we are being compared to direct to chip today as a technology provider. So first of all, you need to basically convince the the other party that your technology is superior in terms of performance. Because at the end of the day, if it's superior in terms of performance, performance is converted into a commercial outcome. We talk about we talk a lot about tokens per watt. If you can, let's say, save power that goes into cooling and push that power into compute, you get more tokens per watt.

You have more, you generate, let's say, more more revenues for the for the operator. So that is something that's important to do in the first phase. The second phase is convincing the the other party that immersion or if you will, liquid inside of a box, which is what we have, right? We have ins inside of a closed box, we have liquid floating around the components, is is reliable and it can last for years and years, which is what you know a lot of these operators worry about is reliability of these new technologies, even with direct to chip for that matter, right? A lot of the initial direct-to-chip deployments have had leaks.

that is something that is a problem. So it is going through them a lot of testing, a lot of work around you know the the design, the mechanics of how it all works, a lot of work in terms of testing the data, which then brings them and convinces them that this is the technology and the solution they should they should go with. And you know what's interesting, you asked me about how is this going? So if you asked me that a couple of years ago, I would say. we've been, you know, developing, perfecting this technology that we have today. but we weren't able really to push it into the market at scale.

Now the market is coming to us. There are we see more and more opportunities for this technology coming to us through the likes of OEM's ODM because they know us, because we worked with them in the past on certain, let's say Development RD projects just showing them the benefits of our technology. They know what it can do. Now they're saying, can you guys cool this device? We haven't found a better way to cool it. It needs to live in very extreme conditions. It needs to be, it needs to be operational all the time, it needs to be cooled properly, et cetera, et cetera. And we're seeing opportunities coming our way.

So this is now starting to open up more and more. and we see, you know, there is a place for this technology. But I would say in this point that it is something that's a as a technology play, it's something that's future-proof, that lives beyond the direct-to-chip technology in the future, what we have. This is a very important point because as these operators think about what comes next after direct-to-chip, it will have to be some form of liquid immersion, which means using using liquid to basically cool all the components inside of a IT device. So that's where we're heading for sure.

Immersion cooling and crypto mining

Jeremy Gilbertson So Mark, for context, the industry's been putting IT equipment in liquid for a long time. Like there I remember ten years ago, there were these there were these racks. So normally you have like a eighty four inch high rack, you know, twenty four by thirty six or whatever it is. These were laid instead of standing tall, these were actually laid down and the servers would get dropped into this liquid, which I'm guessing is pretty similar to this this dielectric liquid that you're doing and is it all rooted in how transformers, electrical transformers were initially cooled? Is that the same concept? Is that where the stuff came from?

Simon Jesenko I believe so. You are correct. And the first sort of, if you will, instances of where this liquid cooling way or way of cooling, which is called immersion, was successful, was in the crypto mining space. That's where the first successes sort of started. And what you've described there, Jeremy, I think you're talking about the big tanks, right? So there's, and this is what we call in our world tank immersion, right? So these were big tanks. Filled with dielectric liquid, you would drop servers inside, cool them, super efficient in terms of cooling. Of course, you have the server inside, then you would lift it up and you would drain And they had initial successes.

This technology had initial success of successes in the crypto mining space. And it tried to push itself into the data center space over the years, worked with several hyperscalers and OD OEMs. But it changes the problem with that technology is that it changes the structure of the data center. It's no longer based anymore, right? So suddenly you have these bots that you're introducing into the data center. So these tanks that you're introducing into the data center, which means you have to completely change the way you maintain the data center, the way you sort of organize yourself, you have to have a crane there that lifts the compute the the equipment up, etc.

So it is quite complex to operate. So it hasn't been successful in terms of large scale adoption. And our our pos our sort of solution is different from a from that perspective that it uses the same way of cooling, which is immersing the components into liquid, but we don't use full immersion. We use a small amount of liquid. That's where the precision worth comes from, because we precisely target components inside of the system with this same liquid. It's super efficient in terms of heat capture, but it doesn't change the form factor of the device. or the structure of the data center is still RAC-based. So it's each individual chassis that has a server has its own liquid and its own loop that it circulates in.

So it is different to what you've just just described. And that's where we see sort of you know our benefit as well because we we basically don't change the way the operator is is used to maintaining and operating a a data center.

Mark Fielding it'd be interesting to do a show on the importance of Bitcoin mining in the development of AI data centers because it was responsible for a lot of the the trans transition to GPUs. It was used early days in that and here

Simon Jesenko Correct? Correct.

What dielectric liquid is

Mark Fielding as well. you you both keep using the word liquid and then immersing immersing immersing Yeah. your chips in liquid or if it's precision directed liquid? What is this? liquid. How can non tech people understand what this dialectic the electric liquid is sure but think of it as dialectric liquid is? And where's it come from?

Simon Jesenko think of it as a highly refined oil which is which is environmentally safe, human safe, and it's manufactured by all the the oil and gas companies of the world. All the big names mana

Mark Fielding Is it a highly refined oil then? If I think about it as a highly refined oil, because it is a highly refined oil?

Simon Jesenko Yes, it is. It is exactly that.

Jeremy Gilbertson What's it made out of?

Simon Jesenko Well, you have now different versions of it.

Mark Fielding Oil.

Simon Jesenko Some of it is yeah, some of it now it's it's it's it they're actually starting to make this sort of oil which is completely sort of natural, which I you know I don't know the specifics of it really. not being an expert in this field, but there are there are some now some options of this that are That are even more environmentally friendly than the normal sort of oils that are, you know, that that all these companies are making.

Mark Fielding How much would a give me a give me a a a a power source? How much do you a ten gigawatt, is that right? A ten gigawatt data center, how much would it need?

Simon Jesenko How much oil would it need with our solution?

Mark Fielding Ham Your Well, yeah, you're both.

Jeremy Gilbertson Maybe look at it by Iraq.

Simon Jesenko well well I would I would

Mark Fielding Yeah, rack.

Simon Jesenko I would tell you by sort of on a server level so we use approximately between 15 and 20 liters of that oil per server. So then you can

Mark Fielding Okay.

Simon Jesenko multiply by the number of servers you have in the rack, etc. etc. But by chassis, it's around 15 to 20 liters that we would use of this oil. And this oil doesn't leave the chassis, by the way. It it's constantly circulating in a loop. And it's taking the temperature away and it's basically connected with the infrastructure, the water loop of the infrastructure. The water then takes the temperature from the from the oil away and and into the infrastructure. So that oil is com is constantly circulating. And we've seen years and years of this oil being operational with no degradation, no issues for the you from the units that we that we have installed.

The oil companies behind the fluid

Jeremy Gilbertson here's a question. So if it's if it's kind of an output of like the exons of the world, how does that factor in Mm-hmm. Mm-hmm. to like if we were to look at it holistically, not just, you its use in the data center, the actual use of that as a piece in the equation of things Yeah,

Simon Jesenko I think you can look at it as, you know, they are the polluters of this world in a way. And they're giving back some of those credits, if you will, by by producing this sort of type of liquid that helps save energy and save water consumption. So it's a you know it's a from from their perspective, policy perspective, if you think about it's how many carbon emissions do they save by doing something like this. So it's an important for it all of these fluid vendors that we've talked about and we've we've worked with a lot of them, the likes of Shell, Exxon, Petronas, etc. It's important for them to have an A positive impact on the environment, considering the bad reputation that they're getting in terms of what they do in principle, right?

So for them, this is an important part of their strategy for sure. Introducing those types of liquids, helping the data centers be more efficient in terms of consumption of energy and the consumption of water. That is for them also an important part of their sort of ESG, if you will, policy and how they look at things, right? Yeah.

Jeremy Gilbertson So it's not like a it's not like an extra output over the traditional refining process. It's a totally different, totally

Simon Jesenko No, different.

Power supplies and the edge

Jeremy Gilbertson different thing. Got it. Okay. all right. So you mentioned the the HP thing was a long time ago. Do you have any more recent examples or could you talk to us about how the adoption is going? Because you have to go to these OEMs. They have to believe in supporting it, and they also have to convince their the Yeah. people that are buying it that it's it's the right move too. So give us an example of a couple of new ones that you're working with.

Simon Jesenko I can't name name names in terms of the OEMs that we're working with, because we're under NDA, but I can tell you a couple of examples. So what we found that is that our technology has a benefit in when we looked at the data center structure and data center world. And I mentioned before that data centers when it comes to liquid cooling are moving into direct to chip technology first, because it's the most most comfortable technology and it's something that they've settled on. But when we looked at the data center and and sort of every the whole the whole sort of IT that sits inside of a data center, we've identified the power architecture, the power rack to be a focus area for us and our technology.

And why is that? Because certain components, certain IT components that sit inside of a power rack are very complex in terms of cooling, how to cool them. So you can't just literally apply a cold plate on those components. You can't get the benefits of the cold plate as you do with a GPU or a CPU. So we're focusing on the power rack as one segment or one pillar with our technology, and more specifically on the in rack power supply unit. So those are the units that basically distribute the power to the compute. And what we're seeing there is because of the device complexity, is we're getting a huge benefit using our technology versus direct-to-chip.

And what that's gonna do from a business case is it's gonna basically unlock that extra power that normal traditional PSUs use when they're air cooled and is going to release that power into compute, which means more more dollars per kilowatt for the the operator, right? So it's a very, very good example of where our technology has a important play today. And then we are looking at the other elements of the power rack as well. The battery backup units are one which we believe are s have similar issues in terms of how to cool them efficiently or how to cool them with with the direct to chip technology.

So all when we are we are looking at where we play and our technology plays today, it's always comparing ourselves to direct to chip, right? So where do we believe our technology has a benefit versus direct to chip? And this is sort of you know in the in the data center, that is the the area that we're focusing on power. And then outside of the data center, and this is becoming more and more apparent and more and more important, is the edge deployments. Edge deployments, where people want to do want to basically do AI inferencing at the edge, they want to process information you know, at the edge, and they need powerful, more powerful compute.

They don't have the infrastructure needed to to basically bring that power to the edge, is our our sweet spot. And then there's other elements at the edge. That we solve for, things like noise. If you want to have compute in the office, because if it's running and it's air-cooled, it's gonna be super loud. Whereas with our technology, it's completely silent. If you have a a sort of area where you're trying to deploy compute that is very dusty or humid, again, you know, all these air ventilators bring in that dust and humidity and it it destroys the the the the IT. So again, our solution is completely sealed.

is able to operate in those environments.

Jeremy Gilbertson The power piece is interesting, Mark, for this reason. Just like in data centers, water's coming in and running through these entire systems, power's coming in and running through these entire systems of that comes in comes in from a utility, goes through a a generator farm likely, it goes to UPSs and through all of these things you have losses. And then the ratio is like how much power the IT stuff is using versus how much power the building itself is using to support the IT system. So you're talking about power distribution, which is down along the chain there. Focusing on the power rack is the idea to make that more efficient so you can actually use more power to power more things.

Simon Jesenko Correct. Like

Jeremy Gilbertson talk me through exactly what you're doing to solve at the PDU level.

Simon Jesenko Correct, it's it's the PSU level. Correct. It's it's basically to make those more those components more efficient and use more power less power to to to cool them and more power can be then distributed to compute. You know, so it's so it's it's an it's a numbers game. When you look at all of the the the technology that's being deployed today, it's all about speed. And then the second phase that's gonna come in those deployments is gonna be efficiency. So this is an efficiency game, right? How do you make the deployment you have as efficient as possible. How can you get as much revenue per per kilowatt as as as you can as an operator, as somebody that has that compute sitting there, And you have a you have the power distribution and everything else connected to it.

All the elements matter. Even it's if it's a small difference in terms of that power efficiency, it makes a it it basically accumulates when you look at the full deployment, you know, from a data center perspective quickly, right? So that's what we're looking at. Those small gains equate to big, big numbers when you look at the the data center as a whole. So that's that's

Jeremy Gilbertson Is it

Simon Jesenko the focus really.

Jeremy Gilbertson is it removing heat from the PSU or is it actually making the conver okay.

Simon Jesenko It's removing it's removing heat from the PSU but in more efficient way. So, similar to the problem that we talked about earlier, you know, we have PSUs that are being air cooled today. Those PSUs are absorbing a significant amount of energy in order to be air-cooled. So you want to make sure that you don't use that energy for cooling, but you want to distribute it to compute. It's the whole

Jeremy Gilbertson So we're pulling the fans out.

Simon Jesenko yeah, it's correct. Exactly that.

Jeremy Gilbertson Yeah, yeah, yeah. Got it.

Simon Jesenko You're pulling the fan out. You're putting a pump in, using our liquid, and then it's you know, you're cooling it and it's z very efficient and you can then transfer that power that you use to cool that com device to compute. That's basically the the the game.

Is there a ceiling?

Mark Fielding Nice. I'm I'm glad my question was followed with what you said, Jamie. I had a very basic question that was based on that. Yeah, I was thinking of my computer overheating and cooling it down, make it more efficient by using this. Is there a a limit? So how much more powerful as as the as the chips get more powerful, they need more cooling? Can this method does it have a ceiling?

Simon Jesenko We we haven't we haven't actually it's a good question. We haven't actually reached that ceiling yet. We've done a lot of tests in terms of the the GPUs that are available on the market today, and all those tests have passed. So we don't really today know what the ceiling of this technology is. but we believe you know it has definitely longevity and is sort of if you will, future proof in that respect, because it has because it's based on the system as a whole. you you know it's not you're not just talking about the chip. You can you can really sort of get the most that you can out of it.

Now there's another element to what to your question that I would like to answer if you allow me to. What we've also been developing is, you know, from a system cooling perspective is a hybrid where you have direct to chip technology sitting on the chip and you have our technology for all the rest of the components. And we see that potentially as being also very useful considering where the you know industry is going, which is direct to chip first and our technology second. So we have that option as well for the user to to go into if they want to. So we have we have thought about you know what the next steps of let's say the liquid cooling sort of curve, development curve is going to be and where the industry is going and how we can how we can accommodate.

Why power equipment is easier to cool

Jeremy Gilbertson Are the power OEMs like the UPS OEMs, the PSU OEMs easier? Is it easier to integrate your stuff into that versus trying to integrate it into the IT systems? And and if so, why?

Simon Jesenko Well it's easier, yes. I would say yes, it is easier to integrate. I think just the complexity of the IT system of the server, right, is is much more complex. It has many more components. It is much more difficult to I would say build up with our technology. It's of course we we we can do it because we've done it across a number of different platforms on the server side, but the power supply units are more, I would say, similar. when it comes to sort of the inside and ju the the components that sit inside, although they're you know they can be allocated slightly differently across the PSU unit, they're much more similar to than versus a server.

So it is easier to cool. It's easier to adopt and it is easier to

Jeremy Gilbertson Plus the electrical engineers understand the cooling with mineral will yeah,

Simon Jesenko Plus a hundred percent. When yeah. it comes to electronics and when it comes to different components, the PCB and e everything else, it's much easier to to to do it in those units than it is on a server level.

Jeremy Gilbertson Mark, he should talk to Danny Andrev about the power circuits he's putting up in space. That could be a that could be a fun follow up.

The handover question

Mark Fielding Well, th there's many links and threads and common commonalities between every show on Thinking on Paper, Jeremy, as you know. So yeah, electronics in space matches this. Maybe our episodes on StarCloud and data centers in space, or our moon infrastructure programs, or maybe our quantum computing and that energy consumption and how they are using cooling There's some very connected ideas to think on. it's The time for the handover question. So every week we ask our guest to leave a question for the next guest and ask them to answer the question from the last guest. And I think you'll like this question from last week's guest, Nathan Silvernail, former SpaceX engineer who turned his back on the space industry to make wood out of grass.

As you do. and his question for you his question for you, Simon, was what technology do you see that is antiquated and that we could absolutely make better if it was not for capitalism or whatever other structure that is propping it up?

Simon Jesenko That's a that's a that's a good question. what technology? and you know what what I would say to answer this question I would say when because we're talking about liquid cooling and we're talking about what we have right and what how how it changes how it changes the world and how it changes all this this sort of the impact it has. I would say, you know, we should be really doubling down on on liquid cooling as a technology of the future, not just for the reasons that I mentioned just now, which is you know saving on energy, saving on on water, saving in savings when it comes to you know the space that that these these data centers and all these sort of installations use.

It's also you know. using it for capturing heat and reusing that heat for other applications like heating homes, heating facilities, etc. So this technology really is something that provides so many incremental benefits to the to our everyday lives that we are maybe not aware of or have not been have not been aware of, that it's something that we simply cannot, I would say, avoid anymore. And I would say this this would really needs to be adopted at scale because it has so many secondary benefits, you know, on top of the primary ones that I've mentioned before. You know, heat heat recapture and all this heat that's that's being generated by these data centers, if you think about it, what a waste, right?

What a waste that is just suddenly just absorbed and goes into the ambient and it's not reused for for anything else, right? And we could use a lot of that heat to and do a lot with it, especially considering where we are today, where you think about where the world is and how expensive certain things are and how they are moving, right? So I would say, you know, liquid cooling, if I'm a little bit you know, biased in in that in that in that answer is definitely the technology that should be adopted today and should be scaled fully and implemented in full.

Mark Fielding Love it. I'm gonna reverse that question for you, Jeremy, and say what technology are we using today which isn't as good as an older technology? What a a technology is that replaced an infinitely better technology. I was thinking of MP3s and

Jeremy Gilbertson Scape skateboards.

Mark Fielding skateboards. Kid. What's replaced skateboards? E bikes, electronic scooters.

Jeremy Gilbertson E dude, the e bikes drive me crazy. E bikes drive me crazy. No one pedals anymore. Why don't we pedal anymore?

Mark Fielding No one. I think it I think it's indicative of the the world's problem at large that nobody pedals anymore, actually. I think you might be on to something there. Simon, do you have a question to leave for next week's guest?

Simon Jesenko Sure, I will I will use a a question that I think is relevant in today's day and age. So and I would say which activity in your day-to-day life or action in your day to day life has been l replaced by AI? And to what extent, and would you go back? Right? Would you would you would you go back? to how it was before.

Mark Fielding Would you go back is that's an interesting part of that question.

Simon Jesenko Yeah. Yeah.

Mark Fielding Thank you. We will ask it to our next guest. And if you want to know who our next guest is, then you'll just have to tune in next week. But it's a good one. Thinking on paper dot xyz, spotify, YouTube, apple, iHeart Podcast, wherever you're listening, like, subscribe, share it with three friends who need to hear this conversation. And until next time, be disruptive. Stay curious.

Jeremy Gilbertson Keep thinking on paper.