Not Your Father’s Data Center - Dynamic Power Solutions for the Future of Energy Storage

Episode Date: February 17, 2026

In this episode, Raymond Hawkins, host of Not Your Father's Datacenter, sits down with Dr Sai Shivareddy, CEO and founder of Niable. Dr Shivareddy shares his fascinating journey from India, through his academic pursuits in physics and nanomaterials at Cambridge, to founding Niable with a focus on revolutionary battery technology.The discussion centers around the dramatic rise in power density within data centers, the historical shift from lead acid to lithium-ion batteries, and their respective challenges—especially in the face of rapidly spiking AI and GPU workloads. Sai explains Niable’s breakthrough material, enabling batteries to recharge in minutes while maintaining safety and high power density. The pair explore the urgent industry need for new solutions to manage dynamic, unpredictable power loads and introduce Niable’s dynamic response power systems as a novel bridge between supercapacitors and batteries. The episode offers practical insights on how innovative materials science can redefine power management in the next-generation data center.Timestamped Overview00:00 Intro & Dr Shivareddy’s background 04:51 From lab to real world06:38 Data centers growing power demand10:23 Energy Storage Capacitors vs Batteries15:22 UPS vs BBU power comparison17:27 Battery stress at full capacity20:13 GPU power surge trends25:33 Voltage conversion in power systems27:56 Dynamic Power Regulation System30:17 Power System Design and Optimization34:43 Optimizing GPU power efficiency36:06 Power solutions for dynamic workloads

you're talking of like 75 kilowatts of average power or something, and then peak power to 150 kilowatts. So in your latest blackwell racks, it's 150 kilowatts of that peak. But what they have to do is, because of the excess headroom they need, you need almost 250 kilowatts of power. And that's basically what you have in an NV-72-ARAC, right?

We've gone up like 25 times in like five years. And we just heard from NVIDIA that they're going to a megawatt in a rack in the next two years. That's insane power density, Welcome to another edition of Not Your Father's Data Center. I am Raymond Hawkins, your host. Once again, we are going to talk about technology. Today, we are joined by a friend of the program all the way from Cambridge, England, Dr.

Cy Chevroredi, the CEO and founder of Niobolt. They do batteries, and he's going to explain to us how we might want to hang on to some of the energy that we have and how that might change what we do in the data center. Dr. How are you today? I'm great. Thank you. Sai, so nice to meet you. All right. Well, we'll go with Sai the rest of the way if you're okay with it.

Yeah, that's great, please. Cy, you're in Cambridge today recording, right? Indeed. Yep. All right. Yeah, it's getting cold, but not as cold is probably where you are. Well, we're doing our right still. I'm in the southeast, so we get a little bit longer hang on to summer,

but we'll enjoy a very brief winter where I live for, which I am very grateful. So, Si, if you don't mind, can you give us a little bit of insight into you before we get into Naya Bolton, what you guys are doing on the battery technology side? Can you back up and tell us where'd you grow up, where'd you go to school, how'd you get to Cambridge? You go back as far as you like and give as much details you like, but we'd love to hear a little bit about you before we get into batteries in the data center space.

Absolutely. So I was born in India. I grew up there, went to school, at least like middle school, high school there, and also an undergrad program, studied physics, naturally curious about the world around me and how it all came together. So try to dig as deep as possible, and normally you end up doing a PhD if you try to go deep. I found myself on a scholarship to Cambridge because there was a history there. You know, we spoke about some history in the past, and the university goes back 800 years.

and growing up in India, there was a stronger connection to the UK, just historically, right? So I had somehow met a professor who brought me on and, yeah. Help orient us. What year was this? What year did you to leave school in India and go to the UK? Okay, so this was over 20 years ago. Okay. About half my life, been in the world of research and materials and energy. But before that, I had no clue where I would land. everything, science was interesting, but the thing that was changing and, you know, felt interesting

was both the new world of energy and the power that would drive the next century. This was turn of century around 2000. Right, right, right. So there was that world of tech. And then there's a world of infrastructure around, you know, obviously climate and energy were front and center after that first decade. So that influenced what subjects I selected or at least you feel like you can do something useful.

So either you figure it out along the way or you kind of see, okay, this is a general direction. So there was nanotechnology at that time of nanomaterials and this was a very big popular science. You know, around 2000, there was a lot of discoveries in new materials at the nanoscopic level that could fundamentally change many fields, right? From, you know, photonics and computing.

I mean, yeah, actually that's a big part of what I was excited about Cambridge too. So, you know, this was 2007-8, and I got this PhD. This department was looking at the 45 nanometer node, so you're talking about semiconductors and data centers. There was some connection between semiconductor processes, materials that worked at the nano level, and we talk about the two nanometer node now.

Back then it was 45 nanometer, that was a big milestone, and obviously all the downsizing of chips and the power going up. To me, like, you know, that was very exciting. to get into and see how deep you could go and see how you could change things because you actually try and understand how things are the way they are. So what specifically, I don't know if the terminology is the same in the UK, what specifically was your doctoral thesis? They was on nanomaterials for energy storage and conversion. Gotcha. So I ended up on the energy storage front and I think that's probably why you got me on the show. Has something to do with what you do professionally. Very good.

Okay, gotcha. It did not directly take me there straight away after, but it did feel like I could contribute uniquely to progressing the industry at where it was based on some of those learnings. But it's not a straightforward journey because, you know, what you do in the lab never scales. And I think that's one part. It's fundamental to a startup company's early journey.

So essentially, the research that people do, whether in university labs or in companies' research labs, there are things that stay in the lab and there are a lot more things that stay in the lab than make it to the real world. And there are many reasons why they actually don't make it, right? So it's fascinating to see which ones do and which ones don't. So there's like an overlap between research done in universities

and research done in sort of commercial labs, whether they're in companies or there is a certain direction you're going where you may or may not know the outcome. and that time frame is sort of indefinite. In the university lab, you never know when you have a breakthrough, but you only have money for a certain period. So, you know, either you get there, you don't, because usually a project's funded for a certain time.

But the answers to some of these scientific pursuits don't happen in that window. And that's when, you know, commercially and academically, either they align or they don't, and either they click, you make a lot of progress and, you know, get a product out there, or you never get there, right? I've seen this over now, a 20-year period where there's so many things that people want to do

in the university labs or just like the make environment. They may never see the light of day, but, you know, some things can accelerate those innovations and other things just come through rapid iteration and people's experiences. So our world, in the data center space, the number one thing everyone talks about today is access to large-scale power, right?

You know, it's shocking to me how quickly the demand for power has grown, how the demand for access to power has changed where we build data centers. And generally here in the U.S., that's on the grid power because we've got a little more land, a little more power than our friends in Europe. So really, really large projects getting really located because of access to large power generation. However, generation isn't global, right? It's not everybody can drop a data center in the middle of Wyoming and get a couple of gigawatts of power to it.

So we recognize that that's a unique demand profile and supply profile in North America that doesn't exist everywhere else. So then we're asked lots of other, okay, how do I get enough power? How do I hang on to enough power? And I would love to hear how you guys and I both are connecting what you do back to the data center business, because if there's, it is without a doubt the number one subject in our business is, how do we get enough power?

And so much so that there's, you know, microgrids have been talked about and, you know, being able to put power back on the grid because in the data center world, right, we need power 24-7, 365. We can take no downtime. So we always have everything redundant. But yet the grid has times, where it has spikes, at least here in North America, right? Summertime, there's spikes. And so we're trying to put energy back on the grid at times. So there's all kinds of conversation for us around how do we manage power, how do we manage the grid.

I would love to hear how you at Nyebold envision battery technology and improved battery technology fitting into the data center space. So at Nibald, we founded the company, myself and my co-founder, Professor Clay Gray here in Cambridge University. We set out with a vision to reduce the wait times to recharging batteries. So we invented this new material that allows a battery to recharge in under five minutes and do that repeatedly without catching fire, as you've seen in many cases, or actually having a lot more density than is possible. And then these are key aspects of like data center and power.

Like power density is critical here. Like you need a lot more power per unit volume than what was in the past in terms of traditional cloud data centers, right? So our core breakthrough has been on the materials front where you can recharge essentially an energy storage device in as little as microseconds, obviously not fully, but you can get the fastest energy input and fastest energy output. So the rate of energy flow is power, right?

And is that battery agnostic side? So you're talking about getting the energy into the battery. Can I use, and I'm not saying any battery, but is it I could use any rechargeable battery and get the charge? And people think of their cars, right? That's probably the one that people are most interested in. So, Nibel makes batteries.

Okay. So it's your battery technology and your charging technology. And our invention on the materials that goes into the battery cells. Okay. Enables our batteries charge fast. So it's actually a materials limitation that we've overcome and solved. Okay. In the battery.

In the battery. So, you know, we were talking about materials for energy storage, as I said in my PhDs, 20 years ago. My co-founder and I looked at, all kinds of materials over the last decade in terms of how you can get the energy in extremely fast overcoming the traditional problems that you have and they still have those.

Usually it is bottlenecks in charging because it heats up too much and too much heats not good, right? Fire and other things then will run away. If you could solve that problem on the input-powered entity side, you would end up with some kind of a symmetric charge-discharge profile

or an input-output power profile. I mean, you probably are aware about capacitors and other forms of energy storage systems. But while they have more rate of charge, they can charge very fast and discharge and symmetrically with very high power, they have very limited energy stored inside them. Like, let's say it's in the microseconds to milliseconds level.

So every power supply has these little passive energy storage units like capacers and inductors, right, to do that switch mode power conversion. It's a bit of energy stored and then it's chopped to get you DC output. these capacitors have like, I think, three orders of magnitude, lower energy stored inside than lithium-man batteries. But I think it's between three and six, depending on the type of pure dielectric film capacitor, like these other types of electronic capacitors, multi-light capacitors, and electrolytic capacitors, and then there's super-capacitors. And then, you know, we start getting into the battery domain of like seconds, right?

So it's like a fast response energy. You have everything from like microseconds to milliseconds to seconds and to minutes and hours, right? So we looked at it from the microsecond level in terms of fast response all the way to the minutes level in terms of full storage of that energy in that device in number five minutes. That's exactly how we get a rapid charge in five minutes because we looked at the materials that were the limiting factors. I'm going to ask you to back up, if you will. in my business, the idea of batteries inside a UPS pretty normal. We're all used to that. People, you know, hey, my power system's gone down.

My generator's not online yet. My UPS is going to give me some power to my compute so my compute doesn't go down. Why my generator spins up because I've lost utility power. So in the world of data centers, we're all really comfortable with that notion. But we're getting, you know, six, seven, eight minutes of runtime out of that massive room full of batteries. And those batteries, before they became lithium, what were they? Lead acid. Yeah. So they were all rooms full of lead and acid. And rooms and rooms and rooms and that all had a diminishing life, right? Because on day one, they might have given me seven or

eight minutes. But at year three, they might give me three and a half or four minutes. So my, our industry and our listeners are pretty comfortable with the notion of, hey, what's in a UPS, that lead acid, you know, how much runtime, it's got a useful life, it goes away. Then we started making the switch to lithium, and lithium started offering unique solutions and unique problems. Can you help us talk through the switch from lead to lithium and then help us talk through some of the limitations of lithium? Great.

So that actually comes from, you know, I was talking about timescale. So I got to the minutes level. and what I was talking about was input power in the minutes. So if you start with what you have today in those big UPS systems that had lead, they took like eight hours to charge, not eight minutes. Right, right. They could discharge in eight minutes. Well, my discharge was an eight minute window.

That's right. But the recharge times of lead acid is like overnight, right, eight hours, whatever. So you can't do very much with that UPS system if you took eight hours to recharge it. So. Yeah, it's kind of a one-shot wonder. Yeah, I get one shot and then tomorrow it'll be back online. So depending on how you design it, you have redundancy there and you put very many of those units to make one's a little bit than the other ones now, right? Yeah, yeah.

So just to give you an order of scale here, so lead acid has energy stored per unit volume of around 20 watt hours, 20 to 40 watt hours. Lithium ion has 200 to 400 watt hours per liter. Okay. So a factor of 10. For every 10 of those, yeah, so it's 10x the capacity. Okay. Lithium ion, traditional lithium ion has this higher energy density,

but it comes with the increased energy stored per unit volume problems. That is, how do you manage for the thermal? It's hotter. Yeah. In a non-scientist's guy's mind, it's just hotter, right? It is hotter, exactly. So I got the same amount of energy in a tenth of the space. It's freaking hotter.

And now I got to manage that problem. Yeah, and lithium-on can charge faster than lead acid. So, you know, you can get one-hour charge. Some of your best devices out there can charge you 80% in 30 minutes now, right? Like, whichever device you use like a phone or on your E. And even within lithium ion technology, there's like different flavors with various fade-off. So you want to store more or you want to run longer. You know, it's called an energy cell.

And if you want it to use for fast response, it's called a power cell. So those types of power cells made it to the UPSs of today. Or a more, let's say, a recent derivative of that has been the battery backup units, or what they call BUs, right, in some of the large hyperscalerscale. And the big difference between these two is UPS is AC input, AC output. It's giving you that sinusoidal good power quality and all of those power-smoving aspects on the AC front. whereas battery backup units are all DC and it's just for backup.

It's not for power smoothing or things of that, right? Yeah, it's not cleaning up my power at all. I'm just straight DC. Yeah, yeah. So I think a lot of these old, or not I don't say old, let's talk about like cloud with CPUs and normal servers were just AI service, right?

So I think that's where the big difference is coming in terms of the differences in the load and how that affects not just, you know, the density, but all aspects of safety and whether they can actually do the job. So, let us say it was pushed out because the density was not good, right? Energy density was not good. It could give it that backup, but you still even had a lot more needed to get the same amount.

And the useful life was so deteriorated so quickly. Exactly. You couldn't have full depth of discharge cycling, which lithium on offers. Now, in the last seven years, you've had a lot of lithium-batteries come in. Because lithium-on intrinsically has higher energy density, and the old materials that were used on the anode side, so this is a negative electrode inside the lithium in battery, it's usually a graphite form.

It's not as safe when you use that over time, over a period of time, charge, charge, discharge. So you're normally floating, let's say, if you want the UPS to function, it should be fully charged all the time and hold on that charge. When the grid fails, you know, at that time you discharge for your five minutes or so, right? Right.

Until the backup generators come on. So because they're held, at that high potential all the time, high potential is not good. Essentially, the materials inside tend to react more at a higher potential because it has high. Right, right.

They're stressed out, for lack of a better term. Yeah, yeah. Yeah. It's a bit like, you know, you have a glass and you poured it full and the slightest perturbation dropped spill something, right? Whereas if it's below,

it just, you know, shakes inside and it's not spilling outside. So think about the potential of lithium ions going into the electrode that's if it's full. Even if you just hold it there, the slightest perturbation in terms of what's going on

on the other components inside the cell, they tend to react and cause side reactions. And these side reactions over time just get worse and worse and worse because it's always held at full potential. You see? Unlike your phone, which, you know, you charge it overnight. Yeah.

The car, you charge it overnight and then you, you know, you drive or you use it. Spend the whole day running it down. Yeah, it's not average held at full potential. Like the batterying the data isn't. So it's not comfortable doing that. put it mildly, right? Like, as a result, you have degradation pathways

or essentially side reactions that cause impedance growth over time. Impedance growth means essentially more heat when you discharge it and that as resistance increases, the I-square-R losses or the current losses over that resistor causes more and more heat when it's discharging. And that's one pathway for like an unsafe state. And you know, there are these aspects of unsafe that they may have states that you don't want to be in.

And they were okay, but, you know, if you don't get it right, you have nasty fires across the whole data centers, and you've seen that in the past. Customers get upset when that happens, just as a heads up. Yeah, when the data sun catches on fire, they get really mad at us. Yeah. So, again, this is why people don't like lithium ion, right? Absolutely.

This is just like millions of dollars of assets. Well, I mean, every, I mean, you're kneeling. Everyone got excited about lithium ion, and then we said, well, wait a minute. There's an entirely different risk profile here. yes, there are some advantages, but the risk profile is extreme. So I'm excited to hear what's next after lithium. Because it was viewed as, hey, this is a great answer,

but pretty quickly the heat made it not terribly practical. Yeah, yeah. So there's that part of heat, and then this transition to AI data centers. So cloud has a need for backup, right? That's steady loads. And now you have these highly dynamic workloads. You've got 180%, you've got 20%, 180%, like 20%. like 20 times a second.

So if you're doing like training workloads, you have these fluctuations, right? Like you have the GPU, which let's say is one kilowatt. It's contrasted with the normal cloud data center. Let's say you have one kilowatt. It was not a CPU. It was like 100 watts or something.

But let's say you have one megawatt of power or let's say the average cloud, 10, 20 megawatts. So you could have 2,000 of those CPU-based systems or 20,000 if it was 100 watts. It's actually big for CPUs but small for GPUs. Right. So what we're looking at is the same GPU, while its average power is one kilowatt, its peak is almost 2 kilowatts, or 1.8.

For a short duration, which was in the past the transient, but these transients have become bigger and bigger. You know, you're talking of like 75 kilowatts of average power or something, and then peak power to 150 kilowatts. So in your latest blackwell racks, you know, it's 150 kilowatts of that peak. But what they have to do is, because of the excess headroom they need, you need almost 250 kilowatts of power. And that's basically what you have in an NV-72 AI rack, right? And you contrast that with your normal CPU server, your normal data center, which was not even like 10 kilowatts per rack. You've gone up like 25 times in like five years,

or whatever, since the first GPUs game, maybe it's like eight years or so. Yeah, yeah. And we just heard from Nvidia that they're going to a megawatt in a rack in the next two years. That's insane power density rises, right? I'm sure you've covered like aspects of cooling and all that. Yes. There's all kinds of trouble at the idea. I get in a lab in California, hey, I'd like to, I'm going to have enough GPUs in my rack that it needs a megawatt to run it. But there's so many other ramifications, cooling being one, safety being one.

How do I have that much electricity in that small a form factor and it'd be safe for humans to work on? There's just all kinds of questions coming. So I think in the last like 12 to 24 months, so I was talking about a co-location provider or a virtual CPU or virtual, yeah. machine provider recently, and they had like this data center in Phoenix where you had a cloud site that, I think, was 10 megawatts, so you could have your steady loads at 10 megawatts. And when they thought, okay, we can put 10,000 of these GPUs, each one kilowatt, for 10 megawatts, they couldn't. They could only use 5,000 because of these peak problems.

Because of the spikes. 5 megawatt goes to 10 megawatts spike. So you've got like millions of dollars of assets that you've acquired. thinking that's what you could fit in and they didn't even know that they could not do it until they fit the machines in the racks inside. You sat on that inventory that you can't use, right? So you've heard there's many places or you could use

that 10,000 but throttle it down to less than 50% utilization to have that headroom. So these problems are very different from first of all power problem like one is steady load, other one is a highly dynamic peaky load. Now how do you design a data center for you know, native peaky loads. It looks very different.

You can't use the lead acid UPS systems because these peaks are ultra-fast. These transients, they have like tens of amps from microsecond or, you know, and essentially this slur rate or the DIDT that you need for bringing these GPUs switched on, the latency increases the longer the cable is because of the inductees.

And that's basically like the slu-rate that limits. Yeah, every meter counts. Yeah, exactly. It impacts it. Yeah. So essentially, you may have what is on paper great metrics on, you know, the ratings. Okay, power taken care of, but actually that's just static loads.

For dynamic loads, there were no metrics. There's no, like, continuity of compute metric. You can call an uptime of, like, powers available at that site. But actually, it's breaking when the GPU needs it. So you're having brownouts, you're having blackouts. When it spikes and it's asking. the most, that's when we start to get strained. That's exactly right.

Yes. All right, so tell me how Nyboldt helped with that. So coming back to fast energy storage, you need microsecond level peak response to just, like, come on as fast as you can. You can recharge it again as fast as that window is available. So there's 20 milliseconds of like overshooting by 100% or something 80%. And then there's another like maybe 100 milliseconds or a few seconds. So you know how the workloads are trained, right?

So you have the compute cycle. and then you have a comps, a communication cycle, and then you have like a checkpoint cycle with the memory. So each time you're going up and down, it's a repetitive pattern. And you, of course, you can't predict it fully, but... You certainly get a rhythm for what's coming. Correct. Correct.

So if you look at a moving average over five seconds, you need to be able to, like, discharge 20 times a second and charge across that five-second window. So you're net-net neutral. So you're not actually depleting the energy while you're doing it. The bigger problem here is the grid is not able to give power fast enough for craning. My traditional delivery system can't bounce up. I mean, right, the way a grid works, it has to be stable, it has to have predictable workload.

It has to have balance between production and utilization. That's how the grid stays balanced. But whether it's 60 hertz or 50 hertz, it doesn't matter. We've got to stay balanced. So I can't come and say, give me 20 megs. I know, by the way, I'm going to give back to you in about two seconds. I can't do that to my grid. So tell me how you help me.

Because that is a real problem we are all designing for right now. Yeah. So anyway, we're at the core of that problem from the way power is transferred, right? So you have medium voltage coming and you mentioned like DC microgrids going through. So the new architectures are moving to high and higher

voltage to get the power up, not increasing current. Keeping current fixed, let's get the voltage up to 800 rather than like 48 volts, right? So what's happening, whether it's the old 48 volts or the new high voltage site, you have energy conversion or power conversion where you have like high, medium voltage going down to like, let's say, 4th volt or a 8 volt bus and then order 48 volt bus at the rack. And at 48 volts, you know, that rack has not like the server, it goes under 12 volts

and then 0.8 volts at the GPU for transistors, which normally come on at 0.7 volts, right? So there's a lot of like voltage conversion going on here. And the way voltage and all of the power conversion happens and this is usually through the switch mode power supplies, right? Or DC-DC converters in this case, if it's just pure DC. Each of those converters need a bit of energy stored to get the right voltage regulation going on. So the capacitors have played that role in the past.

And you know how in the beginning of our conversation we said? They don't store enough energy. We've invented something that has lithium ion-like energy storage capacity. So that's still more than 10 times out of traditional lead acid batteries. So we get the five-minute backup capability. So it's like a super, can I call it a super capacitor? Yeah, and we have supercapacitor-like power. Interesting.

So it's sort of bridging the two worlds of capacitors and batteries. And the last five years, we've just been focused on finding applications where capacitors were used, where, you know, high power batteries were used, but they couldn't be charged fast enough. So they either they swapped the batteries out because they can't recharge them in that time frame. We've solved that recharging problem to essentially behave like a capacitor in terms of input power, output power being symmetric. How quick is that?

How quick can we do input back to fully charge? Again, I mean, full energy is a few minutes, right? So if you're discharging in five minutes, you can recharge in five minutes. Okay. But that's in a backup mode. Yeah? Okay. But the peaky workload problem that I was talking about is not a full discharge.

Right. I'm not going to zero. That's right. Yeah. Yeah. Yeah. So there's a lot more system level balancing required. And we've been working on the whole stack here.

It's not as simple as, hey, throw that capacitor out and put this thing and it'll work. You need a bit more engineering at the system level. There's a couple of electrical engineers running around helping, yeah. Well, and a lot of embedded layers in terms of the power conversion. stack along with the energy storage stack, right? Because essentially a power conversion has a bunch of transistors switches that are connected to some energy storage that is switching on and off to get you a regulated output.

So that part and our storage unit, when you combine them, we call this a dynamic response system, dynamic power system that allows you to essentially regulate the voltage output such that the 48-world bus stays at 48-wolds without droop and those rapid fluctuations and that DC bus essentially saturates the capacitors of the power supply

and it translates to peaks or like transients in, you know, the AC part and the UPS generally doesn't have enough storage there even for these very fast transients. And then that's not filtered, you know, it starts blowing up substations or transformers.

Right? So on one side you're seeing the hyperscalers and like GPU manufacturers slow things down So you're getting faster and faster GPs that have to be throttle. That's not good. You know, you're paying top dollar for a GB. Why pay for it that you want to dial down to 70%?

That's right. That's right. Yeah. But you're solving another problem. They feel like they have to. Yeah. Can I ask you a very commercially centric question? Sure.

So I'm a sales guy. So I'm thinking about this in terms of my customers, which I'm sure you are or will eventually. So give me a scale understanding. So what's the biggest footprint meaning from an energy perspective that you've built one of these? You gave it a name earlier that I didn't catch. You called it a – what did you call the – A dynamic response power system, basically. Dynamic response power system.

If you have one of the shelves, it's called a dynamic response unit. So give me a couple – scale. What's the biggest one you made, meaning it measured in power? At 48 volt level, we've been actually shipping products to a customer not in the data center, but it does the same function. Okay. And we announced this actually two weeks ago. What's the form factor? What does it look like?

So it's actually a box that you plug into a rack. It is rack mountable. Yeah, yeah, yeah. So, I mean, we had to like change the height such that it fits in a one new and, you know, like, and also change the X, Y such that, you know, either you are like, a 19 inch or you're an OCP, 21 inch, so how deep. So that part is configurable,

but the internal building blocks have our cells and our power electronics that essentially is a power source regulating voltage output, which can decouple the grid power transients and the grid supply to time of use.

So we have kind of time shifting these spikes such that, you know, if you have up, down, up down, separate by whatever time interval, We time shift us at the grid sees a flat output. So I don't want you to give away any secrets. You said you're shipping to a customer today, not in the data center business. Do you have data center customers in this conversation yet?

We're having one maybe now. Yeah, you and I are having one, yeah, which is we're supposed to be recorded a podcast. I'm going, wait a minute. This is interesting stuff here. So you're a sales guy, I'm a sales guy. Yeah, yeah, yeah. But, yeah, what I'll tell you is this, like this problem was amplified in the last 12 months, okay? Yeah.

And the natural place to go to the usual power supply people and say, hey, how can you put more storage inside a power supply and what do we do? And then people are scrambling to like go find energy storage units that can go into a power supply that can solve this problem. And we're like, hey, we've been doing this for the last five years. Yeah, yeah, yeah, yeah. Exactly this type of power problem. It's just been a different application in a different area. Yeah, this is a problem we are wrestling with today. I mean, look at what's happening at the megawatt.

in the rack. I mean, we come from a megawatt power system in an EV. We built an EV that can charge it a megawatt and discharge it a megawatt. This is a mine hole truck. They run at like close to 1,000 volts. As I said, you know, like 48 volt in your data center, you're going to close to 800 volts or like 1,000 volts even more if you can. But the industry is looking at solutions that have been deployed in other sectors like EVs or grid or industrials where this type of high power density. problems are being solved. So you need solutions from other proven sectors. I mean, you didn't have this problem two years ago, you know?

No, that wasn't an issue. Exactly. And big companies that are in this space, on the UPS makers and whoever else, have not reinvented anything in the last, I would say, like, not to offend anybody, but the UPS, as you know, were invented a long time ago. It looks like the same it did 40 years ago. Which is exactly the same. As they were 20 years ago, I will tell you that.

That's exactly right. Having looked at what's been developed over the last 20 years on the material side, as I told you from my background, into increasing power density for the storage systems. And essentially, that's what led to our startup idea and say, hey, how can we do it 10 times better than what people out there are doing today? Because they haven't reinvented anything in the last two decades. Well, your business being six years old, right? There's some freedom in that, right? You don't show up with a four-decade history around a UPS that this is how I've always done it. There's some freedom in the way.

You will never talk to me because, you know, this is like what new is not good, too much risk. But at the moment, the old metrics for stability are gone out of the window. Like, big hyperscalers are not using UPS systems. They're out. You know, there's no space for them. It's not practical. I can't use it. That's exactly right.

So, this podcast probably should have been hosted by either my head of innovation or my head of design. But because I think they're both going to want to talk to you after we do this. This has been really, really, really fascinating. So where we are today is we have a supply chain to build our building blocks. We have a supply chain to produce, you know, like 10,000, 20,000 units per year. And we've been, I mean, I've invested $70 million to just get this going in three and a half years time. You know, like, that's where bulk of the investment went to, like, build capacity. Right.

We were focused on like these sort of adjacent verticals where these power requirements, high-voltage requirements, were already being disrupted to use that word because they were not done before and now that disruption's coming to your world where Oh, it's coming, it's here. It's here.

You nailed it, right? The GPU, by the way, and getting worse, right? Because the next round of GPUs are going to even be that much more power-hungry, which means the peak in the valley are going to be further apart.

Yeah, absolutely. And the thing is, as you said at the beginning, you know, we are constrained by power. Like, tokens are power, right? like our focus at this point has become more like how can we enable you to use more GPUs per power unit that you have at your site because if you want to double it or triple it or like increase it 10 times you're going to wait for what a decade I don't know in some sites you probably can't ever but can we double that today with what we have because we're

time shifting the power needs at that very fast intervals yeah I love that all right my side this was an awesome conversation we're definitely going to have some other conversations that aren't going to get recorded because we're interested in learning about the dynamic response power systems. Really, really cool idea, really, really cool technology. Love that smart guys like you are, we're thinking about this years ago, because it is a real, real problem in our industry. Yeah, so ultimately it's two things, right? Like, you want to do more with what you have and we don't want to, like, put that extra burden on the society around you and around us. That's right. You know, we want to, like, try and essentially

its efficiency gains from what we already have. Yeah, I can power the data center without having to take more power to do it, right? I don't need to leave that headroom in the space, which is going to help us because you're right. I mean, at the end of the day, all of us want what we can do on our cell phones. The demand for the applications is there. We've just got to figure out how do we meet that demand with power responsibly. And this seems like a huge step forward. The fact that you get how dynamic the workload is and how quickly it changes will help

us as an industry. So excited to see how this goes from here. Thank you so much, Ty, for explaining it to us. Thank you for thinking about it 20 years ago and for building the company around it. And I'm excited to see where you guys go and excited to see what a partnership with us might look like as well. Thanks a lot. I'm blown away by the time we spend. I've not had a real conversation with a potential, you know, user of what we've been building. But it's essentially thanks for doing what you're doing, you know, like especially on a show like this where, you know, we help get that awareness out

in terms of what, you know, leaders are doing, like as yourself in the space to bridge some of these problems. Awesome stuff. Yeah, thanks a lot. Sire, I do have one Cambridge question before we go. I know you told us before we started recording that, you know, they no longer run the courtyard. If I come to Cambridge and we get a partnership,

can you and I race around the courtyard? We can do that. one time. Just one time. Yeah, well, something about whatever, you know, our discount could hang on on who wins the race around the courtyard. And the best part, I'll take you to that Vandigraph generator, like the real big one.

Yes, that would be even better. Yes, yes. Yes. Yes. Yes. I want a picture with the big Vandigraph generator at Cambridge. Awesome stuff. Sai, thank you for joining us. We really appreciate it. Thank you.