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

All right. Great. We're going to go and get started. Thank you, everybody. I'm Joe Moore, U.S. semiconductor analyst for Morgan Stanley. Very happy to have with us today the management team of Ambarella, CEO, Fermi Wang; and CFO, Brian White. So thanks, guys, for joining us.

I wonder maybe if you could start out, I think most people know the story, but just computer vision transition is now up to 45% of your revenue this year. Can you talk about the computer vision market, what it is that you guys are trying to do and we'll go into specifics.

Right. Computer vision is really a technology we've been focused on, and we are trying to apply the computer vision to the, we call IoT market first and that transition was very successful. We talked about that in the professional security camera. We have quite a few design wins. Particularly if you look at outside China, majority of the professional security customer using our chip for high-end and mainstream. And also we start talking about on the consumer security camera side, we announced Ring and Vivint as our customer. And that market is growing slower because it's very sensitive to price, but eventually I think that transition will complete in a few years. I really think that with the value added with computer vision technology, this transition will continue and eventually the majority of the products our customers build will be completely vision-based. And of course, we're going to talk about auto later, but definitely, that's another market we focus on.

Yes. I mean I feel like this is one of the really exciting developments in semiconductors is the sort of innovations of machine learning and deep learning has brought to computer vision analytics. And I think a lot of people that are competing with you are using repurposed graphic chips, applications processors, programmable solutions versus your more dedicated solutions. So can you talk about that? And maybe in the context of as this has moved from more of a heuristics model to more AI, deep learning, how have you guys enabled customer success?

Right. So first of all, I think you're absolutely right. In fact, even in the IoT market, the first generation of our customer product was based on FPGA, GPUs, those kind of technology and because it's easy to program and quick -- can have a fast time to market. But right now, in the whole security camera market, our solution pretty much is a dominant supplier in that. And I expect that we will see a similar trend in auto. Right now, first-generation automotive, Level 2+ car, is really dominated by the GPU type of solution. But I think that the power consumption of GPU is a well-known problem for that. And with the EV development in the automotive space, the power consumption obviously is one of the most important considerations for the autonomous driving. So I think that is playing a major role. I think the architecture that we provide really address that problem. And on top of that, I think we are really at the first inning of this AI transition, right? We're seeing so many neural network innovation in the last 10 years, and especially the type of neural network being used in autonomous driving continue to evolve, right? I think at beginning, heuristic was important for the Mobileye and the first-generation product. But today, even Mobileye transitioned to neural network. There's a new type of neural network called a transformer, which was not popular in auto but recently -- just quickly become the most popular neural network and which demand huge amount of performance resources and very difficult to implement. So this kind of innovation will continue. I think the AI technology will continue to evolve, particularly on neural network development, and that will require dedicated performance, dedicated architecture to address that change.

Great. Maybe we could just get some of the near-term issues out of the way. You guys have had a couple of issues in the last couple of quarters around supply chain. Can you maybe first speak to an issue you had with getting Samsung wafers? Are we clearly kind of past that? And I think the fact that, that impacted you, how does that affect your customer conversations around supply chain continuity and things like that [ with major ] focus?

First of all, the supply chain shortage or the sensor foundry problem will be behind us at the end of this year. And I think that will not be an issue anymore. On the contrary, the other side of the problem popping up, right? In the last 3 years, most of our customers really tried to accumulate as much inventory as they can. But now we're seeing a different side of the story. When our lead time dropped -- at the peak of COVID, our lead time was 40 weeks, now dropped to 30-plus. And before COVID, it was 24 weeks. So when the lead time drops, the biggest reaction from our customer is, I don't need to keep that amount of inventory anymore. So I think the conversation with our customer is how -- what's the fair inventory level they want to hit and how we work together to reduce to that level. And while total customer demand is not a big concern yet, and I think really how to adapt to the new inventory level is the biggest concern.

Got it. Okay. Great. Maybe if we could focus on some of the recent product development. You talked about CV3 earlier this year. Can you talk about the importance of that product? And then we'll get into some of the recent announcements about customers there.

Right. CV3 is so important because this is the first chip we developed to target the single-chip domain control market. And when we say single chip, it is not just a camera. This is really all the perception you need for a Level 2+ car or Level 3, Level 4 car and all of the path planning, high-level functions and also control the car. So this is single-chip solution put us in a way to compete with Qualcomm, NVIDIA on this really big market opportunity. And it was so important to invest, of course, heavily on this project. And fortunately, we taped out the chip. In June and July, we taped -- in fact, in June -- sorry, we received the silicon in June and we demoed our first CV3 to our customer in July. And in the demo, we talked about -- was putting multiple camera to our chip and deliver the performance and the power consumption target we promised in the past. And that really is a key milestone that enable our customers to say, okay, you deliver the performance target and then we can start talking about working together on the project. So that's why the Conti has this press release just recently.

Yes. Maybe you could talk to that a little bit. So Continental, a major Tier 1, adopting CV3 for future development. What's the path to turn that to revenue? I assume now there's -- they have to sell that to their OEM customers. It seems like an important endorsement, but what's the path to drive that?

So just since that announcement, we have multiple meetings with the OEM customer with Conti together. And that's definitely -- in there, I think, is really Conti's presence definitely give our customer more credibility as well as confidence this is a solution that will be able to manufacture at the automotive grade. So this is -- we are not in the evaluation process anymore, we are really at a bidding -- project bidding process that we talked about. And any project that we bid on, we win, it will take probably 3 to 4 years to production. So we're talking about 2026 revenues.

Great. Okay. And then can you talk about the process of ramping CV3. I mean given that it's still a couple of years from driving real revenue, when you win customers like Conti, does that mean more expense in the short term that you need to invest to get those programs going? Or just how do you sit with regards to that?

Well, in fact, in the Conti relationship, in fact, we have been spending 2 years' time with Conti before this press release. So in fact, we have spent a lot of engineering resource to not only just Conti, but allow the customer to support and to understand our technology. It's helping them to put software other than CV2, give them CV2 board and there are tons of evaluation on the neural network side. So in fact, we would -- our current engineering resource already have embedded a supporting structure to support those companies already. So in my mind, I think, in addition to Conti, we probably can support a few more customers with similar scales. But if we really become successful, then we'll probably have to talk about additional resource, but we are not there yet.

Yes. I mean you've invested hundreds of millions of dollars to develop this. Still a couple of years from revenue and it seems like a pretty big exciting market.

It is. Computer vision, we said cumulatively we'll be putting $800 million worth. We're a really small company, right? $800 million R&D dollar into computer vision alone, and we're betting on this is going to drive not only this autonomous for the next 4, 5 years in the future in the robotic application, which I truly believe there will be a huge market opportunity for us. And a chip like CV3 can be a domain controller for robot also. So I think that the computer vision investment is going to pave the way for our long-term revenue growth.

Great. And then you mentioned CV3 takes on a bigger role with regards to it's not just vision, but you're taking in other sensory elements. And I know you guys have made investments in radar with the Oculii acquisition. And you announced -- I think, yesterday you had an announcement about centralized radar. So maybe you can talk about the importance of radar in the ecosystem and what Ambarella is bringing to that.

We believe that for any autonomous driving car, it doesn't matter whether it's Level 2 or above, you need a multiple sensor modality to provide enough safety and backup [ storage ]. So -- and when we -- we're a camera company, so we are looking for other sensors that can really complement to our technology. And we identified radar. Particularly 4D imaging radar is the best option for us many years ago. And last year, we identified Oculii really has the best match from a technology point of view and a cultural point of view and then we acquired them. Yesterday is really a major milestone that we announced that we proposed a different architecture we call the centralized radar. If you talk -- listened to other 4D imaging radar companies, they are talking about using this 4D imaging radar for them is a huge antenna array, 48x32 antenna. Very powerful machine. You need a big RF chip and plus a big DSP sitting on the edge. We think that's inefficient. We think that although we agree that 4D imaging radar is important, our approach is instead of this huge 48x32 array, we use 8x6 array using a specified antenna array. But although the data array is lower, we use -- when you send to the multiple radar input into a CV3, we can use algorithm in a mathematic way to reconstruct the point cloud to achieve similar type of resolution that other 4D imaging radar companies are talking about. The benefit of that is clear. On the edge, your radar module, you don't need a powerful RF chip anymore. You don't need DSP. And the data can be transferred to CV3 and CV3 can process that. And so the cost and the power saving is significant. But more importantly, on the technology side, because you can do -- you can build the data point cloud both for video and the radar on CV3, it can do a low-level sensor fusion, that gives you much better performance and resolution at the end. So we offer now which is better cost, but also better technology, in my opinion.

Yes. I mean we had a competitor of yours yesterday here talking about a much larger antenna array. I mean it seems when you say it the way you do, you just have different algorithms to use a sparser set of antenna. It seems easy. How defensible is that? How -- the innovation that Oculii brings you, how unique is that going to be to Ambarella's solutions going forward?

It will be -- it's unique right now because we haven't seen anybody has this approach. A lot of people say, how are you going to defend that? There are a lot of smart people. Yes, there is. But when we started video processing and the video compression 18 years ago, people said, well, how defendable your video processing is. And we are still defendable. Even today, we have a better video processing technology, better than NVIDIA or Qualcomm. And that's because we have a dedicated architecture, better -- dedicated algorithm team to work on video solutions. And I think this radar technology, particularly, is very defendable because it really requires a huge amount of algorithm development as well as silicone like CV3. You have to be both. When we acquired Oculii, it's very clear to Oculii that the algorithm is great but without a silicon like CV3, they cannot achieve what we just proposed yesterday. So when these 2 things come together, it's really powerful. I think definitely any technology can be copied and people -- there are lots of smart people out there. But I think our advantage is we only want to put a video and radar together. At this point, we are first runner on this technology. Hopefully, that will give us enough lead time.

And you had automotive investors in Oculii before you bought it, automotive OEMs. So definitely, it seems like there's interest there. Okay. So I guess if you pull all this together, I mean, I've talked -- I've asked you a lot of questions about CV3 and sensor fusion. But you also have a very nice auto business doing surround view video, in-car driver monitoring, digital review mirrors. There's a big opportunity for you guys. So maybe talk to those opportunities. And it feels like, there, you're actually sort of the incumbent. At this point, you're pretty strong in those markets.

Right. Which is true because if you look at -- we talked about sales funnel, our auto sales funnel, at our earnings call, too, and there are $2.3 billion sales funnel we talked about. Auto is the $2.3 billion. There's $800 million, ONE business. When we talk about that business, really the business you talk about, CV2-based, e-mirrors, recorders, surround view and those kind of -- let's call it traditional automotive applications that can be addressed at CV2. But the difference is in -- I would say before CV2 family, it was really for the human viewing application only. People want to display the video. But now even for the e-mirror, they want to put computer vision there so they can do object detection, give you early warnings. So it's become part of, I would say, cheaper versions of autonomous driving. But CV become more and more important. Even recorders in Japan that people are talking about they want to add computer vision to it because they want to make sure that when they record video, they can erase the license plate because of privacy reason. And in fact, in China, people already passed the law says that in the future when you record any video down the road, if you see the license plate you need to erase it for the privacy reason. So technology is going to become not only providing safety and security, but also we can help on the privacy side.

Great. So I guess, pulling all this together, I mean you talked about the funnel, the $2.3 billion. That's a 6-year kind of revenue pipeline versus for a company that did less than $100 million trailing in automotive revenue. So pretty big growth ramp that you guys are talking about. Can you talk maybe about the methodology of the funnel as I know there's some probability weighting and things like that. How -- and a lot of companies talk about automotive backlog with really big numbers, and I think you guys are more precise than some.

Right. So when we look at the funnel, which carve into 2 portions. The first portion is the ONE business we talked about. The ONE business is the project that we got a letter from our customers say, you are awarded by this project. And there's how many cars they're going to build with this technology in how many years. And the price is negotiated, the project kicked off. So we know there's a real project behind it and we know the forecast for our customer. So the only thing that we are putting our judgment is how fast the thing can go into production, how many cars they can ship over that period of time. So there's a factor we're multiplying, of course, less than 1 factor that we're putting there to make it more conservative than the forecast we get from our customer. But on the pipeline side, we really is the project that we're bidding on. For example, there are some projects that we're bidding on with our partners with the CV3 or CV2. Those projects not awarded yet. So the way we did it is we put even bigger multiplier, lower than 1 multiplier to reduce the risk and be conservative to estimate with the probability what's the potential revenue in those years. So that's how we created what we call funnel. So the real business that we are bidding on is a lot bigger than the funnel we show because of the factor that we put in.

And my understanding is you're thinking of this kind of at an OEM level. So when you have a Tier 1 win, that you're not -- it's the revenue that they pull through to the OEM.

If there's a multiple Tier 1 bidding on the same project company-wise. That's basically what I'm saying.

Okay. Okay. And then can you talk to that process then, the probability part of it? As you're competing -- I mean you compete with, we know the names, Qualcomm, Mobileye, NVIDIA, these are all big companies, pretty established. Can you talk about how challenging it is to be a smaller company competing with them? And I mean I feel like it's clear that your technology is unique and innovative and brings power consumption advantages that those guys can't hit, but you're still competing with those companies, and in particular in the current market where supply chain continuity is such a big focus.

Right. First of all, I think everybody knows scale matters in semiconductor. And so the only reason -- the only way you can compete with the much bigger competitor is you have to have much better technology. Fortunately, with the CV3 real silicon, real hardware, we prove that we have a better technology. So even with that, we still need to overcome the perception that because you're small, you are more risky. And also you're more small, you are more -- you are less known to the top management. So that's why working with a company like Conti really bring a lot of benefit to us, quickly give us credibility in front of OEMs. That definitely helped us tremendously after our press release. And so being a small company, we try to work with the customers who really appreciate our technology and focus on them to introduce our product. Our strategy is if we can show off with a small number of products quickly into the market that will prove the concept that we can produce a better product, that would definitely help. That's one thing. The other thing you talked about, supply chain. I think with TSMC sensor building so many fabs around the world, I think in the predictable future, I don't think we'll see another shortage of semiconductor supply anymore.

Yes. I'm inclined to agree with you there. So my last automotive question, if we could talk to China. I think there's a lot of focus in the U.S. around U.S. OEMs and European luxury cars. But China seems like there's a more pragmatic approach, which is that when the stuff is safer than a human, it tends to find its way into cars quicker. And you guys have a lot of partnerships in China and you have a deep base of people in China. Can you talk about that opportunity?

Yes. First of all, in U.S., in Europe, the design cycle for any car usually is more than 4 years. In China, it's cut to half. It's not that because it's easier. It's just that I think the Chinese culture is just they are aggressively trying to put the product to test in a real environment faster than other vision. There's no pros and cons on that in terms of product development. But -- however, just because Chinese companies are more aggressive, we really want to work with some of them to prove the concept. We talk about I want to make sure those people can -- we can demo real product to prove the point, which maybe we can do it faster in China, that's why it's important for us. Not to mention the automotive market in China is still 20% worldwide market. I think that's another reason. Although there's a lot of conflict between other geopolitical level that China and the U.S. addressing, but I don't think that impact comes to the automotive market yet, right? So in fact, we were in the security camera, our Hikvision Dahua business was 30% of revenue 3 years ago; today, 0. That's how much impact to us with the geopolitical situation. But I truly believe that impact is not the area for the auto. Hopefully, it will never be there.

Great. Maybe shifting to professional surveillance, and I'll ask one more question then I'll open up to the audience in case there's questions there. But professional surveillance is still your biggest market and still pretty good growth story. I think a lot -- you guys have a dominant position in video processing. And I think when you talk to your customers that are moving intelligence into the camera, they're enthused that it can grow the market a lot. It certainly grows your competitive position and content a lot. And so can you just talk about where we are in that transition, and how much growth do you think is still ahead of us?

I think we are at the very early -- at the beginning of the phase of the transition. If you look at -- we talk about 40% -- 45% of total revenue will be CV this year. But our ASP is twice higher, so you can say that 20% of our units is CV today. That's just a very small chunk of the total size of the market on the IoT side. So it will take probably another few years for the CV to become the dominant portion of security camera, but I do believe the momentum is there and we're getting faster and faster. So with that, the switching from video processing only to the computer vision, it helps us 2 ways. First of all, the replacement cycle; two is really ASP is twice higher. So that's why we're so excited. And also because we have a great position in this computer vision solution and we do believe that we have better technology than our competitors in this space.

Great. So let's pause and see if there's questions from the audience? Can you wait for the mic?

Thanks for the update, Fermi. I really appreciate it. I mean in recent news concerning Hikvision, right, and the issues with the security industry in the U.S., I mean, as you know, most of the cameras that we're using in Europe and in the U.S.A. are basically made by Hikvision. Apparently, they're about to get banned. So how does that affect you? What opportunity do you see there? Can you run that, please?

Yes. So first of all, I think our Hikvision business is down to 0, but that doesn't mean we lost the market because a lot of the U.S. and the Europe business that's covered by Hikvision will be replaced by other vendors like Bosch, Axis or Motorola who are really using our solution in a major way, right? So I think from that point of view, we probably can pick up some market share that Hikvision lost in Europe and U.S. But however, at the same time, we lost the whole China market. So that's the pros and cons of this.

Other questions from the audience? Maybe if we could talk to the consumer side of surveillance, you mentioned Ring, some of your design wins there. You mentioned the price sensitivity. But the utility of having intelligence in these cameras for doorbells and things like that or even home surveillance where, right now, I get a lot of alerts that are sort of not useful and the ability to do target alerts, it seems like a very valuable function. Where are you in terms of bringing that to market?

Right. So in fact, that consumer security camera really become very obvious to me that it's going into 2 different segments. One segment is cost sensitive. You see a $29 Amazon camera readily available. And there's another one, Ring is selling a $249 computer vision-enabled camera. So these 2 tracks, I think that the low end side is really attractive to people who just want to put camera and protect themselves. But like you said, there's a different customer, they understand the benefit of having a real-time alarm and with accurate alarm with embedded computer vision in the camera. So these 2 markets will coexist, and it would take a little time to develop a system because right now we see that a lot of volume go to this $29 business. But at the same time, I think a lot of the customers are developing this computer vision. So when these 2 markets continue to develop, we would like to see how these 2 market share moves between these 2 markets. We don't have data on that. But definitely, that's the current situation. Ring and Vivint already announced product, there will be a few more coming out with such solutions.

Great. And then the final kind of catch-all of categories when you think about applicability of computer vision to robotics, to store automation, access control, things like that, how do you make sure that you are addressing those opportunities given the focus that you have on surveillance and automotive. For a small company, it's already a big investment. How do you make sure you're addressing that broader range of opportunities?

For any new computer vision application that we want to focus on, the first condition that we decide to do is they can use our existing silicon. If we need to develop brand-new silicon for that application, we're probably going to walk away. But if I look at robotics, particularly for example, the mobile robots that people are talking about, it's just -- it's autonomous driving car moving at a slower pace. That's it. They still need perception. They still need sensor modality. They need all the function for Level 2, Level 3 car. So a CV3 chip with different software can enable that. So for us, building silicon is always taking longest lead time and very difficult to do. But as soon as we have silicon, if we can use a silicon to address new applications with software, that is a new opportunity. So all the markets you mentioned is we can use our existing chip, CV2 family or CV3 family, to address. And we are really putting business development team and the software team to address those opportunities. Some markets will take a longer time. Robots is very segmented. It will take a longer time to develop. But it's my belief it's going to be -- in 10 years, it's going to be a huge opportunity for us.

Great. And maybe we'll close with a question for Brian. If you could talk a little bit to the expense controls, how do you balance the need to invest in all of these opportunities the fact that your revenue growth has been kind of at least not what I thought it would be a couple of quarters ago. How are you balancing those objectives going forward?

Right. So we've talked about the very large opportunities that we're pursuing and the traction that we have in capturing those opportunities. So there will be some additional spending along the path. And as we move into next year, we would expect to see some R&D increases just because CV3, for example, is a family of products, it's not a single chip. And next year, we'll be taping out additional automotive-grade 5-nanometer chip. So we'll see some continued uplift in OpEx. But beyond that, we're being very diligent in managing our spending. Obviously, it's a difficult environment. Revenue is depressed. But we don't want to delay product introductions. It's the lifeblood of the company, and the future opportunity depends on keeping the timing of those product introductions.

Great. That sounds good. Well, thank you very much for your time, and have a great day.

Thank you.

Thank you.