In The Arena by TechArena - Cooling the AI Heat: JetCool’s Strategy for Data Centers

Episode Date: November 13, 2025

Discover how JetCool’s proprietary liquid cooling is solving AI’s toughest heat challenges—keeping data centers efficient as workloads and power densities skyrocket.

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 to the arena. My name's Allison Klein, and this is a Data Insights episode coming to you from the AI Improsite in Santa Clara. Data Insights means that I am here with Janice Norowski. Welcome to the pro-redgenies. Thank you, Allison. It's great to be here. And we are having such an awesome week in Santa Clara talking to practitioners and developers of infrastructure across the AI continuum. Who do we have with us today? We have someone literally and figuratively cool.

We have Mr. Scott to me. Welcome, Scott. Thank you both very much. Thank you for joining us. Scott is the senior director of business development for Jet Cool. No, Jet Cool has been on the program before. but Scott, you have not. And liquid cooling has taken flight in a way that I didn't really even think was possible.

Jet Cool has a really interesting differentiated solution. That's a central element of that advancement. Can you take a few moments to just introduce Jet Cool and a little bit about how you were helping propel these solutions in the marketplace? Sure. Thank you. So almost a year ago, Flexronics had acquired Jet Cool. Jet Cool is a pioneer in the direct chip liquid cooling arena, single phase. And so we're combining Flectronics as global manufacturing scale to help accelerate the adoption of Jet Cools technology with customers across data center and enterprise and communications.

Excellent. And today's AI workloads are really driving higher compute density more than ever. What new thermal challenges are emerging as a result, in your opinion? I'd say you hear a lot of the hypers talking about rack power densities going to one megawatt, which frankly even a year ago, think people said that won't happen for a while. So that clearly is leading to heat as well as the power and obviously the cooling associated with that. So one of the nice things that we see is leveraging Jet Cool's proprietary technology to help address those concerns for the higher power chips and the heat that they provide and then bring in and layer in some of the additional

services that Flex has in our portfolio to help support that. No, I know you're talking to a range of customers about the solutions that you're delivering. Are there common themes in terms of these cooling limitations challenging your customer's ability to deploy advanced AI across training and imprints? Yeah, I'd say two that come to mind. The first is, and it's similar for both training and inference, is the heat flux. And for those that don't know, really what the heat flux I mean by that is, in a given chip, a GPU chip, a high-end custom ASIC chip, the power density can vary depending on the workload. And so the ability to have a solution that can adapt to that changing workload

and the heat generated off that is key. Jet Cool, of course, can support that. Others cannot. I'd say then the second one is traditionally cooling and infrastructure and power many times at a customer level are handed by different teams and you really need an integrated partner that can bring that all together because you're not just solving for the one and again, a value add that Flex brings to the table

is that we can be that partner to have that combined conversation. Amazing. And Jet Coal has developed a really unique liquid cooling solutions, you know, different than anybody else in the market, but what makes your approach different from traditional air and immersion cooling

methods? I'd say so our proprietary jet nozzle jet impingement technology enables us to find those hotspots on a given chip processor and to have what I call a bespoke solution that will provide the greatest amount of cooling, therefore the overall chip is operating more efficiently. Whether it's immersions or even microchannel technologies that people may be a little bit more familiar with. Those generally are kind of a blanket. I'm just going to throw liquid across the chip and the chip's going to feel better. You get some level of efficiencies, but now that the wattages or the TDPs on chips are continuing to increase, they're running out of their bandwidth and we're just starting to gain our momentum. Now, one of the things you've talked about GPUs,

you've talked about high-end A6. One of the things that we know is compute density and the rack continues to grow, and that's challenging this environment even more. What do you see in terms of requirements from customers in terms of power, cooling, and reliability in these dense racks, especially when you consider the amount of investment that's going into these dense racks? And how do you see this playing out in the hyperscale as well as enterprise? Great question. It gets a little bit about what I said a minute ago, but this idea of it's no longer a single issue or topic you're solving for. And I'll give you an example.

Again, whether it's a GPU or an ASIC, there is the heat profile of that and a solution to solve for that. But you also have items like voltage regulation modules that also are converting the different power to support all the chips that are surrounding the processor. Those clearly, because all the power is coming in, generate a lot of heat. And so I'm starting to lovingly call it a safe.

sandwich solution, you will have a cold plate or a jet cool smart plate on top of the processor. These voltage regulating chips sometimes now have to go on the bottom side of the board. You need to now have a second cooling solution on that so that holistically, that whole solution, what we call vertical power delivery, that whole solution is happy versus you just address the processor, but now you're having other challenges. And as we know with anything, you know, now move into what is the next thing. And it could be memory, it could be hard drives. It could be storage cards.

Again, back to the holistic approach. We work with the customers to look at an end-to-end solution that we feel we are uniquely positioned to address that for them versus other competitors. It's amazing. So can you share an example of a deployment where you've improved cooling directly, enabled better efficiency and or performance? We work with a number of the hyperscalers. out there for confidentiality. I can't say who they are, but they're the who's who of who we all

know. And what we say is within the flex portfolio, we cover about 80% of what a data center customer would need. We have the products that we've talked about, the Jet Cool, we have a flex power division, we have mechanicals capabilities. So we've sat down with these hyperscalers and said, we need to look at the design one of two ways. You either look at it from the grid, working your way back down to the chip because it's all going to work together on power and heating and cooling or the reverse. We're okay, but cost we need to look at it. Generally now they are looking at it from the chip, working their way out to the data center out to the grid because there's so much power requirement that it also gets back to how much power are they going to be able to get into

their own facility to support the growth and expansion of AI. And by working together, that led to more conversations around you need to decommission some of the current infrastructure you have in place at your data centers. Again, flex from our global services and solution offering, we're able to sit down at the table and talk about how we can assist that same hyperscale or customer for those types of solution. So I'm not personally aware of anyone, but I tell most people, it's the only company that I know in the world that has that full range of capabilities to do all of that. And we haven't even address probably flexes historically core business of pure EMS manufacturing, which obviously we can do for the customers that need that.

You know, you said something really interesting that I want to double-click down on a little bit ago. We started the conversation with GPUs, but then you said something that really perked up my antenna, which is heat is then going to other places in the platform that need to be addressed. Can you tell me about the state of the state in terms of addressing different components on the platforms, and where do you see this going? So A, I know it won't stop because we're going to be getting parts of the whole AI journey, which is great. But where I was first addressed to this was high bandwidth memory. And there's high bandwidth memory on the processors. And so now you took care of the processing brains and those cores have been cooled and they're

happy. Immediately, typically the high bandwidth memory is the next area that customers will want to talk about, again, whether it's a GPU-related conversation or a customer. basic type solution. We have another customer where we addressed cooling their processor that they then came back and said, these two chips that are a little farther out on the printed circuit board, can we go get those cooled? And so we had to make a bigger solution for them, but they're happy because now holistically, their entire solution works better than if we had just cooled their main processor. They didn't have anybody else they could find in the world that could do that for them. Wow. How are you guys, you know, Jet Coal working within the industry to kind

ramp solution development. And where are we in 2025? I'd say through, again, the acquisition from Flex, we're ramping the global manufacturing capabilities. A lot of people were very impressed with the Jet Cool technology. And probably the biggest hurdle to overcome was as a newer company, the ability to scale. Obviously, now Flex with our 100 factories and 30 countries around the world can easily address the ability to manufacture at scale. but I think the other key component that's come out of all the conversations we've had is it's not just providing the smart plate, but we also provide quick disconnects, we provide CDUs, and we provide manifolds. Why do we do that? Is that customers have asked us for that

holistic solution because they don't want to buy from four different vendors, and whenever there may be a problem, the four vendors are pointing fingers at somebody else, they just want the problem fixed. I say with us, if there is ever a problem, and there many times is not, it's one phone call and we're there to take care of it. Now we're at AIMFRA, one of the things that we're talking about, we're in the gigawatt era, and this is only getting bigger, it's only getting hotter. There's no way around it. How do you see the technology evolving and are Jet Cool's direct solutions having the

headroom to address the coming heat in dense gigawatt range data centers? Great question. You're absolutely right. We know this engineering won't stop. We're not done. It always has to get better, faster, more capacity. And we do. One of the many reasons that we looked at Jet Cool from an acquisition standpoint is that they did have a portfolio of solutions to address the here and the now as well as the future.

Let me give a couple of examples. In our smart plate technology, today we can probably get up to right around a 3,000 watt TDP processor chip that we can cool. Obviously, not everybody has that, but there are a few people. The next evolution beyond that is what we call smart lid. That can get us up to 5,000 watt TDP. And once we get past that, and again, not too many people are even there yet, but there are conversations, is what we call smart silicon. And now this is the real high-techy stuff,

where you're now embedding a layer of liquid cooling inside the CPU or the GPU before it gets encapsulated as a finished product. So now the conversation not only is with that end customer, again, and an NVIDIA and an Intel or an AMD, but it's also now with the Foundry Service because we're another slice in that overall solution. And again, customers like that because they can see a roadmap of future possibilities

that other technologies just don't have. Amazing. I think our listeners are going to want to hear more. So where can they go to learn, more about the solutions you've talked about today. Thank you. Yes, obviously, if you go to Jetcool.com, it has all the information that I've discussed here today. Again, for those that may be here at the show, we're here at the show. We have a booth. And then next month at the OCP Global Summit, we will have the full offering on display Jet Cool, as well as some of the

additional flex sister companies that we have. So please come by and stop to see us there. That's fantastic. That wraps another episode of Data Insights. Janice. It was fantastic. Thank you so much for getting on the show. Thank you both for having. Thank you, Alison. Thank you, Scott. Thanks for joining Tech Arena. Subscribe and engage at our website,

techorina. All content is copyright by Techarena. Thank you.