Navitas Semiconductor Corporation (NVTS) Earnings Call Transcript & Summary

Welcome to Citi Day 1 TMT Conference. My name is Kelsey, one of the analysts here at Citi covering U.S. semis. We are very pleased to have Chris Allexandre from Navitas Semiconductor here with us today.

So Chris, it's been roughly a year since you step in as CEO and launched Navitas 2.0, shifting Navitas away from lower-margin consumer markets to higher power, higher-value applications. So perhaps we can start there. Tell us where are we in a transformation strategy.

Okay. So thank you for having me and welcome, everybody. As you said, when I came on board a year ago, the prime goal was to transform and pivot the company from mobile, low-end consumer to a high power, right? And what I would say today is that pivot is pretty much done. So now the job is basically, from there, scale the company to, of course, bigger, scaled, more valuable and profitable company ultimately. The change was, first, go-to-market as we move from mobile, essentially focused on China, OEM, ODM, moving to the high-power market, which are AI data center, grid infrastructure, high-performance compute and industrial electrification. We have to completely pivot the go-to-market, right? So focusing on the hyperscalers, the XPUs companies, the OEM, OEM merchant power companies as well as anybody in the ecosystem. So that was a big change. The other change was capitalizing on the underlying technology of the company, i.e., GaN, because we were one of the leader or pioneer in GaN as well as high-voltage and ultra-voltage SiC to basically use those technologies to drive the disruption that we're going to talk about in the AI infrastructure. That was fundamentally more accelerating some of the investments that we already made, partly in getting GaN pivoted from mobile, as I said, to AI infrastructure as well as the ultra-voltage SiC. The last thing was to kind of make sure that the team is in place, a lot of work was done into upgrading the application team and the leadership team to build the culture. I said earlier that essentially today, the pivot is done, and you can see that in the revenue. I'm sure you're going to ask a question about where are we in that transition, right? At the end of the year, we started the journey where mobile was nearly almost the entire company, up to 85% of the company. At the end of the year, mobile will be -- is insignificant. So basically essentially gone despite having grown double digit per quarter over the last few quarters, right? So that's kind of where we are. Year #1 was transition the company, and now we can start to build from there.

Okay. So before we head into the details, perhaps can you give us a simple version of why is GaN and silicon carbide gaining traction today, right now? And also, if there is a rule of thumb to think about which part of the power stack, are we using GaN, which part of the power stack are we using silicon carbide? And how do you see both materials accelerating from here?

I assume you want answers without getting into a [ BE ] type of plan.

Yes -- simple framework.

Simple framework. So as I talked about, we moved to high power. The 2 main markets we're focusing on within the high-power markets are the AI data center and the grid infrastructure. We call that AI infrastructure. The reason we call that AI infrastructure is you can't transform data center without changing the grid and data center is the reason why you need transform the grid, okay? So that's number one. Two technologies that we have. One is SiC and the other one is GaN. I'll come to the -- why they are using some form of conversion or power conversion. But think about power equals current multiplied by voltage. If you increase the power and you don't increase the power, if you increase the power, you don't increase the voltage, you're going to increase the current. The loss, which is what basically makes the power inefficient is proportional to the square root -- square of current. So think about, let's take an 100-kilowatt DC load. If you run it at 50 volt, which is today, most data centers are using 50 volt, that gives you 2,000 amp. If you move it to 800 volt, you go down to 125 amp. So by construction, the system is a lot more effective. That's why there is a move to 800 volt. Like we've seen 12-volt to 50-volt back in the days of server. Now we're moving from 50 to 800 , right? So that's the first thing to keep in mind. SiC is a very good technology. If you're running at very high voltage in a consistent way, so where power gets sustained and where robustness makes a difference. So think about SiC as a very good technology for AC/DC conversion where you're converting high-voltage AC into DC. Very good for high-power AC/DC conversion, as well as anything that goes into grid tied application. We'll come to other applications that are kind of growing in the future, which is, as you raise higher voltage, circuit protection and power protection becomes a big thing, okay? We'll talk about JFET, which is another technology. But that's kind of the SiC. The GaN is very good in high power, high-density type of application where you're switching the frequency at much higher. So if you think about how you're moving DC-DC conversion closer to the GPU or XPU or compute load, that's where GaN kind of gets its way. So the best way to think about it is both of them operate in different part of the conversion. There is a bit of a gray zone. If you think about some BBUs, okay, some DC-DCs would kind of use GaN and SiC as well. But directionally, both technologies are very complementary to enable the grid rack conversion about the 800 volt, which I mentioned earlier.

So as you think about the rollout of the 800-volt architecture in phases, like is there a material that will see a greater adoption first because of like the sicar that we'll be implementing before we move on to the SST final state?

So go back to what I said earlier, right, so if you're trying to make more efficient systems and to deliver more power, you have to increase the voltage. So we see multiple drives, multiple inflection points, as we call them, up to 800 volt, okay? The first one is actually very interesting, is pre-800 volt. Most systems today run at 50 volt, which means the buzz bar of your rack is actually 50 volts, which has been around for quite some time. But as we deploy more data centers, as we deploy more powerful and high compute power data centers, you need more power. To deliver more power, you need to have higher density of power because the space available for the conversion is the same. So the first inflection is actually 50 volt, I would say, classic systems moving to high power, and that drives an acceleration of the SiC content. Essentially, when you go from a couple of kilowatts of AC/DCs to much higher power AC/DC replacing silicon by silicon carbide. So that drives the first inflection. But again, this is kind of 800-volt. Now you mentioned sicar. This is not very creative name, but it means what it is, okay? You're moving the AC/DC power shelves and the DC-DC power shelves from the top of the rack to a sicar. Why do that? First of all, you make space in the rack. You keep the 50-volt bus bar, but you then introduce the 800-volt. So think about this as introduction of the 800-volt. And while the first one, pre-800 volt is happening now, this one is happening in the first half next year. So what is the benefit of that? As I said, if you move the 800 volt as the output of your sicar, you can raise the power one step more, right? So we moved from a few kilowatts to tens of kilowatts. Now we move it to 20 to 30-plus type of kilowatt PSUs, right? Now you still need to convert this 800 volt back to 50, and you're doing that in the DC-DC space outside of the compute rate. So that's kind of the sidecar. So the sidecar is a very quick, not dirty, but a very quick way to implement 800 volt, get some of the benefit without moving to the third inflection, which is where you move the DC-DC inside the compute rate. When you do that, you're actually essentially making the full rack 800-volt native. The bus bar of the rack is 800 volt. When you do this, you have to move the DC-DC into the compute rate. When you do that, this is where technology becomes essentially GaN. So inflection 1, replace silicon with silicon carbide, get high content, thanks to that, raise the power level. Inflection 2 is you're getting one step further, 800-volt output, still kind of classic DC-DC conversion that kind of step the GaN to next level and then the -- to next level, a bit of GaN. Third one is really acceleration of the GaN. The fourth one, which is kind of far out is if you think about it, the sicar is kind of add-on to basically put something in between the grid and the data center. But ultimately, the plan is to improve the quality of the -- the grid or kind of modernize the grid, as we talked about, which is the replacement of the transformers by something I'm sure you heard, which is a solid-state transformer. At that point, you're delivering 800 volt straight to the data center. That's 28 and beyond because you need to upgrade the grid, which is a bit more challenging, okay, from a reliability standpoint and of course, from a security standpoint. But think about those transition in 4 steps, which I think from an investor point of view, derisk the transition from an AI data center side, derisk the transition, but also outline the importance of having both GaN and SiC because both technologies, as we talked about, live in their own space with some overlap in the middle. But if you don't have both, you don't participate to the big conversation. And if you're an hyperscaler, the only thing you care is how do I move 35,000 volt all the way down to sub-1 volt in the most effective way, which is less conversion, higher power conversion, higher efficient conversion. This is where GaN and SiC have something in common.

How do you think about the growth rate of this material market, GaN, silicon carbide? Or broadly, like what is Navitas content per kilowatt or per megawatt? How do you think about that?

So everybody is asking what's the content of SiC and GaN and depending on the stage and kind of very difficult to model, right? So we've published models to give investors kind of a reference, right? And the way I would talk about it is if you think -- so the models are based on assumptions. So the -- you have to take an assumption on how many data centers we're going to deploy between now and 2030. It could be 220 gigawatts, it could be 300 gigawatts, okay? It's getting closer to 300, okay, by the day. But it's kind of a window, which is a range which is reasonable.

And that is the assumption you're using, 200 to 300?

Then you have to factor how many of those will be 800 volt. How many of those will be 800-volt cycle, how many of them will be 800-volt native. And of course, this depends on time line. You can't -- when you started at a data center, you're going to stick to an architecture. You don't change it halfway through. So data centers that are deployed in a year from now, might actually well used sicar, okay? And actually, some of them will move to native. So you have to make the assumption there. But high level, what we've modeled is if you think about inside data center, the GaN gets you to $15,000 per megawatt. That's replacing silicon. So that's the content which is today in silicon, okay, moving to GaN. When you look at SiC, there are 2 portions. One is the inside data center, the AC/DC conversion I talked about, which is about 10,000 per megawatt, which includes both the SiC for AC/DC conversion as well as the JFET, which I mentioned earlier for the power protection. And then you have another 10, maybe up to 15 in the long run for the green modernization. And that SiC is at much high power, okay? So that's kind of 2,000 volt, 3,000 volt. So that gives you a sense, total is 25-ish, which is still a small portion of the total power that is being consumed per megawatt in the center. But as you guys understood from the discussion we had earlier about the 800-volt transition, this is basically gradually replacement of silicon. Second, we'll continue to exist for a long time, okay, for multiple reasons, okay? Because not everything switched on day 1. Nothing is actually a diesel switch and control and many technologies need silicon. But when it comes to this poor conversion, what you're going to see is gradually a replacement of silicon on the first stage and the second stage with GaN and SiC. And we'll talk about the third stage, which is as you get closer to the GPU or the XPU.

Okay. So you talked about it's going to be a gradual replacement. So what is the key metric that a customer looks at to drive that decision to do that switch?

First of all, it's the need that drives the transition. So customers don't decide if they're going to use 800 volt or not. It depends on the level of power density you're trying to achieve and the level of the rack level. So we always refer to what is driving all of this is the raise of the power density of the rack. If you want to do a high-power rack, you have no choice to book there. I think what people -- customers are looking for is, number one, do you know you're doing okay? It's a good place to start from, right? So we didn't start GaN and SiC yesterday, okay? Actually, Navitas started as a GaN company, pioneered GaN with another company that got acquired by a very big German company, kind of pionereed GaN into the first market that got GaN to mainstream. So we shipped 350 million units of GaN. Different market, but the underlying need is the same. It's how do we -- it was when the chargers were moving from 20, 30, 40 watts to 100 watts, okay? And you had to move from silicon to GaN, right? So that's number one. Do you know what you're doing, okay? You're starting -- are you planning to start? Or do you have years behind you? Number two is, are you ready? And I think particularly in the AI data center, but the grid as well because the grid used to be slow, so that's low anymore. Customers expect technology to be ready to use, and so you can ramp. Third one, which I mentioned earlier, is the fact that because you can't ignore -- you can't look at each conversion independently, having both the ultra high-voltage SiC, the high-voltage SiC and the GaN makes a difference. And we compete with different people on the 6 side and on the gain side. The SiC competition is traditionally the SI vendors that were focusing on EV that are pivoting to the data. In reality, this is not the same because the voltage are different -- so EV is lower voltage, high voltage, but lower voltage, where data center is 1.2 kV and above and the grid is more than that. The aspect of reliable technologies is also different. So we compete with different set of people. On the GaN side, we're also a different set of people. But people who have in both is actually a very small core, which I think is also very important and probably underappreciated for those don't meet customers like I do every day. I'm talking about the merchant powers, okay? The big -- the Deltas, the Flex, the Lion, the Vertiv, the [indiscernible], those guys do care. And of course, the hyperscalers, right? But those guys do care if you have both technologies.

Right. So maybe you can elaborate on that. How do you work directly with the customers, the hyperscalers? Or do you work more with the Flex Deltas, lions of the world?

All of the above. So if you think about hyperscalers, and they are different type of hyperscales, okay? So let's make a difference between the Microsoft of this world, the metals of this world, the Amazons of this world and the Google this well, which are more vertically integrated. I may have this kind of in between. NVIDIA is a GPU company, but it's actually more than a GPU company. They are doing racks, okay? And when you do racks, you cannot almost do data centers. So I put those last 2 in kind of into the same cap. When you talk about silicon to GaN, silicon to silicon carbide, moving from 5-kilowatt PSUs to 30-kilowatt PSUs, 50-kilowatt PSUs with [indiscernible] core, high integration of the DC-DC close to the compute load, they are involved.

Okay. The Google and NVIDIA.