Perfect. Well, thanks. Thanks. Let's go ahead and get started. I'm Joe Quatrochi, the semiconductor semi-cap analyst here at Wells Fargo. Excited to have Ambarella joining us for this session. Louis Gerhardy, thank you for joining us.
Yeah. Well, thank you, Joe. Thanks for having us. It's nice to be here. It's a bit of a homecoming for me because I grew up in the PV Peninsula.
Oh, OK.
I was just thinking the last time I was in this location, I was probably this tall and watching Orky and Corky jump through the killer whales jump through the hoops in the pool that used to be right here.
Very nice. Very nice. Yeah, we miss that. So maybe to get started, you guys just reported, I think, last week, right? Solid results. Can you talk about just give us maybe a rundown of what you reported in the quarter, what's going well, where do you need to see improvement, or just kind of give us kind of the rundown from earnings?
Sure. Yeah. So it was our third consecutive quarter of beat and raise. Our revenue grew 30% sequentially and 63% year over year. And our guidance for next quarter was for seasonally down sequentially, but that was still $10 million above the consensus estimate. And we also expressed confidence in the consensus view that we can grow our revenue next year, calendar 2025, which is our fiscal 2026, by 15%. And so our results really stand in contrast to what a lot of other semiconductor companies are reporting. And I attribute that to new products that are occurring for us, in particular, new products that command a much higher ASP. And those new products are all AI inference chips. And AI inference this year will be about 70% of our total revenue. So we're seeing some nice trends there.
Maybe kind of double-click on what is the end driver of those edge AI chips?
The beauty about the edge and IoT is it's a lot of different markets, and even some large ones we don't serve. IoT markets would include handsets. It would include PCs. We're not in those markets. We serve everything else. What are those markets? It would range from enterprise or home security cameras. It would be access control products, things about ingress, going in and out of doors automatically, work environment, or in a retail store, for example. It would be portable video cameras, which have found some very interesting ways to actually use a lot of AI inference processing. We can talk about that if you want. Then there's customers like HP Poly, which use one of our more expensive chips for enterprise-class video conferencing, where those systems are using AI in a lot of very interesting ways also.
And so we're beginning to see this whole AI at the edge story for Ambarella started with enterprise-class security cameras. And now we're seeing other IoT edge verticals begin to adopt AI inference technology. And right now, it's a select group of customers. But I would expect over time that customer list using those more advanced products to broaden in the number of IoT verticals we can address with AI to continue to expand also.
OK. I think you said 70% of the total revenue this year come from Edge AI. I mean, can you break down maybe just kind of the end markets? What's the biggest? Just break down that 70% of the end market.
Sure. Yeah. So about 70% of our revenue just by application is IoT, and 30% is auto. So maybe within each of those, how much is AI to answer your question? And so I would say in IoT, maybe 80% of the 70% is AI already. And for auto, maybe it's 60%, something like that.
OK, and I guess when I think about the content growth opportunity there, I mean, walk us through the ASP that you get for driving an AI inference-related chip versus maybe your competitor or something like that. What's that differential look like?
Yeah, well, go back maybe even bigger perspective and go back to pre-AI, what were our ASPs, and that would take you back five or six years, and our average selling price before we had any AI products was maybe $6-$7, in that range, and today with AI, it's 70% of our revenue, so we still have 30% with those lower-priced chips. Our blended ASP is around $13, so it's gone up quite a bit, but we do expect our ASP to continue to rise going forward, and the reason is that today, about 70% of our revenue is AI inference across both markets, and all of that AI that we're doing today is computer vision, which means at the highest level, you're doing object detection and classification of different things, and the utility is high, and that's a great growth business.
We've introduced within AI, there's a new generation of AI products that are capable of doing transformer network processing. That enables, by the way, Gen AI to be run on our chips. It enables vision language models where there's camera input and text input, like LLMs, ChatGPT-type stuff. CLIP is another example of an application that would run on these chips. Again, this generation of chips will command ASPs anywhere from $25-$400. None of those, as we speak, are generating revenue. The first product in the family, which is our CV7 family, is expected to begin to generate production revenue for the first time this quarter and become a material contributor to our business next year. We're quite confident that on an annual basis, our ASP is going to continue to rise.
You might say, it's fair, like, OK, your ASP has risen for five years now, and it's doubled. Why hasn't your revenue done better? Well, one of the problems that was happening is that in our legacy business, we were bleeding units. And so now that the legacy business is 30% of revenue and in a controlled descent, AI is 70% of revenue with all these new products coming on board, we're in a much better position so that this year our ASP is growing. And our units, actually, this year for the first time in a while have stabilized or will grow. And next year, you should have both ASP and units growing. So in essence, if I had two minutes with a portfolio manager, that's what I would tell them about.
Yeah. And that 30% of the business, from an absolute revenue standpoint, does it just kind of bounce around here at the current level?
It probably has a long tail because from here, what is the growth rate? Maybe I would say from a five-year revenue CAGR perspective, that video processor or legacy business, as we call it, maybe it's a negative 10% to negative 20% CAGR on a five-year basis. Somewhere in that range is the right way to think about it. It's very different than calendar 2023, which is our fiscal 2024, when the cycle was really difficult for everyone. That business declined 50%. And our AI business was much better off in the downturn. Yes, it had a couple of bad quarters, but nothing like video processors, which were down 50% year over year. That stopped that type of decline.
Yeah. I mean, I guess on that basis, at least you kind of found this more or less kind of bottom that you can. You're still working down. It sounds like it still is going to work down, but now it's no longer.
It's a smaller number.
Yeah. That's right. That's right.
Maybe let's shift gears a little bit. Can you talk about automotive, the opportunities there? Just help us understand what that looks like.
Yeah. So we have two different types of auto business. The first one is 100% of our auto revenue today. And these would be chips that sell anywhere from $10-$50. And think of this business, let's call it ADAS or safety. It includes a lot of different things, by the way. That business, five-year CAGR for it might be a positive 15%. It's a good business. It's like our IoT edge business we were just talking about before. And that business, to give you an idea of some of the applications, it would, again, it's a wide variety of things, some of which are beginning to show some very encouraging signs. For example, driver monitoring in cars is one area. That's actually how we got started in auto. But then you have other areas like electronic mirrors, which basically become displays and improve the field of view.
If you take the left and right mirrors out of the car and put them as displays inside the car on the A-pillars, then you reduce drag. There are advantages that are a lot more than just improved field of view in difficult lighting conditions and improved blind spot detection. You also actually can improve the range of an EV by reducing the wind drag. That's E-mirrors. There are encouraging things. We announced a win with Dongfeng Honda, a joint venture company in China, for exactly that, the left and right mirrors this last quarter. Another market would be driver monitoring. We'll have some news on this market in the not-too-distant future. It's taken a while for that market to take off, just like E-mirrors.
But again, there's some encouraging signs, and there's regulations in certain regions of the world that will now make that market finally happen. And so driver monitoring is typically one camera facing the driver. But we do see applications emerging now where there might be three cameras inside the car in addition to all the cameras outside the car. And those three cameras, of course, monitoring the driver is the most important thing. But we'll also be looking at occupants. And these interior cameras can be used for a lot of purposes in vehicles ranging from seatbelt tightening if an accident's happening or understanding which direction to blow the airbags based on where the people are seated, or if there's an animal or a child left behind in a car, or driver scoring, driver training.
And even insurance may give you a lower rate if you have driver monitoring in your car and you're willing to share that information. So many use cases for driver monitoring. Another market in this, I'm still talking about our first auto business that's all of our revenue today, would be front ADAS. And that's a market that Mobileye dominates. Mobileye is not in these other markets, but Mobileye is in this market. And we have some business there, but not much. And then there's a few other applications for these ADAS products where the ASP is $10-$50. The big opportunity in auto, the second segment, is currently zero revenue for us. We don't expect first revenue until calendar 2026. And these are chips that will sell for $100-$400.
They are the central domain controller that sits on top of the active safety autonomy domain in a vehicle. We have some wins today, for example, in commercial trucking. We talked about activity with Leapmotor. There's been a huge focus of investors on this market because the value of winning a Western OEM would significantly change the slope of our growth story I articulated of 15%-20% type range because the lifetime value of these could be lifetime values over five or six years could be $500 million to $1 billion. In the past, lifetime value of $20-$30 million would be really good for us. A huge increase in the opportunity. We're making a lot of progress in this area, but we still have not yet announced a Western OEM that would provide that size of opportunity.
So maybe double-click on the use case of that and the importance of that and what made you win your initial deal. And how does that compare to the Western OEMs that you're trying to work with?
Our proposition in this market is similar to the other markets in that it really comes down to our performance per watt. And for example, that can increase the range of a car or allow an OEM to put fewer batteries in. And we can also reduce the cost of the bill of material because while a chip price might not be different than some of the others, we don't need liquid cooling. That adds a lot of weight. It lands a lot of cost to the car. We're more efficient in how we utilize the DRAM. Our form factor is smaller, so you can actually fit our chip on the back seat of a car. You don't need a special box that takes up trunk area. So there's a lot of reasons beyond just comparing the chip price.
But we expect a win in this market because of our efficiency and superior performance per watt because we're coming from a place where we developed all this technology we've been talking about for these edge AI markets. IoT devices that are the terminal device in a network, they have to be very power efficient because they're often running off a battery. And they have to be economical so you and I or businesses can buy them and practically deploy them. And so our AI inference technology has been built from the ground up for that edge market, whereas there's other edge AI technology that's being taken from other markets and trying to be reused in some of these markets. And because we're so purpose-built for efficiency at the edge, we think that's going to be why we win in this market eventually.
You said the main competitor there is Mobileye?
I'd say for the type of product we're talking about now, which is we call it CV3 as our product, but the application would be L2 plus autonomy. What is L2 plus autonomy? Think of it as eight cameras and maybe five radars, and it would deliver FSD-like performance using Tesla as a benchmark that we all understand. The challenge of this L2 plus market is that it has currently not penetrated the overall auto market as quickly as anyone anticipated, not just us, but other players in this market. For example, I'd say L2 plus this year is maybe 4% of all the cars produced. So I don't know what the right number is. 88 million cars produced this year, and 4% of that is L2 plus. Probably three, four years ago, we and many others were saying maybe it would be 10%-15%.
It's not that yet. The answer is why? Maybe a possible explanation is that premium tier that can afford to pay $8,000 or $15,000. Maybe it's starting to get saturated. In order to get L2+ more broadly deployed, you need a lower-cost system. We believe we have a solution for that, not just because of our efficiency, but because we've got a scalable family of chips we've introduced from a chip called CV3-AD635 at the low end to CV3-AD685 for the L4 commercial trucking. The advantage to OEMs is that they can deploy their software across all of those chips, reuse it, rather than use a different architecture for every model or tier of vehicles, which would require them to go through a very significant software porting exercise. It's expensive. That's the state of the market.
We do think we have some solutions that will help solve this problem of L2+ market adoption being slower, and that's the first thing that has to happen before we can start seeing the market take off for us.
So, is there like a TCO benefit of your chip relative to what has penetrated that 4% of cars? I guess how do you think about that in reducing kind of maybe it's not the cost of the chip, but it's the value that you're providing, right, across the board?
The suppliers in that 4% today, almost half of it belongs to Tesla, who did their own chip. Then most of the rest belongs to a variety of Chinese OEMs that are moving very, very fast, one of which we're engaged with. That's the current state of the market. We've got an angle in China, as I mentioned, but we've got also strong relationships with two of the major tier ones in the Western world, Bosch, who's the largest automotive tier one in the world, and Continental, who's, I think, in the top 10 for sure. Both of those companies are bidding on OEM programs based on our CV3 chip. We've got irons in the fire, most of the continents of the world. The biggest frustration in this market has been that there's been program delays. There's been cancellations.
There's also, by the way, new opportunities that have emerged recently as well that we're still ascertaining what it means for us. It's a very dynamic environment at L2+, and we're still very focused on it.
Okay. That's helpful. I mean, maybe kind of sticking with auto a little bit longer, talk about your penetration in China EVs. I mean, how should we think about that as a percentage of the auto revenue today? Where do you think that could go?
Yeah. Well, CV3, as I mentioned, it's 0% of our revenue today. It won't start until next year. But we don't really track our business like how much is EVs and how much is ICE. But we do have, because China is now, what, I don't know, 30%-40% EVs sold this year. That's just off the top of my head. I don't know if that's right.
Physical 50 right now.
Fiscal 50. So we're in many of those vehicles. For example, this quarter, we announced XPeng, which is one of these new NEV companies in China, emerging very rapidly. They're using one of our chips for the rear view. Coming back to e-mirrors for a minute, they're using one of our chips for the rear view. I'll call it mirror, but it's no longer reflective glass. It's a display, right, with the camera and the shark fin looking back, which significantly improves the field of view, especially if you have kids in the back seat or a dog or whatever. So that's an example. And we have many other examples like that. I believe Dongfeng Honda when it was an electric vehicle. We announced that last quarter also. That was the camera monitoring system I mentioned earlier. But we have both ICE and EV. Don't really separate it that way.
I don't even know the answer.
Yeah. I guess as you think about the opportunities to kind of grow your pipeline in China, given that they're definitely focused on innovating faster, can you talk about that dynamic? I mean, have you seen that kind of enter your pipeline fairly significantly?
Yeah. Well, it has lots of interesting implications. First of all, because we're bringing advanced technology to the market, being involved with China and the access to turning product cycles quickly and getting that feedback is very valuable. But here's the other side of the argument. We provide an estimate of our auto revenue once a year. We do that in November. We call it our automotive revenue funnel. It's actually a very rough estimate of what our auto revenue might be at that given point of time. And it's been changing a lot recently. But if you look at China in that funnel, it's only 15% of the funnel, which China is 15%-20% of our revenue today.
You might say, "Well, how does that make sense given what you said about moving faster and adopting technology sooner?" One thing about the business in China for us, especially thinking about the funnel, is that XPeng win I was talking about, that might only last two years. In the Western world, it might last five. And so we have two years of our revenue estimate for that one program in our funnel. But we don't have anything for years three, four, five, and six because we'll put it out for rebid in two years. That RFQ is not out yet. And our discipline says, therefore, we can't put it in our funnel.
And so my point is that China is probably underrepresented in our funnel because you only get two, if you're lucky, three years of visibility for any win, and then you don't have anything in the back half. The Western world is very different.
Right. But maybe push back a little bit on that because that, I guess, requires you to then rebid every two years, right?
Yeah. Very fast moving.
One of the kind of nice things about the automotive industry is that these win cycles are longer. And so it almost becomes analogous to, say, a smartphone where you've got to win your socket every other year or every year. How do you kind of balance, I guess, that dynamic?
That's fair. But remember what you're getting in exchange for having to rewin it every two years is you're getting these quick product cycles and that feedback very quickly. So I don't know anyone that's really serious about overall global auto business that's not super engaged in China for that reason.
Right. Right. Okay. Thinking about just the go-to-market in China, do you have to? Have you been focused on trying to make a China for China dynamic? I think remind me, I think you guys are kind of more on the leading edge in terms of node technology. How does that dynamic affect that?
So there's no restrictions for us. Our supply chain, by the way, our own supply chain, not our customers. Our own supply chain is that we purchase wafers from Samsung, from their factories in Texas and Korea, and we do assembly in Taiwan or Southeast Asia, so we're pretty well hedged from some of those geopolitical risks, at least in our own supply chain. I think when you look at our customer supply chain, it's a different answer, and it varies widely based on the customer because they might be assembling boards in China, but in terms of geopolitical risks, yeah, there's just a lot of them. I would point out five, six years ago, we had two customers that were 30% of our revenue in China, and given what happened with geopolitics back then, they're now 0% of revenue. They went away very quickly.
So we've dealt with this before. And so we've thought about the business we have and the business we're trying to win. And there's a lot of things to think about. We have contingency plans for worst case, but it's still a market you need to be engaged in on a global basis, given the things we talked about before.
Okay. Okay. Maybe shift gears a little bit back to IoT. Sorry. Can you talk about the strategy there to leverage the leadership in edge AI and then kind of what you've honed in on? We talked about enterprise security market. How should we think about the movement into adjacent markets, kind of leveraging that expertise?
Yeah. So Moore's Law, the cost to do newer, more advanced nodes, becomes very expensive. And so one thing that our R&D engineering team has done very successfully is to our five-nanometer products now, CV5 as an example. They've created this SoC where there's two major functional blocks. There's the AI accelerator, and then there's a perception engine that collects the data from a camera. And then the AI accelerator takes that data and makes decisions based on the software that sits on our chip. Well, if we designed CV5 with just one vertical in mind, it would be very hard to get an ROI on it. And so this is very difficult to do, what I'm describing, is that CV5 has been designed with these dedicated accelerators, but there's enough programmability in the chip so that Rivian is using it in an auto application for R1T, R1S refresh.
Motorola Solutions is using this chip in enterprise-class security cameras. HP Poly is using this in an enterprise-class video conference system, and a company called Insta360 is using it for portable video cameras, and by the way, the AI applications that they're running on these are all so different than each other. And so that's to serve this fragmented IoT market, you have to have that very clever design methodology where you leave enough programmability to accommodate a variety of different markets, and I think with CV5, we've demonstrated we can do that. But it's a constant challenge with every new product, so we talk about our new product waves, by the way, new products in waves, and the two waves that we clearly see right now would be CV5, which I mentioned, and about 70%-80% of the revenue opportunity from that is this IoT market.
But then the other 20-30% is coming from auto. The second wave we see is CV7. And same thing. Because of this programmability we have in the chip and the ability to serve multiple markets, we expect 70-80% of revenue to be in IoT and 20-30% to be in auto. It's not until our third wave where that central domain controller for L2+ that needs to be a pure auto chip with what's called functional safety, which is an automotive thing integrated into it. And it's not until that third wave that begins in calendar 2026 does that product family become specific to auto. Maybe autonomous mobile robots could use it too because it's like an autonomous car.
But that product family is much narrower than, say, CV5 or CV7, which can serve multiple applications in the IoT space as well as auto.
Okay. That's helpful. Maybe the roadmap in terms of node technology, I think you said you're at five-nanometer. How fast do you need to progress to three and eventually two?
We're going to skip three. I think there's a lot of problems that are known about that, but we have announced, and it's already built into our OpEx run rate because it is expensive. We have announced a two-nanometer project, and in that case, we have a lead customer engaged with us who will pay some of the bills, and that will be really the foundation for a product family or product generation that we haven't really articulated what it will be other than at this point in time, all of our products have leveraged our heritage with perception and collecting data through the lens of a camera, and then the AI accelerator has done all the processing behind it. The AI accelerator is becoming such a large part of our chip that Fermi Wang, our CEO, has hinted that in the future, maybe it's at two-nanometer, not the first product.
We will have a chip that's a pure AI accelerator without the perception integrated into it. In other words, you can imagine that's going to open up our opportunities into a lot of non-camera related areas. Don't get me wrong. Camera is super important for all the products, all the revenue we're going to have for the next two years. But because our AI accelerator is so respected as a very efficient edge AI processor, you may see us come out with a product for like that.
Okay. That's interesting. Maybe in the couple of minutes we have left, and we've talked a lot about revenue and opportunities and growth. Let's talk about long-term margin opportunities. Where do you see that kind of going? Obviously, nice content growth, ASP increases.
Yeah. So our revenue expectation CAGR is that our SAM is growing 17%-18%, IoT a little lower than that, auto a little faster, and potential for upside depending on the degree of success with CV3. Our gross margin, non-GAAP gross margin target is 59%-62%. And currently, we're operating above it. And our guidance has been that we'll walk gross margins down slowly in the next year into that range because we want to come into our range so we can drive more positive operating leverage. Because we're not a fixed cost store. We don't have a lot of depreciation in our COGS. We are an R&D story, right? And our R&D number is high because we've made this upfront investment in AI that's just starting to pay off. And then on the operating margin line, our goal is to get to 30% non-GAAP operating margins.
We've been there before. I think it was 2017, 2018, we were at 36%-37% operating margins. And this last quarter, we just turned the corner in profitability. So the goal is to drive with revenue growth a lot of positive operating leverage. We're not a gross margin story. We're an operating margin story. And the way to achieve that is with revenue growth.
Okay. That's helpful. Maybe in the few seconds we have left here, I mean, anything that you think we talked about a lot today, but things that investors maybe don't understand or underappreciate from the story?
Yeah. I think there's been a huge amount of focus on CV3, which is wave three, is that L2 plus central domain controller that I said starts in 2026. And I understand why because it can significantly change the slope of our growth curve. But I think it took attention away from wave one and two, which is CV5 and CV7, CV5 of which is on a roll right now and driving this outperformance. And then CV7, which really starts this quarter and becomes material next year. And so I think coming out with CV3 took a lot of attention off of those products, which are going to drive our growth more than 100% of it because keep in mind, they're offsetting the decline in the legacy, right? And so maybe a little more attention could have been spent on those. But for sure, CV3 longer term is very important.
Okay. Perfect. Well, we'll stop there.
Yeah. Thank you.