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

Hi. Good morning. Welcome back. I'm Joe Moore, Morgan Stanley semiconductor team. Very happy to have with us today Feng-Ming Wang, the CEO of Ambarella. Always one of the more fun conversations, whether the stock is 30 or 200, there's always so much good stuff to talk about with you guys.

So maybe if you could just start out and talk through a little bit the transition that you guys have gone through. It's getting now to be the meat of your business, 45% of your business, computer vision, this year. But can you just talk to the transition that you've been going through and the focus -- the strategic focus you guys have?

Right. So when we started this company, in fact, our Series A investors here 18 years ago, we were focusing on digital video. And throughout years, on the first 10 years, we did just video cameras for capturing and storing video. But at IPO time, almost 10 years now, we focus -- we started to realize that we need to focus on analyzing video real time, which we call computer vision or AI for video. But that technology that we spent the last 6, 7 years to develop invest more than $600 million total on that technology. And the idea is really any camera using our chip in there, not only can capture video, but analyze a video to apply smart functions or analyze the environment to perform higher level functions. For example, for the autonomous driving, for security camera, they can detect objects like a license plate or colors of car or the people passing your house. So all of that is enabled by our new chip. And like you said, we start seeing a material revenue like 2 years ago, it was 10% and become 25% total revenue last year. This year, we are forecast to be 45% of total revenue coming from our computer vision chip. And we are penetrating now in just our existing markets, like security camera. We think we're going to penetrate the automotive market. And also in the future, we think this similar technology will be used in a new market like any robotic type of locations.

Can you talk about -- I guess, one of the things that's made computer vision really exciting to me was the breakthrough of machine learning and deep learning in terms of solving some of these vision problems. And that happened really right around the time you guys had first silicon available. And I know you have a software approach that's very tailored to that. Can you talk about your software force approach to this and sort of why this is a good development for you guys?

Right. So for us, it's -- although we are a semiconductor company, but we are trying to solve a real problem. At beginning, a real problem is how to build the best video quad camera. So that means you have to provide better algorithm or software on the video processing and compression. And then for AI, you have to provide the best highest possible performance of computer vision and lowest power consumption because that's which is important for the any H device. Our approach is always build algorithm software first. And that after we reach to the performance level, we think is good, then we'll start optimize. And optimize target to get smaller die size and smaller power consumption without impacting the performance. And this unique approach gives us huge advantage in terms of power consumption, particularly power consumption at sometimes even die size savings. So that has really become our differentiation. Today, if you look at just H device like camera, security camera or any other camera device, power consumption is important. But even for automotive location, that driving a car with a 200-watt chip, it's just not a good value proposition. And if we can deliver a similar performance at the 40 watts, that help our automotive customer not only saving the battery life, but also much easier to deal with [indiscernible] problems in the system level.

Great. Maybe we can talk to a couple of the recent strategic developments, starting with CV3, which was a chip that you announced at CES. Can you talk about what that does to sort of widen out your scope of your ambitions in this space?

Right. Like I said, until CV3 announcement, Ambarella is focused on digital video than go through a video perception. But we know that for the most of our allocation moving forward, we need to be the -- we call the domain controller, meaning it's in addition to just a video perception, you need to be able to handle all the sensor modality like radar, like other sensor type. And also on top of that, you need to be able to do a fusion of those sensors, then you do a planning and control of the application. So basically you want to become the main processor of any device you want to go through. So CV3 is really the first chip that would deliver this idea. And basically, the performance level can take in multiple 5, 10, 15 cameras, plus multiple radars into a chip. And we can take the raw data video and the radar at the same time and building a point cloud for both of them, then we do AI on the point cloud, then we do the sensor future from there. So we really think that CV3 give us a change your company from just a video perception company to really a domain control company.

Yes. I mean, you really -- you go from in a Level 4 autonomous car. Today, you're sort of doing video perception offload with maybe an in video or an Intel in the trunk doing the -- you're now going right after the centerpiece of that. I guess what's the time frame for that to start to generate revenue? I assume it's going to take a while.

So we're definitely bidding RFQ on with OEMs right now. Any design win with a major U.S. or European customer will be 25, 26 type of revenues.

Great. And then the other big development was the acquisition of Oculii, which kind of gives you a stronger position in radar and I think give you entry points into maybe some of these other markets. Can you talk a little bit about that and why it's important?

We -- like I said, when we start looking at CV3 and try to become a sensor fusion supplier, then we start looking at what other sensor types we need to support. And it's become very obvious that the radar technology is the best friend of video camera. A lot of people are using LiDAR today, but I think LiDAR and the camera are similar type of technology. They all use optical sensors so that the disadvantage of a camera and the disadvantage of LiDAR are very similar. But radar is different at night at the traffic really bad weathers, radar can perform quite well. So that's when we realize we need to have an internal radar development. But more importantly, I think we understand that today, if you look at all the atom-driving car, their video and the radar sensor fusions happen at optimist level. We believe that's suboptimal. In the future, I really think that the video and the radar future is -- have to be happening at the raw data level. Meaning you're taking 10 video sensor data and 5 radar sensor, you build a coin plus on both of them, then you apply your new network on that. That will give you the best result. And that's why we think it's important for us to acquire because that really provides the best technology solution to our customers.

Yes, it seems like this is something suddenly that everyone's kind of talking about the importance of radar in these applications. And I guess when you look at markets that have here been pretty competitive with forward-facing ADAS single camera. It seems like radar could be an interesting entry point for you guys to start getting traction there?

Yes. One of the reasons that radar has been criticized is that the current generation radar doesn't have the resolution or distance of LiDAR. But however, the current 4D imaging radar that we acquired and also there are some other companies like NXP talking about, the technology always show that the resolution of radar distance can be as good as LiDAR. So with that technology, I think it can be used in combination with -- even with ADAS market, like front-facing, single camera. If you put a radar in there, you can probably even help you to build an even better product in terms of coverage and accuracy.

Great. Maybe we could talk a little bit about your automotive design win pipeline. You guys -- I think your automotive business was something like $70 million last quarter, and you've talked about a funnel of $1.8 billion. So obviously, a pretty big ramp ahead of you in that market. And that funnel I think, tripled in the last year. So fairly impressive growth of the funnel. And I think still a conservative number because there's no radar in there.

No software licensing.

No software licensing. There's some probability weighted stuff that I think you've been pretty conservative on. So can you just talk to how you think about that funnel? And where you are today in terms of securing those design wins?

Right. So one of the problems we face is any design we win, we cannot talk about until the customer in production. For example, we start engaging with Rivian 2018. And with -- the only time we can talk about it 2 quarters ago, when they go into production. And that's where we -- to address that concern, that's where we created a sales funnel. In sales form, there are 2 parts of value. One is the product we really want. The other one is really the pipeline. The product line is very clear that we -- our customer issue a letter to us and kick off project, we know which part model we're going to work on, what's the value and what's the production date. So that can be modeled part quite easily. On the pipeline side, it's really all the IPO we're engaging and also then we start applying a multiplier that reflect the possibility in the design and the possibility of the size we can win based on the volume. So it's heavily reduced based on the assumption. So with this true, we are showing that we think the next 6 years of automotive revenue combined will be $1.8 billion. But the truth is, if anything in the pipeline becomes reality, and the confidence that become 100%, that multiplier can easily change the total value in there. And -- but plus they're going to continue their new RFQ going to bid on in the next few years. So I think that you are right, we try to be conservative, providing guidance. But however, I also want to show people the realistic number that we're thinking and instead of trying to show a very big number compared to others.

Yes, I think it's very helpful that -- the color and detail you guys give us there because I think a lot of companies have funnels and pipelines and they're ill defined. So it's your -- the way you're driving it is very clear. Can you talk to some of the recent design wins? And I guess, I don't want to go specifically into individual customers. But I think when you look at Rivian, the arrival motion, all these different things, I'd like to use a clear trend of people who are starting from scratch, who don't have the incumbency of having worked with somebody else. They're choosing Ambarella a lot in these applications. People who are doing battery-powered vehicles choosing Ambarella a lot. Obviously, you're competing with big companies that have incumbent positions, that have more resources. But it seems really telling to me that guys who are sort of starting with a blank slate are choosing you. Do you think that's a fair representation? And I guess, where does that leave you in terms of penetrating some of the bigger OEMs with the Level 2+ or Level 3, 4 types of wins?

Right. It was a fair assessment considering the design win was at 2018. In 2018, our CV2 just sampled and our -- to a customer, and we are a young company for automotive industry. That's why I think a lot of start-up automotive company wanting work look at us, but because we don't have the reputation in the industry yet at that time, so a lot of the big customers won't consider us because they have other options. However, things are quite different today, right? We have been established not only as a viable automotive supplier, but our CV2 has been tested by real market, by real customer showing up at the product. And more importantly, in revenue, look at we grew revenue -- CV revenue from 25% last year to 45% this year. That shows you how many customers we have. We talk about we have 250 customers evaluate our CV technology, only 100 of them in production today. So the momentum and the reputation we built in the last 3, 4 years really help us to kill people, we have a real technology they can use and our CV3, is 10x better than the CV2 in many different fronts in terms of performance and so on. So I think that reputation we built in the last 4 years, the progress we made in the last 4 years definitely make a huge difference. And I can say that we are in some of the major design we are building on -- we are surely in the final.

And how do you think about -- I mean the competition in this space, when you think about CV3 types of applications, you're dealing with programmable GPU type solutions. You've got like in one high-profile case in Tesla. They have their own custom silicon. It does seem to me that dedicating silicon specifically to this application, particularly in a battery-powered car where power consumption is an issue is going to be better than something that's programmable. I guess do you agree with that? And do you see more companies going the way of Tesla and sort of developing their own customize? So can -- we heard from GM Cruise earlier today that that's their strategy as well.

Right. So first of all, if programmable architecture is always there for the first generation because if you want to do self Level 2+, you want to build a dedicated like CV3, take years to do it. So that first generation that's [indiscernible] because they have a lot of customers know how to program at GPU. But when we look at how much performance improvement you need to get in the next-generation cars. When you talk about 10-plus cameras plus, 5-plus radars, each one of them generate gigabits or more input to your chip and you need to process all of them in real time. This requires application-specific chip to do that, to do in an efficient way. I think that's the biggest difference. I think we are talking about power efficiency 6x better than the programmable architecture. That's because we took the algorithm, we understand where is the pinpoint. We optimize that. That's how we can get to this power efficiency and which is very important for our current customer. So I think that eventually, the -- my 30 years, let me -- combined experience tell me, one thing is on the high-volume market, application-specific solution always win at the end. Programmer architecture doesn't mean FPGA or other things, it's great for the first generation, maybe even for second generation, but when you go to high volume, opportunity specifically give you better price, better performance and lower power. That's the history. So in terms of a system company want to do their own chip. I think 5 years ago, I would say I won't give them a lot of chance because at that time, nobody want to do semiconductor company. Everybody thinks, well, dinosaur, we're going to die, right? But today, how many semiconductor company people are building and how -- VSI people engineers get a huge amount of salary to -- if you want to hire them. So for a system company that want to build a brand-new VSI team that doesn't have IP, it doesn't have a resource. I think it's going to be challenging to size up the team they want and also build up the experienced team to get collect old IP. For example, we need a video processing, radar image processing, CPU, everything, you have to put all the IP together, I think it's going to be challenging for a system company to do it.

They're going to have to have a better chip than you who have just invested $600 million on top of a foundation of 20 years of video processing expertise. So it's going to be challenging. Can you talk about -- CV3 is pretty far out. Can you talk about the growth in the other automotive applications. And we have -- our revenue buildup actually has relatively little progress in even L2 in the next couple of years. But when you think about driver monitoring, where you've secured the wins already surround you, you're already very strong. And then other categories, logistics and package stuff that you've seen with FedEx and UPS as well as fleet management. There's a lot of opportunity there kind of on the way to that bigger pie road.

So I think CV2 silicon will be address those needs, and that's our short-term revenue opportunities. I think I really -- on the automotive side, we are looking at data logger in Japan, which become material in Q4 last year, and also looking at the DMS, OMS plus ADAS. And those 3 things will become one function in the future. I really think that the single camera facing forward, the ADAS market need to integrate DMS function, maybe even OMS function, and we have the solution on that with our CV2 family. So that's definitely an area I think we can penetrate and drive revenue in the next 2, 3 years.

Great. Maybe shifting to surveillance a little bit. You've seen a lot of growth there that's -- you've kind of had Wave 1, 2 and 3, but it seems like Wave 1, 2 and 3 are all hitting sort of at the same time. But wave 1 was professional surveillance, you've seen very strong adoption there, your customers there are quite positive on it. Where are you in terms of that adoption? And then I'll follow up with sort of specific what's happening in China, but the success you've had so far in professional surveillance.

I think it's -- when you talk about security camera, you have to separate China and non-China. For the non-China portion, I think we did quite well. In fact, all the major system houses out there are using our solution, either exclusively or as a big portion of that business. And the reason for that is really our CV family of solution that now we are -- when I say CV is not just one product -- one point of products, we have a whole family a different performance and the price point. And we start at a high level and going down to the lower price points. So we are the only one that I know in that market provide a family of chip that you can invest software once but you can get the whole family product out. So I think that's definitely a strength of us in addition to our performance and the power consumption. So in that market, we're doing quite well.

Great. And then China, your 2 biggest Chinese customers were kind of 25% of revenues a couple of years ago, 6% last year, ended the year at 2%. So obviously, significantly diminished. But that's a big surveillance market, and you were 1 of 2 suppliers. The other one has had their own issues with Huawei. So what do you see happening there? Do you see opportunities still obviously significantly derisked from your perspective? But is there upside opportunity down the road?

Well, because I think the reason we've been going from 25% to 2% is really geopolitical situation. Not only Chinese want to use a Chinese supplier, U.S. put those company on entity list, were squeezed in the between. So unless the geopolitical situation changed, I don't think that we don't -- I don't look at that market as an upside. Do you think that we want to continue to stay neutral on that. And if the situation change, we might have different considerations.

Great. And then consumer surveillance, where you've done very well with people like Amazon, Google, big customers there, probably a little tougher to get computer vision penetration as high because the price points are high, but you've had some initial progress there. Can you speak to that? And it seems like you can add a lot to the utility of these cameras with CV functionality.

Right. So we already announced Ring using our chip for their first generation of computer vision solutions. And we believe there will be a few more coming out. And the benefit to that is very important. Because we noticed that a lot of our customers as soon -- including professional guys, as soon as they turn on the AI at the edge, they start reducing the video being transferred back to the cloud to a different level. Because as soon as you know, what's the important part of the video, you can start reducing the storage costs at the server side. So that's a huge benefit for our customer. I think that same thing will apply to the consumer security camera customer when they really can come over. I think the cost of the camera needs to come down a little bit for them to start looking at the benefit, but the benefit is obvious. When they turn around that, I think eventually when the price is right, the majority of the consumer security camera world go into AI.

Okay. Great. Great. So there's a couple of other growth opportunities that I'll come back to you if there's time, but I guess I did want to touch on some of the short-term issues that you flagged on the last call that tripped up the stock. I guess, first, Samsung Foundry, you sort of mentioned that you're not getting all the wafers that you'd like, $5 million negative impact in the July quarter. Can you talk to that? And I guess any sense for how long that will be a lingering challenge?

I think that the only bad news that we talk about this quarter versus the previous quarter, we talked about is a sensor problem, right? So there are a lot of speculation of why this happened and trying to speculate, there are other problems. I want to be clear that the only problem we're talking about in this earnings call is this Samsung Foundry, they're going to show supply us about $5 million revenue in July quarter. And the reason happens, they haven't told us reason, but we noticed that they didn't load the wafer after Chinese New Year, and we asked the question then they come to -- they become clear that they're going to supply to the PO that we could give to them. What they told us is they think Q2 is the worst. I think Q3 continue -- the problem will persist but will be in a lower level. Hopefully, Q4 will become much better than that. So that's a trend, I think we're telling our customers and investors at this point.

Okay. Great. And then yes, the other thing that I agree, it wasn't a negative commentary on anything numbers-wise. But you did comment on the overall supply chain, challenges that people are seeing with really other people's components. And to me, you mentioned one positive, one negative. There may be people deferring purchasing from Ambarella waiting for those components. So there's some upside demand there. There might be other people that are building inventory, their stuff they can't ship out because they can't get those components, which is a negative you've actually said that for several quarters. People noticed it this last quarter because a, the market is much more negative now and b, there are other contextual things around the quarter. But you've framed it this way for a while. But -- and I think you made the comment like people ask is there a problem and you say, well, not yet. And that's always for us, a little anxiety inducing to hear a CEO say not yet. But it does seem like this has been a consistent language. And I guess, in the context of computer vision going from my numbers, something like $80 million last year to about $180 million this year, it seems like $5 million of Samsung Foundry issue and maybe some of these inventory issues don't seem like they should be the defining factor for the stock, but just I'd be curious to see your comments on that.

I think market overreact, but I'm not in a position to comment on that. I don't want to comment on our stock because it's not at that level.

That was I think...

So however, I really think that if you look at our long-term trajectory didn't change at all. In fact, that we announced CV3, which is a $10 billion transistor chip and building a 5-nanometer, which is, I think, is a great achievement for a company our size. And we deliver that chip on time, and we got the sample on time and hopefully that we can get a design win there. I think that's trajectory people should focus on, and find a short-term delivery problem is unfortunate. We need to put a handle -- our handle around it, so that to minimize our impact to our customer. But it didn't change our long-term plan at all.

Great. Let me stop and see if there are questions from the audience. Great. Then maybe if we could go back to the growth theme. Could you talk to some of the other markets that you guys have talked about, whether it's -- I guess what's exciting you the most between robotics, sort of access points, some of the different applications that you guys have started to talk about. Drones?

Well, I have -- drones.

Industrial drones. Yes.

Industrial drones. As technology people I really truly fundamentally believe that robotics will become a major, major application in the future. I don't know it's 5 years, almost 10 years. But I think that any robots out there will require a platform like we built for CV3. All derivatives with CV3 depends on the performance level you want. And we want to make sure we don't miss the opportunity when that happens. So we have a dedicated business development team on that market, trying to talk to as many customers as possible and try to identify customers that we need to work with and try to win design wins with them. That process is ongoing for us. And I continue to monitor it myself because I think although automotive is greatly important. We need to win there. But I think if you want to build a long-term company, robotic will be there for us if we continue to work on it.

And it's really kind of related because I mean, I feel like if you can teach a car, how to drive, which is a very complicated thing with lots of externalities and surprises, you can do so many other things that are simpler than that and whether it's like military robots or industrial robots where you can replace human functionality, that's much simpler than driving a car and much cheaper than driving a car.

Exactly. I want to give you -- one minute, give me -- I will give you an example that I run into. There's a Stanford research team trying to put the camera radar on the can of a full blind, right? And the camera and the radar control the wheels, which is connected to the can. So the can is leading by the camera and by the wheel. So that suddenly, the mobility of a blind increased by 50%, just with a small -- if you can optimize that device, think about that. I'm not saying it's a big market, but I'm just talking about a brand new application can help human beings get better lifestyles just with this one little device. And I think there are tons of different applications we can have with our platform and I'm looking forward to work with those teams to make it happen.

And what about like when you think about the existing markets, the state of kind of software development, professional surveillance is able to do at the camera level can do face detection, license plate detection. But it seems like down the road, you should be able to do like incident prediction, you could see muggings in progress, send alerts and things like that. Do you see people working on kind of the next generation of those applications?

Absolutely. In fact, that we believe the demand for the AI performance is going to continue to increase because you want to do higher and higher level of intelligence that require higher level of performance. So I really believe that's a driver for our road map is, when you go to -- when you build the technology, if a customer don't want to pay for any additional performance, you know this market is going to die. But I think AI is -- well, at the beginning of AI age, that people -- our customers continue to demand higher performance because of the reasons you just mentioned. And that will continue to drive the performance, therefore, driving our road map of innovation in the future. For example, at the CES, we demo how to use AI to do image processing and the video quality is have a step-function jump in -- at the low light. That's just one example, but to implement a simple feature require huge amount of AI performance. So can you imagine that all the AI demands from different angles will continue to drive the performance of the chip and drive innovation of the company.

Yes. All right. Good stuff. Well, that brings us to the end of our time. Feng-Ming, thank you very much.

Thank you. Thank you, Joe.

Thanks.