theCUBE + NYSE Wired: AI Factories - Data Centers of the Future | Duncan Abbot, Zenbul Labs

Clips
More from theCUBE + NYSE Wired: AI Factories - Data Centers of the Future

play_circle_outline NIMBY Backlash Over Data Center Low-Frequency Noise and Failed Barrier Walls

play_circle_outline Low frequency hums cause sleep disruption and vertigo, sparking class action lawsuits and stalling data center projects

play_circle_outline Site-Specific Tuning of Fan Low-Frequency Tones with Zenbul Labs Resonator Metamaterials

play_circle_outline AI-Designed 3D-Printed Metamaterials for Drones, Airports, and Urban Noise Reduction

play_circle_outline Retrofit duct silencers that allow airflow while stripping sound

Info

Duncan Abbot, Zenbul Labs

search

(INTRO)

John Furrier

>> . Hello, I'm John Furrier, host of theCUBE, here at theCUBE's NYSE studio. For theCUBE, of course, we have our Palo Alto studio, where we connect Silicon Valley to Wall Street, and Wall Street to Silicon Valley. This is our AI Factory series. We talk to leaders, startups, innovators who are making it happen, building the future of the AI era, pouring the concrete, paving the roads for the AI era. This is what's going on, it's the hottest area. Duncan Abbot here is the CEO and founder of Zenbul Labs, who's got an innovative solution around helping AI factories get online faster, mitigating noise for the neighbors. Everyone's protesting data centers these days. Don't know why, they're awesome, but there are issues. Duncan, thanks for coming on our AI Factory Series. You're a hot startup, thanks for coming on.

Duncan Abbot

>> Thanks so much, thanks for having me.

John Furrier

>> I really am psyched you're here because one of the things going on right now is the protesting of data centers being built. Reminds me of the old days back when cellular technology was rolling out and towers would go up. We don't want a tower in my town, I'm going to get brain damage from that radio frequency. There's a lot of misinformation around that, things like that, but data centers actually have a problem. You're solving a problem that these big data centers have, which is noise. Noise that the neighbors get pissed off about, people are like, wait a minute, there's all kinds of levels of frequency. So it's not like, wow, it's a concert playing, it's more of a different kind of noise. Explain what you're doing.

Duncan Abbot

>> Definitely, so yeah, as you say, there's a big NIMBY pushback on data centers and their noise profile, and a lot of these data centers need to be built near urban or suburban environments, and so they're near people's homes. and the fans that cool these data centers are very loud. And not only are they very loud, but they produce these hums at low frequencies that sort of cut through walls and concrete barriers and into your home. And it's actually those hums that we're seeing cause a lot of lawsuits and moratoriums and sort of big residential complaints and it's mostly an unsolved problem and is really blocking go-to-market speed for a lot of these necessary deployments.

John Furrier

>> Yeah, it's usually after the fact, so it's kind of like there's mitigating techniques. Explain what people tend to do today because, obviously, urban areas, that's where the power is or in an area. What is the current state of the art to mitigate the noise? Do you build walls? Is it like special?

Duncan Abbot

>> It's always less complicated or less advanced than you might think, right? So it's walls generally, and depending on whether this is on the roof or on the ground, the wall is going to be different heights. But yeah, you're building a barrier, and the traditional thinking was, well, higher frequencies are what humans hear better, and so we build a wall and that blocks higher frequencies, and so we're good, but it's really brought out this revelation that it's the low frequencies that are the problem, and low frequencies bend over the walls, and so it just doesn't work.

John Furrier

>> Yeah, I'm not an audio expert on the physics of sound, but in my place in Palo Alto, we are literally two towns away from Shoreline Amphitheatre, and they have a big wall. Some nights I can actually hear the concert through my windows almost clearly. Something about the way it comes in, the physics, just the way the environment is at that time.

Duncan Abbot

>> The weather of the day might affect it.

John Furrier

>> Talk about that aspect of it, because that's something that's hard to control if you don't nip it in the bud. Is that some of the factors that go into this noise dispersion?

Duncan Abbot

>> Definitely.I think a lot of these traditional solutions, they come in, they sort of blanket it with maybe an over -engineered wall, and they don't think about how is the temperature going to fluctuate throughout the year. They don't think about what is the cloud cover going to look like, maybe what weather events might happen. So they have these issues. What we're trying to do is come in and actually guarantee a sound level at the property line and monitor all these things and react and update accordingly if we need to come in and improve the wall or improve our own low frequency dampening solution or adjust it based on that location. You need someone who's responsible for it, not just on day one, but for the next 10 years.

John Furrier

>> Ongoing. I want to get into the solution, but before we do that, explain the noise problem in layman's terms. You got a property line, you'd have homes or businesses, maybe a couple towns over if that's an issue. What is the impact of this noise problem to the surrounding environment that you guys are seeing and solving? What would people say would be the complaint?

Duncan Abbot

>> So the complaint more and more now is that 24/7 there's this hum that reaches inside your home. So you can't sleep, maybe it's giving you vertigo. I know people who actually are more sensitive to low frequency tones in your inner ear and that causes vertigo than for higher frequency noise. And because of anti -AI sentiment across the world, coupled with maybe anti -data center sentiment, this is something they can latch on to. You're sitting there, you're hearing it. Maybe you wouldn't get up and take action normally, but now you're going to get up because this is in your home. And we throw out lots of numbers. One number is that just this year, $150 billion or more of data center projects have been slowed down or canceled because of NIMBYism, right? And this is a big part of that.

John Furrier

>> Yeah, and we've all seen, we all kind of see on the Instagram and TikTok or the internet, the council meetings, they're not experts, but you got the residents complaining, wait, don't put the data center, it's going to be, make up all these things. Noise is probably one of them. There's been some class action. Talk about some recent.

Duncan Abbot

>> The first two class action lawsuits for noise at data centers ever just this year in May and July in Michigan and Wisconsin. And it's exactly this. They sort of came together in a way that I think the communities wouldn't have been able to do without all this clout and are pointing at the noise issue. And saying, we're going to stop you from moving forward or stop an existing data center because of this problem. we can just fix that.

John Furrier

>> Yeah. Yeah, well, certainly I've been inside data centers with a lot of GPUs, and it is loud as hell inside the data center. This is outside. Get into the solution because it's a unique proposition. Talk about how you guys got here and what specifically are you doing because you've got to cool the machines. There's liquid cooling. Air has got to come out, somewhere. That's the key thing. You can't just block it up. Talk about how you guys solved this, and what's your approach to helping not slow down the AI build-out?

Duncan Abbot

>> Definitely. Definitely. So, you really hit the nail on the head. The problem is you're cooling these data centers and cooling almost always is going to end up with air being rejected to the atmosphere and the sound is in the air, so to speak. And so you could cover it with a big lead shield and that would get rid of all the sound, but then you'd also stop pumping air out into the atmosphere and it wouldn't work. So the problem is you need to let the air out and strip out these sounds, especially at low frequencies. And you have to do this with different techniques, silencers and resonators, and what we've done is use some technology I developed with my co-founder to really build novel resonators that not only can strip out these low frequency tones, but can do it in a way that is targetable per site. So your site in Michigan is different than your site in Wisconsin. The exact tones that the fans produce are maybe even different if it's the same fan, depending on the site and installation. And we're able to very carefully tune it to match your fan and really effectively strip out that sort of peak that you hear at your home.

John Furrier

>> So you attack the problem frequencies, the low frequencies.

John Furrier

>> Yes.

John Furrier

>> The high frequencies can be mitigated by...

Duncan Abbot

>> The barriers work, although I think we want to be involved in the orchestration and implementation of the walls in concert with our low frequency tonal solution because then if we're going to have this SLA for residential property line sound levels, we need to be aware of and responsible for everything going into that. So the unsolved piece we bring to the table for low frequency and we oversee the solved piece at high frequency and then we wrap it in a guarantee that we manage and supervise.

John Furrier

>> All right, so I'm a customer, pretend I'm a customer. Hey Duncan, I'm doing this major CapEx build out, I'm going to XYZ town, it's in a suburb, I got good power, I'm totally stoked, and next thing you know we run into all these things. how would you engage them? Because I definitely want you to come in and take this off my plate, because I'm too busy building. What do you do? You say, hey, buy this material, put it over the fan. Do I pay for that? What's the service to me?

Duncan Abbot

>> Well, I think the first thing is, we would want to come in and establish a baseline. We'd want to come in and say, both to the data center and the residents, look, let's measure what the sound is now, and then we can prove to you that we're upholding our end of the bargain. And I think just that as a first step is something that we're not seeing everywhere unless it's required by law and shows a lot of good faith. And we follow that up with saying, okay, now we're going to commit to exactly this low -frequency tonal dampening and the higher -frequency dampening and publish that to residents. And once that's agreed to, we go in and install the range of solutions, including our own proprietary technology, and start monitoring it. So I think whether it's a new build, which will take however long it takes to finish the build, or an existing build that's starting to have pushback, we can go in and retrofit and add on top or be part of the build solution and market that from day one and not wait for it to be a complaint that happens at the last moment.

John Furrier

>> And do you guarantee measurement? Now, I'm like, okay, thanks for coming in. You helped me out. Let's just say that worked. What do you do? Do I go to your dashboard? Is there a system? What's the SLA? How do I prove that the before and after levels, I'm cool?

Duncan Abbot

>> So we are going to install, we're installing these third -party very official sound measurement devices at the property line. So it's not our device. We're not double dipping and then claiming our own measurements. And if that falls outside of the SLA, we have a tiered approach to resolution. So there's immediate, within the hour alerting, and that's very public, and within a day root cause analysis, within a week commitment to a resolution, within a month a full resolution. And then if we fall out of that multiple times in a year, then we're going to break our SLA.

John Furrier

>> So you're a service provider to the data center. And thenwe can all relate to town council, I mentioned that earlier. So the old model was some civil engineering firm or environmental engineering firm would come in, make a presentation, oh, everything's going to be good. But there was no actual actionable data. It was just more of a consultant opinion. You're actually kind of doing environmental engineering on behalf of the data center on the audio piece. That's what you guys do.

Duncan Abbot

>> And we're putting our money where our mouth is too.

John Furrier

>> And you're servicing it, so you're an always-on provider.

Duncan Abbot

>> Correct, and if it doesn't work, if we break the SLA and we're not delivering, we lose money. So I think we want to have these win-win incentives to keep everyone moving as fast as they can.

John Furrier

>> Is there a liabilities piece to this? Because sound is sound, but if you can mitigate the sound and guarantee SLA, I don't think there's much risk on the back end Because it's really more of guaranteeing to the town or city or the municipality or whatever, the region, that everything's good. You've got the before and after.

John Furrier

>> Exactly.

John Furrier

>> So there's really no, it's more of, okay, we check the box.

Duncan Abbot

>> You check the box and I think almost we are a layer of insurance that the data center can present to the town and if we're failing, then we're remitting a lot of our fees and costs to the data center, which maybe they remit to the town. It's up to them how they want to do it.

John Furrier

>> You know what I love about the physical AI world, because you're kind of in the physical AI world meets the AI build out, is that the physics side of our life is coming into the digital. And robotics, we're seeing this in drones. They have noise too, you probably have some solutions there, but this is a real world example of material science meets problem. Talk about how you guys got here, what was the origination? No one wakes up and says, hey, I want to mitigate the noise barrier for AI data centers. It's not obvious on the bingo card.

Duncan Abbot

>> Yes.

John Furrier

>> So, how did you get here?

Duncan Abbot

>> I didn't see us ending up here either at first, for sure. So, my co -founder and I have worked on a number of projects together. He's a professor at the University of Michigan, and one thing we had worked on before is using AI to design what are called metamaterials, really structures where the behavior of the system is governed by the shape and not the underlying material as much, so we're not developing a new steel, we're developing a new mesh, or a new sound dampening membrane in this case, and you can actually 3D print these things, so you can print a lot of them, do tests, and feed that data back into an AI, and so we were experimenting with all sorts of metamaterials, and realized that sound was an area that was very untapped. You could develop, there's a very high ceiling of undiscovered metamaterial behavior for sound. And then looking at areas that needed a lot of novel sound attenuation, this just jumped out, right? I think the timing was perfect. We were using AI.

John Furrier

>> Yeah, airports, you got data centers. There's a slew of problem areas.

Duncan Abbot

>> you mentioned drones, actually a great point. We had worked on sort of drone propellers and wing structures before. I think it very well could be in the same realm of technology and something we look at later, which is as we get maybe drone taxis and that becomes more prevalent, you're going to need to see pretty advanced propeller technology to make them quiet and that is in our wheelhouse.

John Furrier

>> We've all had drones humming over our heads and you know what drones sound like. We know that sound. On the data center side, getting back to here, what are some of the site sizes that you're looking at? have you done or worked on? Can you share some of that momentum, some of the conversations you're having?

Duncan Abbot

>> Yeah, I can't name names yet, but I can say we're looking at sites from 15 megawatts up to 100 megawatts as first deployment sizes. The beauty here is the bigger sites need this even more, and so there's a lot of demand from even gigawatt scale sites to fix this quickly.

John Furrier

>> What's the solution at deployment? So again, assuming I'm the customer, you come in and put a device around the fan? Is it in the build?

Duncan Abbot

>> Exactly.So one of the nice things here is it can be retrofit. We don't have to be built in during construction. And we basically are putting, you could look at it as a duct, a vertical duct on top of the fan. The duct is filled with things that strip out the sound, but it's mostly empty air. And so the duct just sits on the fan and air goes right out and it really doesn't hurt anything in the data center. But by the time the air exits the fan through the duct, it's been stripped of any of those peaks in low frequency tones. And so that's the main innovation. Beyond that, you do these walls. The walls are typically built around the chillers on the roof or on the ground if it's a ground deployment. And those are more traditional, but the combination of those two systems allow us to sort of match this SLA level at the bottom.

John Furrier

>> So it's a combination of mitigating techniques. You take care of the hard problem, where you see the most potential complaints, which is either annoyance or actually some sort of physical reaction.

Duncan Abbot

>> We're starting to see more physical reactions as well. I think the studies for that are in their early stages, but at least anecdotally, the low frequency has more of an effect on people.

John Furrier

>> So at the University of Michigan, you guys are beavering away, discover this, and then AI infrastructure, why that selection? Did that jump out at you as an obvious thing? Obviously the CapEx numbers are big. As they say, start with a big TAM, right? What was the reasoning for the AI infrastructure besides the obvious need?

Duncan Abbot

>> Well, I think it's obviously the TAM, but it's also the concentration of clients, right? So you have 20 major players in the data center world, and if you get a few of them, you can even get into one site and then land and expand. You prove that you work for a CyrusOne or for Google, and you can deploy out to more of their facilities and you're sort of their player and their solution, similar to the company I was just at, Cadence, sort of land and expand in top 100 health systems. I think that works really well. It's hard to get the initial deal, but I think we're in a place now where it's so acute that they're willing to work with startups. They're willing to work with whatever solution solves the problem. And so we can get in the door a lot earlier than we would be able to in a maybe less acute market. And that's really, really valuable.

John Furrier

>> All right, talk about the payment side now. I just want to make sure I nail this piece. If I'm the data center customer, you're going to guarantee an SLA. You get paid when what happens? Fees are a little bit upfront, because I'm like, okay, I got a lot of stuff being built, but I got this residential thing I want to mitigate that problem. So what's the package? What do I do?

Duncan Abbot

>> So we take a chunk upfront, I would say, over five years, it's about half up front and then half over five years. And the chunk up front is just to help us actually manufacture and deploy. And after that, it's an annual fee that we just get. It's the same fee every year. And we can break that down. Basically, part of it is just us actually being out there and going and doing site visits. And part of it is a reserve so that we can tackle issues as they arise. And if we fail to do that, we're going to start dipping past that reserve into our own funds and remediate it at our own cost.

John Furrier

>> How much of the mix do you expect the business to be between reactive to an oh, shoot moment, the town's on my butt, I've got to respond, or is it more proactive, hey, I want to be proactive and have this mitigated out of the gate?

Duncan Abbot

>> It's a lot more proactive. The guarantee is so that it never has to be touched. You don't want to, you don't want to crash your car because you have insurance on the car. You want to never do that. But you also want to know that if you did, you can pay for a new car. And that's entirely the model here is that they want to know we're incentivized to keep this sort of green litBut that if it doesn't we have the sort of reserves

John Furrier

>> and the ability to mitigate It's interesting that I've been following the data center buildouts in the beginning absolutely since it started But then it's fun to see all my friends who had data centers or have been in the business It was never really like this We have a big neocloud here in the building at NYSE. I was just upstairs listening to the analyst day and They've got big gigawatt factories going on and a gigawatt on the horizon. They actually have an executive that's called Global Affairs. Basically someone who has to manage the policy side, which is government regulations. But also one of them is the towns and cities, which is why I brought up this whole protesting. People don't even know why they're protesting these days. It's like the telecom towers. There are real concerns in the large deployments that they're actually putting personnel in charge to do that. So I'm assuming you work with either the global affairs person or someone on the ops side?

Duncan Abbot

>> Exactly, it depends on the customer a little bit, but sometimes it's very much at the executive level because this is such a big problem. But some groups we're working with, sort of the sustainability group that is responsible for engaging with local communities. sometimes it's the more like PR side, the group that engages with the local municipalities and the governments, and yeah we're.

John Furrier

>> And there's always that old school consultant mentality that they would do on any kind of project, changes to the water table, putting a new sewer system in. infrastructure, there's a playbook for rolling out infrastructure in the US, for sure. You hit that too, this is like a normal playbook, but you come in saying okay, here's the pitch, any questions? Okay, next question, next debate.

Duncan Abbot

>> Yeah, ideally we come in and make it very easy and make it clear that we're not going to leave you after we sell you stuff to put on your facility.

John Furrier

>> All right, so what's next? Obviously you're going to go land and expand. On the product side, any new innovations on the horizon that you guys are experimenting with that you can share or thoughts on where this goes?

Duncan Abbot

>> Yeah, so maybe on the business side, right, We're trying to sort of balance going and expanding too far because we have a limited capacity of fabrication right now. And we really want to get out there and just show data. We want to get out there, show data, show the model works, and do that before the end of the year. And then expand as fast as we can.

John Furrier

>> Well, you certainly add to the acceleration. No one wants to slow down. The demand curve is so high. Supply is short. Any disruption is a problem. So expensive. Duncan, thanks for coming on our AI Factory series. Again, the AI infrastructure build -out's creating a whole new supply chain and ecosystem as AI factories are just another infrastructure rollout. They're pouring the concrete, they're paving the roads with AI factories, the digital version, to bring in the innovation of the AI era. I'm John Furrier, host of theCUBE. Thanks for watching.

Duncan Abbot, Zenbul Labs

search

(INTRO)

John Furrier

>> . Hello, I'm John Furrier, host of theCUBE, here at theCUBE's NYSE studio. For theCUBE, of course, we have our Palo Alto studio, where we connect Silicon Valley to Wall Street, and Wall Street to Silicon Valley. This is our AI Factory series. We talk to leaders, startups, innovators who are making it happen, building the future of the AI era, pouring the concrete, paving the roads for the AI era. This is what's going on, it's the hottest area. Duncan Abbot here is the CEO and founder of Zenbul Labs, who's got an innovative solution around helping AI factories get online faster, mitigating noise for the neighbors. Everyone's protesting data centers these days. Don't know why, they're awesome, but there are issues. Duncan, thanks for coming on our AI Factory Series. You're a hot startup, thanks for coming on.

Duncan Abbot

>> Thanks so much, thanks for having me.

John Furrier

>> I really am psyched you're here because one of the things going on right now is the protesting of data centers being built. Reminds me of the old days back when cellular technology was rolling out and towers would go up. We don't want a tower in my town, I'm going to get brain damage from that radio frequency. There's a lot of misinformation around that, things like that, but data centers actually have a problem. You're solving a problem that these big data centers have, which is noise. Noise that the neighbors get pissed off about, people are like, wait a minute, there's all kinds of levels of frequency. So it's not like, wow, it's a concert playing, it's more of a different kind of noise. Explain what you're doing.

Duncan Abbot

>> Definitely, so yeah, as you say, there's a big NIMBY pushback on data centers and their noise profile, and a lot of these data centers need to be built near urban or suburban environments, and so they're near people's homes. and the fans that cool these data centers are very loud. And not only are they very loud, but they produce these hums at low frequencies that sort of cut through walls and concrete barriers and into your home. And it's actually those hums that we're seeing cause a lot of lawsuits and moratoriums and sort of big residential complaints and it's mostly an unsolved problem and is really blocking go-to-market speed for a lot of these necessary deployments.

John Furrier

>> Yeah, it's usually after the fact, so it's kind of like there's mitigating techniques. Explain what people tend to do today because, obviously, urban areas, that's where the power is or in an area. What is the current state of the art to mitigate the noise? Do you build walls? Is it like special?

Duncan Abbot

>> It's always less complicated or less advanced than you might think, right? So it's walls generally, and depending on whether this is on the roof or on the ground, the wall is going to be different heights. But yeah, you're building a barrier, and the traditional thinking was, well, higher frequencies are what humans hear better, and so we build a wall and that blocks higher frequencies, and so we're good, but it's really brought out this revelation that it's the low frequencies that are the problem, and low frequencies bend over the walls, and so it just doesn't work.

John Furrier

>> Yeah, I'm not an audio expert on the physics of sound, but in my place in Palo Alto, we are literally two towns away from Shoreline Amphitheatre, and they have a big wall. Some nights I can actually hear the concert through my windows almost clearly. Something about the way it comes in, the physics, just the way the environment is at that time.

Duncan Abbot

>> The weather of the day might affect it.

John Furrier

>> Talk about that aspect of it, because that's something that's hard to control if you don't nip it in the bud. Is that some of the factors that go into this noise dispersion?

Duncan Abbot

>> Definitely.I think a lot of these traditional solutions, they come in, they sort of blanket it with maybe an over -engineered wall, and they don't think about how is the temperature going to fluctuate throughout the year. They don't think about what is the cloud cover going to look like, maybe what weather events might happen. So they have these issues. What we're trying to do is come in and actually guarantee a sound level at the property line and monitor all these things and react and update accordingly if we need to come in and improve the wall or improve our own low frequency dampening solution or adjust it based on that location. You need someone who's responsible for it, not just on day one, but for the next 10 years.

John Furrier

>> Ongoing. I want to get into the solution, but before we do that, explain the noise problem in layman's terms. You got a property line, you'd have homes or businesses, maybe a couple towns over if that's an issue. What is the impact of this noise problem to the surrounding environment that you guys are seeing and solving? What would people say would be the complaint?

Duncan Abbot

>> So the complaint more and more now is that 24/7 there's this hum that reaches inside your home. So you can't sleep, maybe it's giving you vertigo. I know people who actually are more sensitive to low frequency tones in your inner ear and that causes vertigo than for higher frequency noise. And because of anti -AI sentiment across the world, coupled with maybe anti -data center sentiment, this is something they can latch on to. You're sitting there, you're hearing it. Maybe you wouldn't get up and take action normally, but now you're going to get up because this is in your home. And we throw out lots of numbers. One number is that just this year, $150 billion or more of data center projects have been slowed down or canceled because of NIMBYism, right? And this is a big part of that.

John Furrier

>> Yeah, and we've all seen, we all kind of see on the Instagram and TikTok or the internet, the council meetings, they're not experts, but you got the residents complaining, wait, don't put the data center, it's going to be, make up all these things. Noise is probably one of them. There's been some class action. Talk about some recent.

Duncan Abbot

>> The first two class action lawsuits for noise at data centers ever just this year in May and July in Michigan and Wisconsin. And it's exactly this. They sort of came together in a way that I think the communities wouldn't have been able to do without all this clout and are pointing at the noise issue. And saying, we're going to stop you from moving forward or stop an existing data center because of this problem. we can just fix that.

John Furrier

>> Yeah. Yeah, well, certainly I've been inside data centers with a lot of GPUs, and it is loud as hell inside the data center. This is outside. Get into the solution because it's a unique proposition. Talk about how you guys got here and what specifically are you doing because you've got to cool the machines. There's liquid cooling. Air has got to come out, somewhere. That's the key thing. You can't just block it up. Talk about how you guys solved this, and what's your approach to helping not slow down the AI build-out?

Duncan Abbot

>> Definitely. Definitely. So, you really hit the nail on the head. The problem is you're cooling these data centers and cooling almost always is going to end up with air being rejected to the atmosphere and the sound is in the air, so to speak. And so you could cover it with a big lead shield and that would get rid of all the sound, but then you'd also stop pumping air out into the atmosphere and it wouldn't work. So the problem is you need to let the air out and strip out these sounds, especially at low frequencies. And you have to do this with different techniques, silencers and resonators, and what we've done is use some technology I developed with my co-founder to really build novel resonators that not only can strip out these low frequency tones, but can do it in a way that is targetable per site. So your site in Michigan is different than your site in Wisconsin. The exact tones that the fans produce are maybe even different if it's the same fan, depending on the site and installation. And we're able to very carefully tune it to match your fan and really effectively strip out that sort of peak that you hear at your home.

John Furrier

>> So you attack the problem frequencies, the low frequencies.

John Furrier

>> Yes.

John Furrier

>> The high frequencies can be mitigated by...

Duncan Abbot

>> The barriers work, although I think we want to be involved in the orchestration and implementation of the walls in concert with our low frequency tonal solution because then if we're going to have this SLA for residential property line sound levels, we need to be aware of and responsible for everything going into that. So the unsolved piece we bring to the table for low frequency and we oversee the solved piece at high frequency and then we wrap it in a guarantee that we manage and supervise.

John Furrier

>> All right, so I'm a customer, pretend I'm a customer. Hey Duncan, I'm doing this major CapEx build out, I'm going to XYZ town, it's in a suburb, I got good power, I'm totally stoked, and next thing you know we run into all these things. how would you engage them? Because I definitely want you to come in and take this off my plate, because I'm too busy building. What do you do? You say, hey, buy this material, put it over the fan. Do I pay for that? What's the service to me?

Duncan Abbot

>> Well, I think the first thing is, we would want to come in and establish a baseline. We'd want to come in and say, both to the data center and the residents, look, let's measure what the sound is now, and then we can prove to you that we're upholding our end of the bargain. And I think just that as a first step is something that we're not seeing everywhere unless it's required by law and shows a lot of good faith. And we follow that up with saying, okay, now we're going to commit to exactly this low -frequency tonal dampening and the higher -frequency dampening and publish that to residents. And once that's agreed to, we go in and install the range of solutions, including our own proprietary technology, and start monitoring it. So I think whether it's a new build, which will take however long it takes to finish the build, or an existing build that's starting to have pushback, we can go in and retrofit and add on top or be part of the build solution and market that from day one and not wait for it to be a complaint that happens at the last moment.

John Furrier

>> And do you guarantee measurement? Now, I'm like, okay, thanks for coming in. You helped me out. Let's just say that worked. What do you do? Do I go to your dashboard? Is there a system? What's the SLA? How do I prove that the before and after levels, I'm cool?

Duncan Abbot

>> So we are going to install, we're installing these third -party very official sound measurement devices at the property line. So it's not our device. We're not double dipping and then claiming our own measurements. And if that falls outside of the SLA, we have a tiered approach to resolution. So there's immediate, within the hour alerting, and that's very public, and within a day root cause analysis, within a week commitment to a resolution, within a month a full resolution. And then if we fall out of that multiple times in a year, then we're going to break our SLA.

John Furrier

>> So you're a service provider to the data center. And thenwe can all relate to town council, I mentioned that earlier. So the old model was some civil engineering firm or environmental engineering firm would come in, make a presentation, oh, everything's going to be good. But there was no actual actionable data. It was just more of a consultant opinion. You're actually kind of doing environmental engineering on behalf of the data center on the audio piece. That's what you guys do.

Duncan Abbot

>> And we're putting our money where our mouth is too.

John Furrier

>> And you're servicing it, so you're an always-on provider.

Duncan Abbot

>> Correct, and if it doesn't work, if we break the SLA and we're not delivering, we lose money. So I think we want to have these win-win incentives to keep everyone moving as fast as they can.

John Furrier

>> Is there a liabilities piece to this? Because sound is sound, but if you can mitigate the sound and guarantee SLA, I don't think there's much risk on the back end Because it's really more of guaranteeing to the town or city or the municipality or whatever, the region, that everything's good. You've got the before and after.

John Furrier

>> Exactly.

John Furrier

>> So there's really no, it's more of, okay, we check the box.

Duncan Abbot

>> You check the box and I think almost we are a layer of insurance that the data center can present to the town and if we're failing, then we're remitting a lot of our fees and costs to the data center, which maybe they remit to the town. It's up to them how they want to do it.

John Furrier

>> You know what I love about the physical AI world, because you're kind of in the physical AI world meets the AI build out, is that the physics side of our life is coming into the digital. And robotics, we're seeing this in drones. They have noise too, you probably have some solutions there, but this is a real world example of material science meets problem. Talk about how you guys got here, what was the origination? No one wakes up and says, hey, I want to mitigate the noise barrier for AI data centers. It's not obvious on the bingo card.

Duncan Abbot

>> Yes.

John Furrier

>> So, how did you get here?

Duncan Abbot

>> I didn't see us ending up here either at first, for sure. So, my co -founder and I have worked on a number of projects together. He's a professor at the University of Michigan, and one thing we had worked on before is using AI to design what are called metamaterials, really structures where the behavior of the system is governed by the shape and not the underlying material as much, so we're not developing a new steel, we're developing a new mesh, or a new sound dampening membrane in this case, and you can actually 3D print these things, so you can print a lot of them, do tests, and feed that data back into an AI, and so we were experimenting with all sorts of metamaterials, and realized that sound was an area that was very untapped. You could develop, there's a very high ceiling of undiscovered metamaterial behavior for sound. And then looking at areas that needed a lot of novel sound attenuation, this just jumped out, right? I think the timing was perfect. We were using AI.

John Furrier

>> Yeah, airports, you got data centers. There's a slew of problem areas.

Duncan Abbot

>> you mentioned drones, actually a great point. We had worked on sort of drone propellers and wing structures before. I think it very well could be in the same realm of technology and something we look at later, which is as we get maybe drone taxis and that becomes more prevalent, you're going to need to see pretty advanced propeller technology to make them quiet and that is in our wheelhouse.

John Furrier

>> We've all had drones humming over our heads and you know what drones sound like. We know that sound. On the data center side, getting back to here, what are some of the site sizes that you're looking at? have you done or worked on? Can you share some of that momentum, some of the conversations you're having?

Duncan Abbot

>> Yeah, I can't name names yet, but I can say we're looking at sites from 15 megawatts up to 100 megawatts as first deployment sizes. The beauty here is the bigger sites need this even more, and so there's a lot of demand from even gigawatt scale sites to fix this quickly.

John Furrier

>> What's the solution at deployment? So again, assuming I'm the customer, you come in and put a device around the fan? Is it in the build?

Duncan Abbot

>> Exactly.So one of the nice things here is it can be retrofit. We don't have to be built in during construction. And we basically are putting, you could look at it as a duct, a vertical duct on top of the fan. The duct is filled with things that strip out the sound, but it's mostly empty air. And so the duct just sits on the fan and air goes right out and it really doesn't hurt anything in the data center. But by the time the air exits the fan through the duct, it's been stripped of any of those peaks in low frequency tones. And so that's the main innovation. Beyond that, you do these walls. The walls are typically built around the chillers on the roof or on the ground if it's a ground deployment. And those are more traditional, but the combination of those two systems allow us to sort of match this SLA level at the bottom.

John Furrier

>> So it's a combination of mitigating techniques. You take care of the hard problem, where you see the most potential complaints, which is either annoyance or actually some sort of physical reaction.

Duncan Abbot

>> We're starting to see more physical reactions as well. I think the studies for that are in their early stages, but at least anecdotally, the low frequency has more of an effect on people.

John Furrier

>> So at the University of Michigan, you guys are beavering away, discover this, and then AI infrastructure, why that selection? Did that jump out at you as an obvious thing? Obviously the CapEx numbers are big. As they say, start with a big TAM, right? What was the reasoning for the AI infrastructure besides the obvious need?

Duncan Abbot

>> Well, I think it's obviously the TAM, but it's also the concentration of clients, right? So you have 20 major players in the data center world, and if you get a few of them, you can even get into one site and then land and expand. You prove that you work for a CyrusOne or for Google, and you can deploy out to more of their facilities and you're sort of their player and their solution, similar to the company I was just at, Cadence, sort of land and expand in top 100 health systems. I think that works really well. It's hard to get the initial deal, but I think we're in a place now where it's so acute that they're willing to work with startups. They're willing to work with whatever solution solves the problem. And so we can get in the door a lot earlier than we would be able to in a maybe less acute market. And that's really, really valuable.

John Furrier

>> All right, talk about the payment side now. I just want to make sure I nail this piece. If I'm the data center customer, you're going to guarantee an SLA. You get paid when what happens? Fees are a little bit upfront, because I'm like, okay, I got a lot of stuff being built, but I got this residential thing I want to mitigate that problem. So what's the package? What do I do?

Duncan Abbot

>> So we take a chunk upfront, I would say, over five years, it's about half up front and then half over five years. And the chunk up front is just to help us actually manufacture and deploy. And after that, it's an annual fee that we just get. It's the same fee every year. And we can break that down. Basically, part of it is just us actually being out there and going and doing site visits. And part of it is a reserve so that we can tackle issues as they arise. And if we fail to do that, we're going to start dipping past that reserve into our own funds and remediate it at our own cost.

John Furrier

>> How much of the mix do you expect the business to be between reactive to an oh, shoot moment, the town's on my butt, I've got to respond, or is it more proactive, hey, I want to be proactive and have this mitigated out of the gate?

Duncan Abbot

>> It's a lot more proactive. The guarantee is so that it never has to be touched. You don't want to, you don't want to crash your car because you have insurance on the car. You want to never do that. But you also want to know that if you did, you can pay for a new car. And that's entirely the model here is that they want to know we're incentivized to keep this sort of green litBut that if it doesn't we have the sort of reserves

John Furrier

>> and the ability to mitigate It's interesting that I've been following the data center buildouts in the beginning absolutely since it started But then it's fun to see all my friends who had data centers or have been in the business It was never really like this We have a big neocloud here in the building at NYSE. I was just upstairs listening to the analyst day and They've got big gigawatt factories going on and a gigawatt on the horizon. They actually have an executive that's called Global Affairs. Basically someone who has to manage the policy side, which is government regulations. But also one of them is the towns and cities, which is why I brought up this whole protesting. People don't even know why they're protesting these days. It's like the telecom towers. There are real concerns in the large deployments that they're actually putting personnel in charge to do that. So I'm assuming you work with either the global affairs person or someone on the ops side?

Duncan Abbot

>> Exactly, it depends on the customer a little bit, but sometimes it's very much at the executive level because this is such a big problem. But some groups we're working with, sort of the sustainability group that is responsible for engaging with local communities. sometimes it's the more like PR side, the group that engages with the local municipalities and the governments, and yeah we're.

John Furrier

>> And there's always that old school consultant mentality that they would do on any kind of project, changes to the water table, putting a new sewer system in. infrastructure, there's a playbook for rolling out infrastructure in the US, for sure. You hit that too, this is like a normal playbook, but you come in saying okay, here's the pitch, any questions? Okay, next question, next debate.

Duncan Abbot

>> Yeah, ideally we come in and make it very easy and make it clear that we're not going to leave you after we sell you stuff to put on your facility.

John Furrier

>> All right, so what's next? Obviously you're going to go land and expand. On the product side, any new innovations on the horizon that you guys are experimenting with that you can share or thoughts on where this goes?

Duncan Abbot

>> Yeah, so maybe on the business side, right, We're trying to sort of balance going and expanding too far because we have a limited capacity of fabrication right now. And we really want to get out there and just show data. We want to get out there, show data, show the model works, and do that before the end of the year. And then expand as fast as we can.

John Furrier

>> Well, you certainly add to the acceleration. No one wants to slow down. The demand curve is so high. Supply is short. Any disruption is a problem. So expensive. Duncan, thanks for coming on our AI Factory series. Again, the AI infrastructure build -out's creating a whole new supply chain and ecosystem as AI factories are just another infrastructure rollout. They're pouring the concrete, they're paving the roads with AI factories, the digital version, to bring in the innovation of the AI era. I'm John Furrier, host of theCUBE. Thanks for watching.