Ambarella, Inc. (AMBA) Earnings Call Transcript & Summary

Good morning, everybody. Welcome to the second day of the EdTech Conference. I'm Ross Seymore, and we're pleased to kick it off. At this time I get to interact with Ambarella. We have CFO, Brian White, and Head of Investor Relations, Louis Gerhardy, with us. So we'll kick it off with Q&A led by me, if you all have questions in the audience here live. Just raise your hand, we'll get a mic over to you. If I don't notice you just wave that hand or yell out to me.

So guys, thank you very much for coming to Vegas and joining us. You guys just very recently reported earnings. So why don't we start off with a little bit of a summary on that. From my side, the quarter looked absolutely fine, in line with your guidance on the top line, even better on the bottom line, great gross margins. But the guidance was a little bit weak, a couple of million dollars. And I think what people are really focused on is the commentary on demand rather than supply. So just talk a little bit about what you're seeing in the forward-looking guidance that you gave on that kind of supply and demand side.

Sure. We did guide to revenue growth for the quarter in both our automotive and IoT end markets. But at the same time, we talked about some inputs from certain customers electing some pushouts and timing of deliveries. And it's not pervasive. It's at a few different customers, primarily in the IoT end market less in automotive. But I think what we're seeing is customers have been in an environment where there's been a considerable amount of supply disruption and they've had difficulty getting parts. Some of those bottlenecks are starting to break free and it's becoming easier to get parts, lead times are starting to shrink. And while the customers are talking about strong demand, we're not having customers talk about weakening of demand. They are getting concerned about all the dialogue around weakening macroeconomic conditions and the potential for recession. So while they're not seeing the demand decreases today, they're more concerned about what's going to happen in the future, and they're getting more comfortable with their supply situation. So they're focusing on managing their inventory. They're turning some of the attention that they didn't put on inventory management -- back on inventory management. And so consequently we're seeing some level of pushouts, and that's what we provided some transparency to on our way [indiscernible].

And when you guys talk about the demand side, just remind people about the lead time dynamics you talked about in your call because I think it's -- there's lead times, which is part A of my question. And then there's the flexibility to adjust the timing of delivery that you guys give to your customers, that's a little bit different than some of the broad-based semiconductor.

Yes, that's very true. So in terms of the lead time metrics, our lead times were running about 40 weeks. Those have come back into about 30 weeks. Normal historical lead times for us would be about 25. So lead times have been shrinking. But at the same time, Ambarella manages our requirements on customers taking orders on our books differently than I've seen in other companies, quite honestly. We're much more flexible in working with our customers. So even if a customer has an order on our books, we are receptive to request and move the timing of deliveries, trying to manage inventory and not get into a situation where we're shoving inventory on customers ahead of when they really [indiscernible]. And so I think that, that reflects a more timely impact to the revenue as we allow customers to adjust things. And sometimes there is a pulling as well. So it's not all pushouts. The dynamic flex of the timing of shipments. But we are much more flexible than other companies.

You mentioned on the demand side that it was a little more pervasive in your IoT side of your business, that's 75% roughly and the 25% in auto. Is there any geographic diversion that's of note and where you're seeing some of those actions as well?

No, there's really not. We've seen it in multiple geographies.

Last night, we also had some news about companies getting a notice from the U.S. government about the prohibiting shipments without a license of some higher-end seemingly kind of data center centric training, AI chips. Any impact of that on Ambarella?

We don't see any impact. We have licenses currently for all the products that we ship. So at this point in time, we don't see any impact.

Then the supply side of the equation, you guys had a relatively unique scenario where you had about a $5 million headwind in your second quarter from one of your foundry -- well, your main foundry partner being unable to ship. Talk a little bit about was it indeed a $5 million hit in the fiscal second quarter? And how do you expect that to improve in the back half of this year?

Yes. It was $5 million as we expected. That is improving in the current quarter, and we believe by our fiscal fourth quarter that will have a negligible impact.

Got you. And was there anything outside of Austin, was it just the Texas fab or the 14-nanometer side of things?

It's been limited to the 14-nanometer.

Got you. And supply in general, how are you -- if it's loosening up for other folks, any limitations that you guys have on obtaining either front-end capacity, back-end capacity at leading or relatively lagging nodes? Or has that stayed relatively stable for the last few quarters?

Yes. I think really the only challenge worth noting is the 14-nanometer that we talked about. Outside of that issue, we've been in pretty good shape.

Got it. Why don't we shift to a little bit away from the quarter into some kind of bigger picture items. And let's talk about computer vision, your CV business. I believe in the first quarter, you said it was about 40% of your revenues, if I remember correctly, and you said this year, you'll exit with it being -- well, for the whole year, it will be about 45% of revenues. Talk a little bit about where that's going between your IoT side of your business and the automotive side of the business? And what are the applications where customers are truly most interested in using that technology?

We're seeing good adoption of CV in both IoT as well as automotive. And certainly, at a 45% objective for the full year and IoT being 75% of our revenue, you have to assume that IoT is 45% or more along that journey. On the IoT side, the penetration has been strongest in the enterprise and public security area thus far. The value proposition there has been high and the time to get designs going is a lot shorter than some other areas. So penetration there has been strong, although there's still a lot of penetration available. So we'd say single-digit penetration currently. So a lot of room to grow.

And what -- talk about the features that CV brings to the public security market, whether you or Louis want to answer it either one. But just what's the feature set that makes that so appealing to people? I know it's a more expensive chip for them to buy, but the performance of the features it brings along with it are also obviously worth that price. So what are some of those features that appeal to people?

My microphone, is it on?

The light won't be on, but it is on.

Okay. So we provide the processing, the AI processing, the heterogeneous processor that's got -- provides both the eTOPS as well as the DMIPS and that chip will take the high-resolution input and the AI processor will then make decisions based on the software that sits on top of the chip. And that software for that IoT market, we're not providing. So our customers are providing it, and it's an increasingly diverse set of applications, some of which serve the traditional physical security market that we've always been in and will help augment the human viewing aspect of our chips that they continue to provide. But then there's new machine sensing-only applications that utility never existed in these cameras before. And so that would be applications, of course. When we first came out, everyone was talking about face recognition and all the different applications that you could use that for. But now it's broadened into traffic infrastructure and reading license plates and automatically determining if it's a stolen vehicle or know who's driving the car. It's used in school systems for attendance taking or threat detection. It's used in health and medical applications for knowing did someone fall down. Are they still in bed? How many times did they get out of bed? So it's an incredibly diverse set of applications and not only are our customers developing these different software applications, but there's a very vibrant community out there that -- start-up companies that are developing new applications. And I think throughout our product line -- throughout our CV product line, whether it's CV28 for the high-volume market or CV5 now for the high end, in the next year or 2, you're going to continue to see us address a lot of markets and new customers that we never reached before because our opportunity before was strictly human viewing, and now it's human viewing and machine sensing in one chip, and that's going to enable applications where you could have a camera in this room, not for human viewing purposes, but to understand, there's more people on the left side of the room than the right side of the room, and let's redistribute the HVAC system, the cool air to one side rather than the other, and we'll save 20% energy by doing that. So that's just one example of some of the new applications you'll see emerging.

So 2 questions on the technology side. How hard or easy does Ambarella make it to have these customer-specific software layers run on top of your chip. That's kind of openness of your platform, it has to be extremely appealing to it. Just how is that different from your competitors? And how do you guys go about making that relatively easy for the customer base?

Yes. We're focused on the high-volume opportunities. And so we don't flood universities and schools and things of that sort with design kits, but we're focused on identifying the customers that are going to go to high volume and can take advantage of the efficiency of our chips. And so this is a secret sauce. It isn't really discussed very often. It's a very good question, and we have SDK, software development kit that will license to customers. And in this SDK, it has the tools that allows them to port from common networks like onX or TensorFlow into the format that's needed in our chip. And there's actually some important proprietary technology in that tool kit that allows customers to do things like sparsification that further improves the performance of our chip relative to the other options in the market. So it's a very important part, and it does have differentiating features built into this SDK.

And I would think the second part that would be extremely appealing to people is the fact that everybody wants to talk about AI, but AI at the edge, intelligence at the edge, you guys are kind of personification of that. It doesn't have to go into the cloud and do some of the processing there. You guys can do the facial recognition, cloud recognition, whatever the machine vision application processing needed at the edge. So how does the power consumption and the need or lack thereof to go to the core play into the sales process.

Yes. So by doing more on the edge, and you're right, the perception we come across it all the time is that AI, it's been done in servers historically. And so what that means is you have to take all the camera data and transmit it into somewhere wherever the server is, middle of the network, edge of the network, do the processing and then send the conclusions back to the endpoint. We're trying to do as much of that deep learning, as you said, in the endpoint device itself and minimize the bandwidth that's needed, which can be expensive or by the way, cannot be available. And so you save on that bandwidth expense, which can be significant if you're streaming 24/7 into the cloud like some of these applications need. The other advantage is that as you go into the cloud and back, there's a latency and that latency can really cause problems with the performance of the application and its accuracy. And so by doing as much in the endpoint as possible, you minimize that latency. And then the third factor that I don't think is well understood is that by doing as much AI in the endpoint device itself, you can maintain privacy, maintain a much better control of privacy. Because imagine what's happening is the image of whatever object you're looking at is processed in the endpoint device, and that image never goes into the cloud where nefarious things can happen with your personal data. And so by doing as much of the AI processing in the endpoint device itself, you're helping solve the privacy problem. One example we talked about earlier is that, say, a health application and understanding, say, an elderly person in bed and how many times in a day did they get out of bed and did they fall down? Well, in the server type architecture, you'd be sending a video of them into the cloud their image, their personal information. Are they in their pajamas or whatever. And now with doing as much AI in the endpoint device, all that you're sending into the cloud is that a person just fell down. You're not sending a video of them or an image of them. You're just saying a person has fallen down, send help. So that's it. And it's one example of how you can protect privacy by doing more in the endpoint device.

And we talked to your customer base and potential customer base and let's stick with kind of the wave 1, the professional security applications for CV first, what's the most appealing aspect of what you bring? Is it the power consumption? Is it the performance, the privacy? I assume it's somewhat all the above. But when you get down to brass tacks about what differentiates the Ambarella solution some potential competitors, what do you find as the most common point that the customers like?

It varies a lot, as you said, by application. The smart home might be different than the enterprise market. And even within enterprise, there's different applications. But it really boils down to performance per watt and performance per dollar. Let's talk about performance per watt. You could use other chips and achieve the same function in endpoint device, but the camera would be the size of a shoebox, it's right of the size of your iPhone. And because you have to dissipate the heat and you need fans and increasingly sometimes water cooling systems. And it just is an economic, it's not practical. You can develop a proof of concept as a Series A or B, company go out and raise money with that. But when it comes time to going into production, you need a much more efficient silicon. So that would be the advantage, performance per watt or performance per dollar.

Speaking with the technology side, talk a little bit about CV5 and then, more importantly, CV3, what are some of the technology benefits they bring. And I know CV5 is ramping what a quarter ahead of plan. So kind of right now. And then CV3, you just finished a big round of sampling of that.

For those 2 chips, these are our latest processors, CV5 is just going in Q2, our July ending quarter. We had 3 customers and production status, and that's just started. It's our first 5-nanometer chip. And we expect that, that's going to be an important product for us in the next -- in the coming years. And CV5 is a product that we are now just demoing to customers, and we'll soon begin sampling and providing SDKs and EDKs, so customers can port their own software to it. But if we -- the CV5 is primarily targeted -- I'm sorry, CV3 is primarily targeted for the auto market. And should we get a win in that area, the opportunity for revenue is very significant. However, it would not enter production in most auto applications until calendar year '25 or '26. And in terms of the advantages, it's more of the same, but Ambarella's functionality is expanding beyond just camera perception and we're capturing much more processing value in those applications, whether it's ADAS or it's autonomous vehicle. So historically, in the automotive market, up to this point in time, we provided the AI processing for cameras and now perception processing with Oculii. But with CV3, we're still doing that for all of the cameras in the vehicle. Again, this is all in one chip. And then we're also doing the fusion processing layer, and we're doing the path planning processing, also all of that in a single chip. So this is going to command a much higher ASP than where we're at today, which we said on our last call was about $10.

Yes. Talk a little bit about that ASP side of the equation. If you went from kind of traditional human viewing applications that you've done in the past, to kind of -- I don't know if you want to talk about it in your ways version of it or talk about it CV to CV2, CV5, CV3, et cetera. What's been the path of ASPs percentage-wise, absolute dollar-wise, however you guys are willing to discuss it?

Well, I think if we look at the last year, at that $10 that we talked about on our last call, that represents a 30% increase in total company ASP over the course of last year. And that's really driven by this mix up to CV products away from video processing. And that's a trend that's been going on now since CV was introduced. And we expect the CV portion of our mix to continue to increase as a percentage of the total in the future. So that will create additional tailwinds for ASP increases as we move forward. On top of that, the new products that we're bringing out, we're just talking about CV5 and CV3. These chips have much higher functionality than the chips we've had in the past and will come to market at much higher ASPs, particularly CV3 in central domain controller applications in automotive. So we've demonstrated increasing ASPs, which is relatively rare. We have traction to increase ASP in the future with good line of sight to that. And if you combine that with unit growth, I think there is a significant opportunity for revenue growth in the future, combining those 2 elements.

And how that substitution effect work? If the vision processing goes away and/or gets replaced obviously accretive on the ASP side. But if I look at your overall revenue growth as a company, like ASPs are going up more than the revenues are going up. So something has to be going down in the background unit-wise. How is that trade-off working?

Yes, that's fair. And there's been cannibalization of CV with video processing business. particularly in the security area. But moving forward, the new applications that we have traction into, I think, will drive a lot of unit growth in the future. So there is some of that cannibalization on historic business, but moving forward, there's a lot of new opportunities that will drive growth.

And how does that work, the cannibalization if you guys are replacing a non-CV chip was something that's 30% higher price, that's fine. One for one. Obviously, you come out way ahead. But are there competitors coming up from underneath on the video processing side because over the years of covering Ambarella is it what's allowed you to continue to grow as you stay multiple steps usually ahead on the technology side. So that natural competition coming from below might take units from you, but the high end is where you guys continue to dominate. So how does that trade-off work? Are you seeing more competition at the lower end, and you're just staying ahead of them? Or are people -- is it more just a substitution AMBA for AMBA that's occurring?

Do you want to take that?

Sure. So we continue to innovate in the video processing and human viewing market. However, it's interesting in that I'll take one technology development example, which we call NeuroNet ISP where ISP is image signal processing, and that's an integral part of a video processor. And we found by leveraging in our CV chip software technology that we've developed that runs on a CV chip, it enhances the video processing technology for just pure human viewing function. In other words, what I'm saying is someone might buy a CV chip for a pure human viewing function because by running this NeuroNet ISP on the AI processor, it can make the human viewing experience better. And by the way, this is also applicable when you go to machine sensing because think if we can capture a better image of data -- the frame of data is what we call it -- used to call it a picture, but image or video. Now think of a frame of data. If we can capture it more realistically in the environment than someone else can, then we're feeding our AI processor integrated onto the same chip with a much higher quality set of data than someone else can deliver. And therefore, we should be kicking out more accurate results than a competitor's chip. And so it all blends together and I think the conclusion that it's there is that we're seeing a very strong shift to CV, and it may even happen in just some pure human viewing applications. So most of our customer activity right now is on CV. Video processor for us will have a long tail in certain markets like automotive, imagine driver quarters. that should have a long tail because, for example, most of our driver quarter business now is pre-install versus aftermarket. And pre-install is when you get into a program and you're basically shipping for 3 to 5 years, whereas aftermarket driver quarters, that business can come and go pretty quickly. So it's really TV and video processor, there's this overlap now with CV.

So last question on this topic. I think by my math, and I could be wrong, but it seems like with the 45% growth you're talking about -- 45% of sales that CV will be this year. Your CV revenues will be up, I don't know, 80%, 90% year-over-year, nearly doubling. The non-CV stuff, therefore, by definition, seems like it's down kind of 20%, 25% year-over-year. Two-part question. Is that rate of growth in CV sustainable going forward? And how much of the down, say, 20%, 25% on the non-CV is just a cannibalization dynamic, which you're fine with, that's still accretive to you versus something else, macro competition coming in and taking the sockets? How do you guys split the upside downside?

On the second part of the question, in terms of CV cannibalization, there's a really interesting dynamic in video processor business. If you look at last fiscal year with kind of this fiscal year. And so last fiscal year, we had 2 customers [indiscernible] for the human viewing security camera market for the most part. For last year, they were roughly mid-single digits to revenue, I think it was 7 or 8. I have to check that. This year, they're projected to be very low single digits, if not 1%. And so as that business has gone down, the -- and imagine they had lower ASPs. So their units were a much larger proportion than revenue was. And so as that business has gone down, it had an important impact on video processor unit demand. So you asked the question. How would you rank the different factors that are causing your video processor units to decrease this year versus last year? And you gave some numbers that are in the ballpark. And the #1 reason is [indiscernible] business going away. But #2 reason is cannibalization and the first reason is a little more than 50%. So cannibalization maybe it's 30%, 35% of the factor.

Got it. So we talked a lot about what you guys described as kind of wave 1 of CV, which was the professional security. Wave 2 is more of the consumer applications. Let's get into the fun part that's above and beyond that on the automotive side of things. You guys automotive business is about 25% of your revenues. You have what about a $1.8 billion pipeline or funnel as you've described it, you'll be updating that at the end of next quarter. I think, again, talk a little bit about some of the applications that's driven the revenues to the 25% of sales now and then how that morphs into that big pipeline that you guys have discussed?

Yes, there's been 2 areas. Number one, CV last year began to emerge basically in auto from close to 0 to a material part of auto. In fact, in our press release, we mentioned we shipped 10 million CV chips and more than 20% went into auto. And that's a cumulative figure. So you could imagine as auto CV continues to do well, the figure might be higher than that 20% suggested. So the factors causing it was initial penetration of CV into areas like driver monitoring, into front ADAS, into camera perception and in some L2+ wins. But also there's been an increased penetration of driver quarters into vehicles when they're produced. We call that pre-install. And so historically, as I mentioned earlier, most of our driver quarter business had been aftermarket, which can be much more volatile. Now the vast majority of that business is pre-install where the competitive landscape is much lower because we announced, for example, a win at Toyota, CV chip this quarter, CV25. And that is -- it's very difficult to get those wins versus an aftermarket product. And so it's a much higher quality business. And so to answer your question, it was the initial ramp of CV and multiple camera applications in a car as well as the transition of our driver quarter business from aftermarket to pre-install.

How do you guys think it the business transitions, people hear about drive quarters, both from aftermarket to pre-install hereabout the e-mirrors, things like that. I think the dream of self-driving, ADAS, L2+ and all of that is definitely there, and I know you guys are addressing it. But those other applications are happening now and should deliver some significant growth. So when do you think we transition from those kind of e-mirror driver monitoring, drive recording to more of the ADAS applications being the majority of your revenue?

It depends on how you define ADAS. A lot of people define ADAS is just the front ADAS product that mobilize sells. But broadly speaking, all of those areas would be ADAS and that they're used to prevent accidents from happening. And so -- but the front ADAS market, I think to answer your question, is currently the single largest opportunity in 3, 4 years, the single largest opportunity for us in auto becomes L2+ with the CV3 chip and these other applications, it's very clear right now, driver monitoring is starting to take off, but it's only about 2% of the vehicles today globally. E-mirrors is still at 1% or below, but we see solid activity in particular in Asia for e-mirrors, not so much in the U.S. or Europe yet. And so they all have a different cadence by far, the largest opportunity is front ADAS, and we announced a new design win this last earnings call. We need to continue to execute and announce more front ADAS wins in the next year. But they all look like very nice opportunities for us.

We have just a few minutes left. I want to stick on the auto for a little bit and then get into some of the margin and cash flow stuff. Brian, I think got 2 boards it up here. So last one on the -- well, 2 last ones on automotive. Oculii, the ability to do the sensor fusion side of things, handle the radar input, et cetera. Talk a little bit about how you integrate that into your feature set, especially in CV3 and what advantage that gives you versus more traditional radar implication or applications?

Yes. Just briefly, there's 2 paths in our vehicle today. There's 2 separate pieces of hardware. There's the camera, and there's the radar system. They both do their AI perception processing independent of each other. The data coming out of both of those boxes then goes to a fusion processor, which we're not in that business today, but we will be with CV3. And that fusion processor then combines the signals and creates a single point cloud that's updated at 20x a second or something like that. What we're proposing to do is to have one chip that does the processing for camera and radar and by collecting the raw data and doing a deeper type of fusion versus remember, 2 discrete pieces of hardware, we believe that not only can we save cost, but we can also extract much more valuable perception information because we're not discarding too much raw data. And so that's the essence of our long-term strategy with Oculii is to do, let's call it, deeper fusion than what's done today that's going to yield a lower cost and much higher perception. And we think it will have a profound impact on the sensor suite of these cars and help reduce cost.

The last question quickly on the auto side. You're competing in some much, much larger companies, the Mobileye side of Intel, NVIDIA, obviously and Qualcomm seem to be the 3 that get mentioned most frequently. How is a relatively small company [indiscernible] versus those guys, do you believe you can differentiate?

Well, it's a couple of things. It's higher efficiency, which is performance and power, and then it's scalability. So we offer -- we'll be offering a full suite of solutions that allow the customer to leverage on a common software across that. So it's efficiency and scalability is really how we'll win.

Yes. If you notice people trying to change the business model with this as well where some people want to do kind of the revenue sharing model down the road. Some people have more of the walled garden approach to the technology itself. The seemingly different go-to-markets technologically and even business wise. It seems to me, and correct me if I'm wrong, that one of the other advantages you guys have is you kind of open in both senses, you'll do whatever business model makes sense for you and the customer together and you, by definition, have a more open solution technologically. Is that correct?

Yes. Now, that's a great point. I mean there's some solution providers or at least one out there that provides a black box solution, hardware and software bundled together. Our strategy is to provide more of a menu of options for the customers. So we have a full software stack through the VisLab acquisition. And we can provide that to the customer and license that to them if they want it or we can license specific modules that help augment the software that they already have. So we're extremely flexible in what we are available to provide to the customer, whereas other suppliers are not.

Got you. And last question in the few seconds we have left, just on the financial side of things. You guys have, I think, beaten your gross margin guidance for about forever. And I know mix plays a big role in that, but more conceptually down the road as automotive and CV become a bigger part of your mix as the workloads you do within that become more intense as the ASPs rise, all the goodness that you just said, is there a natural tailwind to your gross margin?

Well, we've put forward a long-term gross margin model of 59% to 62%. And we just did 64.5% in the last quarter, and we've been running well above that long-term model. So the question would be are we -- our gross margins are going to decline significantly to our model level sometime in the near future. I'd say the answer is no. In the near term, I would expect our gross margins to continue in the range that we've been delivering, plus or minus. But why that long-term model is lower is really to contemplate the potential impact of the very large opportunities that we're pursuing in the future against large competitors. And so we believe that we need to have some flexibility in that long-term margin model to make sure that we can capture those opportunities in a competitive environment in the future.

Yes. And I don't think anybody can plan if you got many hundreds of millions of dollars of opportunity and it happened to be a 60% gross margin that people [indiscernible].

I think that would be okay.

All right, guys. Well, we're out of time. Thank you so much for joining us here in Las Vegas. And best of luck. We look forward to monitoring how Ambarella grows.

Thanks, Ross.

Thanks, Ross.