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

Good morning, everyone. Thank you for those who have joined our earlier fireside chat sessions, and welcome to those who are just tuning in now. My name is Aileen Smith, and I work alongside John Murphy on BofA's U.S. autos team. As a reminder, through the course of this discussion, we're going to be receiving audience questions, which you can submit through the Veracast system on the right-hand side of your webcast interface. On the right-hand side, you can also access materials, which will include a slide deck that Ambarella will reference in the prepared remarks. You can e-mail myself, [email protected] or ping me on Bloomberg if you'd like to submit questions that way and I'll try and stay on top of everything. We'll be reading these question anonymously and posing them alongside a number that we've compiled already. If you experience any problems with the webcast at any point, please e-mail [email protected], and they'll get back to you in very short order with assistance. We're now very happy to have join us Ambarella, which is covered by our U.S. semiconductors team and analysts in the BAIC area. But we think it's increasingly important to watch from the automotive side of things as this technology is very relevant in the march towards vehicle autonomy. Specifically, Ambarella is a fabulous developer of low-power, high-definition and ultra-HD video compression, image processing and computer vision chips. The company combines its processor design with expertise in video and image processing, computer vision algorithms and software to provide a platform that is scalable across a multitude of applications like security, sports, wearables, drones and very importantly for us, automotive cameras. From Ambarella, we are very pleased to welcome Casey Eichler, the company's Chief Financial Officer, a position that he's held since 2018. Prior to joining Ambarella, he's held a variety of CFO positions, including at Ultra Clean Holdings, Credence Systems, MarketTools and MIPS Technologies. We also have Louis Gerhardy, who is the Senior Director, Corporate Development and Investor Relations. First, thank you, Casey and Louis for joining us today. I'll now pass it over to you guys for some opening remarks and an overview.

All right. I'm going to make -- I appreciate that, and welcome to everyone. I'm going to make a couple of quick remarks. It seems like my technology is a little spotty. So if I jump in and out, Louis, you'll need to jump in for me. But I do want to just quickly, of course, reference everyone to Slide 2, which is our forward-looking statements. I won't read that, but I'd appreciate it if you would take that into your factor and into our discussion today. The company, I think, I'm not going to cover the detail. It was founded in 2004, went public in 2012, initially doing video processing. And that's really the heritage of the company. What I'm going to add in the last few years, we've come out with computer vision products. And that's really where we're going to today because that's where the majority of the discussion around automotive is centered today. The company has been highly profitable, has a strong cash position. And I'm happy to answer any of the financial questions, but I think people are more interested in getting into some of the more detail. About 800 employees, mostly in Asia, mostly engineers today and very, very technology-focused, which is where I'd like to go. Louis... [Technical Difficulty]

I'm not sure if the -- did Casey's video come out?

[Audio Gap] of where we are today with our CV technology, in particular, as it applies to automotive.

Yes. So Casey, you blanked out there, but I think you passed it to me. And Aileen, I'll make a few comments. I'll try to keep it short so we can get right into questions. But fundamentally, at the highest level Ambarella is in the transition from human viewing to computer vision. What that means is that our opportunity is now not just serving -- capturing an image for human eyes to digest, but also we're now, the incremental opportunity is helping electronic systems perceive their environment so that they can make decisions either partially or fully autonomously. And this is playing out for us in multiple IoT markets, automotive is one of the largest, but we're initially realizing revenue from the security camera market and a number of additional non-auto IoT markets are in development for us now. One thing that's very unique about how we're doing this is that we're providing this visual AI horsepower for the IoT devices. A lot of the AI that you've heard about before it's been executed in servers using CPUs or GPUs or FPGAs that either the network edge or the cloud and data centers what's unique about our approach to this market is that we're doing deep learning AI in the endpoint device in the IoT endpoint device itself. And to do so, you need a very different architecture. How are we differentiated? Well, we're differentiated, one is we have a critical mass of the scarce video processing engineers. And there's a short supply of them. We've got critical mass. We have reputation. Our team is growing. The second differentiating characteristic is by our approach to designing a computer vision chip. We take a thoughtful approach, which we call algorithm first, which means before we start laying out transistors for the tip, we understand the software stack, the computer vision software stack we're going to implement and then we design the silicon architecture. And that's in contrast to the way many of our competitors have entered this market. In terms of products, you can see on Slide 7. Our portfolio of computer vision chips, most recently announced products are both ends of the spectrum, CV28 for very high volume, CV5 very high performance, multiple cameras supported for chip. And in terms of markets that we're serving, we've spoken about 3 waves of visual AI revenue. The first wave would occur in the enterprise-class and public infrastructure, cameras, and we began to see material revenue in calendar year '20 from those markets. The second wave we said we expected to see material revenue in calendar year '21 from the home security camera market, and we're still confident that's going to play out this year. And then the third wave we've been talking about has been in the automotive market. We said previously we expected CV revenue in auto to become material in calendar year '22 or calendar year '23. However, we're already providing significant evidence that, that third wave is well underway. For example, by the end of April, we said in our last earnings call that we expect to have shipped more than 300,000 computer vision SoC units by the end of April, with most of those units having shipped in the last year. Finally, we did announce in automotive, and I'll pass it off to you, Aileen, after this. On Slide 16, we have an automotive funnel that's beginning to take shape. About 2/3 of that funnel, which net figure is $600 million over the next 6 years, about 2/3 of that has been won already. And the other 1/3, we're in the bidding process and discounted it for the probability of winning those programs. And that automotive business is coming across all levels of autonomy, ranging from L0 all the way to L5 in an example of a recent L5 win would be an Motional announcement we made just a few weeks ago. So Aileen I'll pause there and we can go into questions.

Absolutely. It's a ton of good content to start with and brings up a lot of questions that we can certainly get into. I think a question to kick off with, just given the audience that we have today, for those of us in the automotive universe that cover the automakers, the Tier 1 supply and the Tier 1 suppliers. The Tier 2-plus supply base, frankly, can be pretty confusing in terms of where the technology is coming from, that is ultimately going into Tier 1-plus component systems and service integrators. Obviously, though, semis and chips are gaining in relevance from a content and from a value-add perspective. Most of us are pretty familiar with Mobileye on the vision front. But could you remind us who you identify as your key competitors and you touched upon this a little bit in your prepared remarks, but I want to drill down a little bit more. Really what differentiates Ambarella's approach versus some of the other key players in the market?

Yes. So first of all, let me define a Tier 2 from our perspective as being both hardware and software. So you can be a Tier 2 supplier in supply chips or you can be a Tier 2 supplier and provide software that runs on chips like ours. Our focus is on providing the computer vision processing chips, and we work with a network of Tier 1s or OEMs or third parties, sometimes a combination of those 2. We're doing the software that runs on our chip. So you can see our focus is clearly on providing the best-in-class hardware resources for our customers or partner software to run upon. And we're differentiated in that we've got this tremendous know how we've built up over the years in terms of treating video as a unique type of data. It's something this company has done since day 1, where video is considered data. And if you can't feed your AI processor with high-quality data, if you can't see everything or if it's blurry or if difficult lighting conditions to store the data going in into the AI processor. It doesn't matter how good your AI processor is, it's garbage in and garbage out. And so we're differentiated in the sense we've got this reputation and technology for that video processor front end. And then we've got this optimized deep learning AI processes that sits behind it. Everything architected to work together to process this data and make decisions so that our customers and third-party software can run very efficiently. And if I contrast that, maybe I'll just -- you had mentioned Mobileye. Mobileye is taking a very different approach that hasn't been commonly used in the industry, where they're doing the software stack and the chip, whereas Ambarella is just focused on providing that best-in-class hardware resource and then customers or third parties can develop software to run on our chip.

Well, I think let's follow-on to that. When we think about the hardware versus software, where you guys are focusing more specifically on the hardware side, Mobileye as a key competitor is really on the software side. Can you talk about how much software support you are receiving for your systems on chip versus perhaps some of the other competition outside of Mobileye?

Sure. Just I'll reference on Slide 4. Of our 785 employees, 81% are engineers. And almost 70% of the engineers or more than half of the company are focused on software and algorithms. And so how does that contrast with what I just said that we're focused on the hardware resources and not the application layer. Well, think about a software stack, as the application layer being on the top of it. And then a lower level, mid-level software enabling blocks that we can provide to accelerate our customers and partners time to market as they develop the application layer. So that's actually where we draw the line is that we have a lot of micro code embedded in our chips, and that we provide SDKs with lower level software blocks that our customers and partners can use to develop their own applications. And furthermore, and this is very important because we have some competitive advantages. We provide the software tools that allow our customers to port their software, their training software they've developed into our inference chip, and then to optimize it doing some unique things like sparsification that other companies have a more difficult time doing, and we can do that given our more thoughtful approach to the market and the algorithm-first strategy that we articulated. In terms of our overall software footprint, in terms of Tier 1s and OEMs and third parties. We have several dozen that have developed a wide variety of applications just for the automotive market. Sometimes, they're focused on just doing the software for an eMirror. Sometimes there's companies that are just focused on having driver monitoring or occupant monitoring software. And then, of course, there's the front ADAS software companies. And eventually, the market moves to higher levels of autonomy like L2+ and L4 and L5 like Motional, as an example. In that case, we work with the OEMS, who we find are focused on developing their own software for those higher level of autonomy applications.

Okay. So you guys describe ourselves as, I think, an artificial intelligence vision silicon company. So when you're using that description and thinking about the value-add that you're providing to your customers on the AI front. It's really these software tools that are providing to the customers this algorithm-first approach that allows it to be essentially modularized at the customer level for the various software applications that they're working on, correct?

Yes. Our software is both in the tools as well as lower level enabling blocks that customers may choose to use to create the final application layer.

Okay. When we think about -- one of the questions that we get a decent amount of people is, why is computer vision an ideal solution for autonomous driving systems, whether it's Level 2+ active safety or Level 4+ fully autonomous capabilities. What specifically can computer vision offer that cannot be solved for by the existing sensor hardware or even software suite?

Yes. Well, first of all, cameras is the technology we all have in our phones, 2 or 3 cameras by now. And so at the highest level, it's a technology that has matured, the cost has been driven out and there's just a lot more advances that we can make. And so if you look at, for example, Slides 9 and 11, we'll refer to this in our deck. The base level of perception, whether you're talking ADAS or higher levels of autonomy, the base level perception, and when I say base, think of -- define it as portion of the point cloud data that's being generated by cameras. The vast majority of it, we think more than 90% is being generated by cameras and then that information is being augmented with data from adjacent sensors. And so there isn't another sensor technology that's capable of doing such a large amount of the data in the software stack at such economical price points. And so that's why cameras, we believe, are the baseline level of perception upon which you can provide incremental information with other sensors that have different characteristics to operate in challenging situations.

Great. That brings up, I think, a really important question from our side of things. Obviously, we've seen this in terms of the emergence of some of the newer specs onto the market. But [indiscernible] is frequently cited as a gating technology for autonomous capability, while camera and radar are perceived as perhaps more mature systems. However, obviously, you're clearly advancing computer vision technology and companies like Magna and others are pursuing more advanced radar as well. So when you think about the incremental progress that can be made in terms of the value-add for cameras, do you think that traditional component systems like vision have much more room to go via your technology or perhaps others?

Yes. So I mean, there's a lot of headroom for cameras and how they're used today. For example, if you look at the front ADAS market, the sweet spot in terms of camera resolution is 2-megapixel sensors, which means they can see so far. But if you use an 8-megapixel sensor, which our new reference platform that we call, Rebel, R-e-b-e-l, uses, you can see farther, you can see more accurately. Some companies might want to use stereo, which will give you more depth information than just a monocular camera. And then there's -- you can use more cameras or you can do fusion of, say, cameras and radars and create optimization with some of these other sensors that didn't exist before. So we think it's true that the cameras are relatively mature technology, but that, by no means, means that technology development is slowing down and very strong outlook for the use of cameras as we move to higher levels of autonomy.

And as a follow-up to that question, maybe a good segue into the competitive landscape here. Mobileye recently unveiled its LiDAR SoC. So do you guys perceive this as an effort that may be indicative at all of a broader change in the competitive landscape and where your peers are directing investment?

So we're agnostic to what other sensors are used in these vehicles, we'll listen to our customers and Tier 1s and what they want. And we're focused today on the camera perception layer. And in the future, the next step for us will be -- to be the domain controller, where, in addition to doing the camera processing, we'll take the output from other sensors. And again, we don't care what is the mix and combination of them, some LiDAR, no LiDAR, more LiDAR, less LiDAR. It doesn't really matter that much to us as long as we're in there with the cameras and then also positioned to do the domain controller in the fusion level of processing. So to us, that doesn't really change the competitive landscape. And again, we're -- we'll support whatever sensor suite our customers tell us to support.

And you mentioned that sensor fusion and the domain controller system, what is the customer receptivity to outsource those systems As we think about autonomous systems in general, whether it's Level 2 or Level 4-plus, they're highly mission-critical systems in getting the car from point A to point B safely. So when you think about customers and their likelihood of outsourcing that system itself or the software to the Tier 2-plus supply base and getting a solution that may be more or less a black box. Is there -- like I said, is there receptivity for them to do that?

So for the semiconductor component, most of the chips are provided to the automotive industry on a merchant basis, meaning outsourced. There are a few companies that are doing their own chip. The question really becomes one of scale because in the semiconductor industry, the cost in all-new chips. Moore's law becomes higher and higher, so you need more and more volumes to justify that expense. And there's a few companies that are still trying to do their own chip. Some are doing it so successfully so far. But we believe in the merchant market where the semiconductor design and the product investment is leveraged across a broad customer base, and that's the way that we can offer investors the best ROI, is by spreading our R&D investment across a much broader base of customers rather than just selling 1 ship to 1 company on a kind of a customer arrangement. So we believe this outsourcing of chips will continue. And furthermore, I think part of your question was about having separate software and separate hardware resources. And we believe that for ADAS, Tier 1s and third parties will develop that software stack to run on chips like ours. But for higher levels of autonomy, we see the OEMs much more involved in that software stack. And we spend time with OEMs as well as Tier 1s for our automotive business.

Great. I'd love to switch gears a bit and kind of talk around the commercialization aspect. So you have a host of compute vision chips for automotive either in the market today or in the pipeline. So CV5, CV25, CV2, CV22 and a bunch of others. But when we think about what you are currently in the market on, obviously, it's very much more so recorders and data loggers. How do you transition your existing automotive business from -- for your chips from what our recorders and data loggers right now to that Level 1, 2-plus, Forward ADAS and then ultimately, Level 2 to Level 4-plus systems in autonomy.

So for reporters and data loggers, you're correct, most of our automotive revenue last year, which was in the $35 million to $40 million range was in that category. In that area, we expect to be -- continue to be a growth market, in particular, on the pre-install side, not as much on the aftermarket. This is the year, and we've guided our automotive revenue to post the strongest growth at our company in calendar year '21. And that's because many of these new markets, you can see listed on Slide 10 in our handout, such as Front ADAS, eMirrors, In-cabin, even some L2-plus and L5 business will begin in earnest this year, and that's what's going to cause our premium revenue growth in the automotive market. And so we've got these products out. And we felt that we hit an inflection point in first half of last year, and we articulated on November 23 last year and automotive revenue funnel. This is represented on Slide 16 of the handout. And this automotive revenue funnel, which I mentioned earlier, is a net figure of $600 million is coming from multiple products, whether it's CV25, CV22, CV2, CV2FS across all levels of autonomy, whether it's an L0 type, say that L0 could be an ADAS product for a commercial fleet that gives you a warning, but it's not integrated into the vehicle control system L1 or L2. But we even have some L2-plus and L5 projects in there that will generate revenue for us this year and are a solid -- making a solid contribution to the $600 million funnel. So how do we make this happen. Well, we continue to introduce a portfolio of products that will increase our serviceable market today in a car or serving the perception layer processing for the cameras. The next phase for us will be to introduce products that can do the fusion layer as well as the perception layer, and then we continue to build upon that. So we would anticipate that our SAM continues to give us robust headroom for revenue growth.

I think it's also important to note that on Slide 7, all of those products, you can move from 1 product to the other fairly seamlessly because they're all based on the same architecture, the same design. So you can pick and choose depending on what your need is across the portfolio of products that you see on Page 7. Yes. On the Slide 7, as well as Slide 10. For each of these discrete camera systems that we articulate on Slide 10. For each of them, we have multiple software partners, whether those partners are Tier 1 or an OEM or a third party, developing software based on optimized to run on our chips. And we can provide the customers with flexibility as to different combinations of these discrete camera functions. After choosing rather than being forced to have 1 box for 1 application, you could have a product that does say, front-facing ADAS, a data logger or front-facing ADAS and In-Cabin camera system.

You bring up a ton of good points there. I certainly want to get into the revenue funnel and the total addressable market estimation. But I do kind of want to close the loop as we think about your existing product set transitioning over time, your customer set should also transition with it as well. So right now, some of your key customers on the automotive side are at the retail and the Tier 1 level. As you are targeting, advancing more towards Level 2-plus Level 4-plus -- excuse me, Level 4-plus products, specifically, how does that validation process work as you transition from retail to Tier 1 to then automaker customers? And what role do your Tier 1 partners, a lot of which you just mentioned you're partnering with on the software side. What role do they play in kind of helping that go-to-market process or getting your foot in the door at the automated level?

So yes, our revenue in auto today is a mix of aftermarket and preinstalled. And the outlook is that the pre-install market has got the, by far, the most significant growth prospects and opportunity ahead of it. We're well-established today with a number of Tier 1s, which you can see on Slide 15 in our automotive footprint for that human vision pre-installed market today, companies like Denso Ten has sold to Toyota or someone like JVC, KENWOOD has sold to Honda or Nissan. And so Ambarella name is quite well understood for -- and has a reputation and reputation for the data logger drive recorder market and so you can imagine what we need to do with those companies is to convince them of the powerness of our computer vision products. It's not so much a story of convincing them about Ambarella's ability to deliver. Of course, then there's other Tier 1s and OEMs that we have not had relationships with before. And I put Motional as an example, in that bucket. And those are companies where we need to convince them of our technology and products and strength of the company is there so that they will design us in and so that's -- so we're well in the process of making that happen. And we're just constrained somewhat about the ability to announce these relationships as things are -- tend to be very confidential in the automotive industry. But again, going back to the funnel on Slide 16. We're engaged with many Tier 1s and OEMS, and a majority of this business represented by this net $600 million in the funnel is coming from these Tier 1s and OEMs in the automotive industry. In terms of the softwares, we're all in helping us win. Well, without good software stack, it's hard to sell just the chip only. And we're constantly on the lookout for state-of-the-art software stacks, and we're constantly adding new software partners. So we don't see that as a bottleneck for our business. And when you do find a major software partner or an established software partner. What ends up happening, for example, with the Rebel reference platform is you have both companies marketing the joint solution, their software running on our chip to Tier 1s and so that can help the snowball move a little bit faster.

I think it also can be an advantage to have that flexibility because a lot of OEM in the future are not going to want to get boxed into a single solution that they don't have flexibility. And what we provide with a platform and our partners is the flexibility to combine different functionality across several of the different markets, and I think that can be an advantage.

Absolutely. I do want to touch upon this or maybe expand upon it a little bit because it's been referenced several times today in our conversation, which is, obviously, a very important development on your end which is Motional. And for those a reminder of everyone on the conference call, that's the joint venture between Hyundai and Aptiv. We'll be using your CVflow vision processors. So as we think about what Motional is right now, which is very much a, I would say, R&D-focused investment to advance autonomous technology. Is it fair to assume that, that win on the CVflow vision processors is still a R&D project award, while Motional is still in testing of its autonomous vehicles. And then how do you convert that project award to eventually a production contract to go back to your point of there's still a proof that you need to have a common customer to really get in the door on a production contract.

Yes. Slide 14 of our deck, and give some information on Motional, and whether this is commercial or an R&D project, it's best to ask them, but I point out on December 16. Motional announce that Lyft in Motional will deploy fully self-driving vehicles in multiple U.S. cities beginning in 2023. And we have indicated that we'll see some commercial revenue from that -- this relationship and design win this year. But wouldn't expect it to be a significant number. Just to make it clear what we're doing and what we're not doing in this vehicle is we're providing -- our chips are part of the central processing module, and we're doing the video processing, the AI processing, the safety online for the exterior cameras in this vehicle and again, this is at the perception level. There's another layer of processing in this vehicle called the domain controller infusion level that we're not doing, hopefully, in the next-gen car, that's some business that we can win with a product that will come out within the future. And then there's central processing in this vehicle as well. Our role in this car is to do the video processing, AI processing, video encoding, safety [ island ] for the cameras on the exterior of this vehicle.

That's very helpful. In the 5 minutes or so that we have left, I do want to touch upon, that $600 million funnel for automotive revenue because it is important to get into. So of that $600 million, you disclosed that your key customer set in there is a lot of the Tier 1 suppliers. So when we kind of bracket the percentage of that $600 million. I mean, one, you said 2/3 of it is booked already, another 1/3 is based on probability of booking it. When you think about it from a product perspective, whether it's data loggers, Cabin monitoring, eMirrors and then more so towards Forward ADAS, Level 2-plus, Level 4-plus. As you're increasingly booking those contracts with Tier 1 suppliers is it fair to assume that, that funnel is increasingly being comprised of the forward ADAS, the Level 2-plus and the Level 4-plus solutions?

Yes, that's correct. The $600 million is net. A majority of it is computer vision. And well, our strength in recorders and data loggers is where we're coming from. More than half of the funnel, easily more than half is coming from all these other markets combined including L4, L5, including some L2+. But yes, in the next 5 years, some of these other markets, in particular, Front ADAS, then followed by eMirrors and In-Cabin type camera systems, those will become more important. But it's pretty well balanced across the applications. And by the way, it's also balanced geographically with Japan, North America, Korea and China, roughly in that order, Tier 1s in those regions and OEMs contributing to that net $600 million figure.

Certainly, the revenue from L2-plus and L4, L5 is further out. Some of the revenue that's more current is, obviously, from some of the ADAS wins that we've had to date. And so -- but there is a balance across that number. But certainly, the revenue that we're seeing today is more related towards that first category.

That's some good color. I think a final question to wrap up with you guys. I'd love to touch upon the SAM or the total addressable market for your products. I think you've outlined that through 2025, perhaps, some numbers that have been put out there, $400 million to $500 million in recorders and data loggers, over $2 billion in Forward ADAS, eMirrors and In-Cabin monitoring combined. Over $2 billion for Level 2-plus and a smaller $200 million level for Level 4-plus. Do you guys internally or could you provide any color on an estimate that you guys have for percentage market share that you think Ambarella could occupy in some of these verticals over time as we think about what that SAM is versus what's really applicable to say, Ambarella?

Yes. So today, our market share in our SAM, I'd say today, Fiscal Year '21 as an example, it's in the 1% to 2% range. And yes, we don't provide targets, as you can imagine, and it can be lumpy as you win 1 program to another. But the only reference point I can give you is in IoT security, a market that we've been in for some time. And in that IoT security market, our market share varies based on the tier of cameras for example, low end versus high end. Our technology is very well positioned at the high end of markets and as the markets are evolving. And in a market, say, like IoT security that's fairly mature but still growing at a very nice pace, and particularly with the new product cycle, CV. Our market share of our SAM there is in the 20% range, and we think it's probably going up a little bit. So hopefully, that will tend to book end things for you and investors can make their own assumptions about market share.

Absolutely. It's a good benchmark to think about. I think with that, we are at time. We very much appreciate the commentary and the information today. It's been very informative. From my end, we appreciate your continued support of our summit and I know you guys participated with us in the past. So I'd just like to thank Casey and Louis again for participating. And we thank everybody for tuning in.

Thank you.

Thanks, guys.