Thermal Breakthrough, with David Tearse (CEO of Karman Industries)

Nice to be here. Thanks for the invite.

Yes, I have my four-month-old Bernese Mountain Dog puppy here in my office. So, I wanted to give the viewers here a little preview of what maybe is some background noise during the podcast here, but yes.

That's right. I was thinking I was like I could put the little line on the wall as he keeps growing taller and taller, but we are opening our new facility here starting this summer. So sadly we'd be moving out of this space and we would have all the little height lines as we move out.

Yeah, that's a good one. I mean, I like, I am definitely an engineer at heart, but I would say I grew up using Legos and I just love building stuff, and I'm not the best at actually building the stuff. So I brought a great team around me to actually go execute on that vision, but at the end of the day, I just want to build cool things with good people that have a positive impact on society.

Yeah, I mean, I do think that we were in this long period of like great technical innovations, but everything was improvements. You had like these Moore's law type technologies improving, but you didn't have like things that were fundamentally unlocking certain new categories. I think SpaceX definitely did that. And but now when we actually think about how long ago that really was when we could start putting commercial satellites up into space, you know, it's now been maybe a decade plus of real reliable launch capabilities. And so there was many of these years where we just saw things keeping incrementally improving and life was getting better and technology was advancing. And then I think we saw this crazy unlock with ChatGPT just kind of started to scratch the surface of what was possible. And now we're starting to really see that accelerate. I still think we're very much in the early innings, but this is going to be one of those fundamental changes in society. When you start looking at the steam engine, electricity, these big inflection points, I think you're going to see that, you know, 10 years from now, looking back, this was a big moment in civilization.

Yeah. So after Skyrise, I was an entrepreneur in residence at Riot Ventures and looking to build something within the hardware space, and I met my eventual co-founder who at the time was head of technology at Antora Energy, which is a big thermal energy storage company in the Bay Area. He had worked at a couple startups prior to that working in the kind of thermal energy space, and then he actually started his industrial career after his PhD in Princeton at GE Research working on supercritical CO2 turbomachinery. So compressor work. And what we started to evaluate was, you know, 50% of all end energy use goes towards thermal management. We're either heating something up or we're cooling something down. And we do it with pretty antiquated tools to do that today. They're inefficient. They waste energy. And we looked at that we could go and build that next generation of thermal infrastructure. And how we were going to do that was we were going to pick better Lego building blocks and we were going to leverage the tens if not hundreds of billions of dollars spent in the aerospace industry, the EV world and advanced manufacturing, and we could bring this all together in a brand new package really. And so that's the kind of insight that we saw was that there was such a massive opportunity for efficiency performance gains across the thermal landscape, and it was something people were not paying attention about. We focus all on power generation. We don't think about the actual thermal side of the equation.

I think that today we view heat as something that we want to discard. We either spend a lot of energy to remove it from situations or we spend a lot of money generating it for industrial applications where you need heat. But in each one of those situations, we do it very inefficiently. And back to the stat of 50% of all end energy goes towards that, that if you can start making improvements on how we actually manage the thermal applications, I think you can start unlocking massive changes across almost everything we touch. You know, from food to our clothing to the compute inside our phones or our computers or the models that we're using. All of this is governed by the laws of thermodynamics. And so being able to make improvements towards how we manage that heat becomes a big unlock.

So that name is actually after a famous physicist, Theodore von Kármán, who had a lot of very influential work in hypersonic and supersonic flow. So it's all aero, and we are, you know, compressors and turbines, which is what we do inside our system, is all aero profiles. Our systems are moving at 30,000 RPM. So a lot of, you know, close to supersonic speeds in there. So it was my co-founder actually came up with the name. I added the Industries part of it at the end of it. Maybe they'll make a movie one day and they'll say you should drop the Industries piece of it, but you know that that is where the name came from.

Yeah. So when I was in entrepreneur residence, the first half of my time doing that, I was going out and trying to figure out a problem that I wanted to solve. And then I assumed that once I found that problem, I would go and find a co-founder that could help me do that on the technical side. While I have a technical degree, that was definitely not my strong suit. I was not going to lead the whole engineering work. And about halfway through that process, I realized that wasn't quite the right framework to be going after. So I inverted the process. I started saying I want to go find the smartest people in a lot of different industries and talk to all of them because for the most part all your low-hanging fruit has been done. Probably your medium-hanging fruit has been done. To truly have defensible products with a large potential outcome, you need to have industry and world experts working on that problem to uncover where the opportunities are. Pretty quickly after I reversed that, a mutual connection of CJ and mine, I said, 'Hey, you know, I'm looking for a potential co-founder and we're going to ideate on things together.' And he said, 'Oh, you got to meet CJ. He's one of the most brilliant people I've ever met.' And we just kind of hit it off right from the beginning. And I think one of the big things that drew me to CJ was of course a very, very intelligent person, but also had so much experience working in this industry that he knew all the mistakes that people have made previously. He knew all the traps that people get down and really had the nice blend of what is possible versus what is practical. And I actually reflect back on like what makes us a good pairing as well is actually a Ben Horowitz talk that he gave at Stanford saying about him and Marc Andre when they started Netscape. And Ben is like, you know, I was the entrepreneur and I could lead the business side of it, but I also, and then Marc on the other side was the world-class engineer that was going to invent something that no one else had ever invented before. Now they didn't stop at those silos though. Marc also had quite a business mind and understood that side of it, and vice versa, Ben had a computer science degree, could understand what was actually, you know, Ben could understand what Marc was actually building. So there was a good enough Venn diagram overlap, but each had their kind of core competencies. And I think that actually speaks a lot to what CJ and I have. I'm shocked all the time at, you know, CJ can be sometimes one of our best salespeople from both a recruiting standpoint as well as I bring him into BD calls and he sells the product very well. And on the flip side of it, you know, I run mostly the business side of the company, but I do have an aerospace engineering degree, so I understand what he's building. I don't need to get into the nitty-gritties of it. And man, when I go into, you know, when I go into design reviews, I'm like, man, everyone in here is so, so smart compared to myself on the technical side.

First off, turbomachinery is a very challenging field to be in. When you look at SpaceX and the advancements that they've made on a propulsion side, that is all turbomachinery. And you probably have seen the Raptor 1, 2, and 3. Massive simplification between those motors. And there's tons of money poured into that. And your top top engineers at SpaceX end up being your propulsion engineers. Now, the ones that have invented a lot of the original Raptor turbopumps, you know, they're probably long gone out of SpaceX, but it's a very, very small field of top people that understand turbomachinery. So when we started the company, we actually doubled our option pool to try to bring and attract highly experienced, highly capable people that could roll their sleeves up and give them a meaningful piece of the company because at the end of the day, like we want to have people that are here for the long run and are motivated. So I think that there's just such a small amount of people that actually can do the turbomachinery side of this, but then there was key technology unlocks as well. One of them being high-speed power electronics, silicon carbide, permanent magnet motors. That was a big unlock. And now the technology's been around for call it a decade. But now as Tesla and Rivian have built out massive supply chains, performance has gone up, cost has just come off like a cliff. So there was right time, right opportunity, and then the rise of AI, the right customer on the other end as well.

Yeah, so what's really interesting is first off, we all know the scale that they're building out at. Each one of these companies seems like they keep saying that they're going to spend, you know, 10, 20 billion more than the other ones whenever they report earnings. We went from, you know, 160 billion, which is, I think, what Google was planning to spend this year, to all the way AWS most recently last, earlier, sorry, last week, saying they're going to spend 200 billion this year on it.

Yes. 30?

Yes. Yeah. It's pretty crazy when you see scale and it's all concentrated to, you know, your big hyperscalers, your neoclouds, your collocators, and primarily most of that spend is happening domestically as well. But what is different is it's not like we're just building the same data centers we did three years ago and it's just a lot more of them and we're building a lot of it. What is actually, it's a totally different data center. All of our previous ones for the most part were air cooled. You didn't have the power densities in them. You could cool them with traditional air conditioning. Then you had the rise of these GPUs and you switch immediately over to liquid cooling. 98% of data centers getting talked about and built today are all liquid cooled. And then you have Nvidia refreshing your chips every 18 months. And every single time it's Moore's law of just densification, densification, densification. So now you're packing more heat into a smaller area.

As well as that, the campuses are getting bigger. We're not building 50-megawatt data centers. We're building 500-megawatt. We're building multi-gigawatt sites. So more power going through a smaller amount of area just creates completely new problems from a thermal management standpoint. And when you look at the infrastructure, the existing cooling solutions out there, they were not designed for a gigawatt-scale campus. And so what we brought to the table was systems level. How would you move a gigawatt worth of heat if you went from a clean sheet design, first principles? And spoiler alert, it doesn't look anything like what they did previously.

All the electricity turns to heat.

Yeah, we reject it to ambient today. So we call what we launched the heat processing unit, an HPU, and that's meant to be an ecosystem of products because depending on your ambient conditions, depending on your operating conditions of the data center, depending on your workloads as well, you may want to process that heat differently. Today, every single customer needs chillers and dry coolers. We are going to remove that heat the most efficient way possible, and you can do that with our system. Now when you start looking at the future where you have high temperatures coming into the chip and you have even higher temperatures coming out, now you start opening up doors for heat reuse as well, whether you actually push that to district heating and cooling or you start doing power generation with that as well. So really we want to be that, I would say, the brain or the node that is what you do with heat, and depending on your configuration you may process that differently.

About a third of that energy is going, or the power to that site is going towards cooling. So it's a massive number. So when you think of a gigawatt campus and if it's pulling a gigawatt off of the grid, you do not have a gigawatt worth of compute. You have probably two-thirds of that going towards compute, which that's your revenue-generating side, and then the other side is opex. So if you can effectively move that needle, which we are able to do with our system, is called your peak loading condition, what is your worst design point you need to be able to cool on your hottest day running full bore. When we can start moving the needle and the line in the sand of how much power is going towards compute versus how much is going towards your cooling system, that starts to unlock more compute, which then means more revenue.

Yeah. So we're kind of high single digit up into, you know, 10, 12, 15% more power can go towards compute. So when you're looking at a gigawatt campus, if I can unlock 100 more megawatts towards compute, that's a big revenue jump up on these systems.

So we actually picked CO2 for its thermodynamic properties. It has some very unique things that are different than the HFOs out there. One of the main things is that in the temperature ranges and the pressures that we're running at, it's running in either transcritical or supercritical. And what that means from like why it differentiates us is that on the expansion side of your refrigeration cycle, there's still a lot of potential energy that you can unlock. So we actually run a turbine that allows us to recapture some of that work being done by the compressive side, and that's what gets our performance delta on the system. It's also a very energy-dense fluid. So instead of, in a standard shipping container for a chiller, instead of two to three megawatts, which is kind of the highest level of what the industry produces today, we can jump up to 10 megawatts. So again, that really reduces that outside infrastructure piece. The knock-on side effects of CO2 though is that it has a global warming potential of one. It's a natural refrigerant. It does not form forever chemicals over time. It's nonflammable. All of those we felt like were going to be nice-to-haves for customers, and maybe their sustainability people would write an article about it like, hey, we've moved over to this, but that wasn't going to be the main selling point. And we've been very surprised at customers latching on to that as strongly as they have. Now, typically, it's not from a sustainability straight-up standpoint. What it is is like, okay, we need to, we can't take lower performance. So we give them higher performance. Great, that unlocks that piece of it. What they view it as is that we're building this data center today, we want it to last for 20 to 30 years. If there's any risk from a regulatory standpoint, we don't want to take it. If we can take that risk off the table, that is great for us because maybe we can build it today, but what happens if in three years they ban it in certain locations. It's the same thing is that this is also a closed-loop cooling system. So we do not consume water with our system. So it's a WUE, which is water utilization effectiveness of zero. So we do not consume water in our system. Again, a regulatory de-risking for them because even if you can go get your water rights today, doesn't mean that the community is not going to pull that permit two or three years from now.

So we remove everything from the exterior of the building, from exterior wall out. We do all the heat rejection out to ambient. What we interface with is the existing cooling loops that go inside the data center. So we feed cold glycol loops back into, or whatever temperature you need, back into what's called a CDU, a cooling distribution unit. We feed it into there. That system inside the building feeds that to chip level. So we replace everything outside. We don't touch anything inside, but we make this very plug-and-play so that we use all the same hookups, glycol loops going into the data center, so you don't need to change anything on the inside of your data center.

Yeah, I think we've seen quite a big shift in strategy from the hyperscalers over the last couple weeks, and maybe it wasn't quite an explicit shift in strategy, but it's like public-facing, this is the strategy we're going forward with. A lot of the community pushback has been, hey, we're not going to get any economic benefit from this. Well, now you're seeing that they're going to pay their way in taxes in those states. They're going to be donating money to local boys and girls clubs and schools and universities. And you will, I think, actually see because they're going to smaller rural locations, watch those places now have some of the best schools in the country, and it becomes almost like a weirdly a destination to want to go move your family towards there. The other big key complaints that people have is around the water consumption and electricity bills. A lot of this is being done behind the meter. I saw some stat, it was a very, very high number of these are being done behind the meter while they wait for grid interconnect. And then water consumption, you've seen them all shift towards closed-loop cooling systems. So again, you're trying to alleviate all that risk of trying to go to a site. There's much more than just getting a plot of land to building a data center. There's so much that goes behind it from fiber, from power, from the community, being able to get approval for it. So everything they can do to de-risk that helps. And for us, it's like, great, you can be a smaller footprint. So that helps. You can go to different locations out there with our system. We have a lower noise signature on our system as well. We don't use water. We don't use harmful refrigerants. So all these things I know that our customers are going to market pretty aggressively. I kind of half-jokingly, half-seriously tell, you know, our hyperscale customers that right now the communities are, not in my backyard, I don't want a data center here. And I go, 'Well, you guys could flip that script pretty easily and say, let's add a half a percent to our capex of the total budget of this data center, we'll build a thousand homes around it and we'll pipe all the heat into it using an HPU and we'll give people free utilities.' Now, all of a sudden, people really want to live next to these systems. So I think you'll see an interesting potential migration of people out of bigger cities to these hubs because there's going to be so much economic value derived where these data centers are. And then if you build...

communities around it, I think you could see quite a big shift from a geographical location of where people live.

Yeah, so throughout the fall, we were pretty stealthy about the product that we were building. If you went and looked at our website, it was pretty much not what the data center focus was on there. And we were going to conferences talking to people there and starting to share what we were working on. The official launch happened January 15th of this year, and since then the reception has been tremendous. And we went from, you know, trying to sign the lease on our facility of, yeah, we should maybe do that in like a month or two for our production facility, to we need to do that right now, to is this facility too small, and the scale that everyone is talking about is just ginormous.

And what I think is the most fun part of my job right now is that I am tip of the spear on sales right now. We don't have a salesperson. It is me. So I'm talking to all these people and within the first five minutes you can say I've said enough to where they're pretty convinced on it. We typically then will flip into, you know, an hour technical discussion. By the end of it they'll bring on 10 engineers from their thermal teams and they will all get exactly what we're doing. It quickly goes then from, okay, we like everything you're saying. This unlocks a lot of, you know, pain points we currently have in our build and construction timelines and performance. Now the question really becomes to us is, can you keep up with us? Can you keep up with our demand? The facility that we're specking out right now, one customer could take that entire production capacity. So the scale up is actually where I put a lot of my focus now is we need to be able to scale to meet that demand.