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

Good morning, everyone, and thank you for joining UBS' Future of Electric Mobility Virtual Conference today. I'm Neeraj Dhanoa with UBS, and I will be hosting today's fireside chat with Ambarella. I'm pleased to be joined by Ambarella's Co-Founder and CEO today, Dr. Fermi Wang. For those not familiar with the name, Ambarella is a provider of AI semiconductors specifically for endpoint applications, such as vehicles. In the auto space, the ADAS and autonomy domains are the company's largest business opportunities. And while there's a strong correlation of ADAS features in EVs, Ambarella's system-on-chips also enhance EV performance. Its SoCs are some of the most significant semiconductors in the market, which reduce energy consumption in vehicles, allowing less battery content and longer range. So I'm excited to have Ambarella with us today to learn more about the company's products and their future -- and their impact on the future of electric mobility. So with that, Fermi, maybe I'll hand it over to you to provide a couple of minute overview of how Ambarella is leveraging their approach and technologies that you've developed over the last 18 years for the evolution of ADAS and autonomy domains in the vehicle.

Thank you, Neeraj. Ambarella founded in 2004 and at beginning, we were focused on doing digital video application particularly for cameras. So for the first 10, 12 years, we focused on to deliver silicon solution to optimize for the best video quality, best compression so that -- for the video processing so that compressed video can be transmitted as well as focused on to provide lowest power consumption for all the application we just mentioned. So basically, in the first 10, 12 years, you can view us as a company to provide human viewing solutions like security camera, like sports camera, like -- even for the early automotive, we focused on human viewing application like e-mirrors and so on. So -- but however, I will say that after the IPO in 2012, we realized that all our customers are talking about how to add AI functions, basically, analyze video real time for video application. So we are -- that's the time we start investing on the computer vision technology based on neural network. And it took us a few years to come up a proprietary solution, which is really optimized for the computer vision AI application for the endpoint applications. For example, in the past, security cameras just capture video and store in the cloud. Today, the security camera can capture and analyze and determine, for example, the simple like face detection, license plate detection for auto, those kind of functions in a camera. Of automotive application, suddenly, we -- instead of just providing a human viewing application, we enable the, we call, machine vision so that the silicon running a car can analyze the video so to provide even more advanced function for back ADAS in Level 2+ car. So this is really a transition. We always [ will come to ] video, but we start with a human vision and try to transfer to a computer vision in the next several years. And we've become very successful on the computer vision side in the last 2, 3 years. You can see that we talk about our computer vision revenue. Last year, it was 25% of total revenue. This year, it's 45%, so show you the progression remained throughout the years.

Got it. And Fermi, what percent of your revenue is auto today? And how do you expect that auto percentage to change over the next 5 and so years? And I believe you've publicly mentioned before that auto becomes about 75% of your SAM over time.

Yes. So today, I think 25% of total revenue comes from automotive. And the main application is like for data log, to record video in a car or e-mirror or DMS, driver monitor system type of application. So that they are the majority of our automotive revenue today. But also, we add to new revenue stream into the automotive market, including ADAS. For example, we announced multiple ADAS design wins in Asia as well as some in Europe. But also, we announced Level 2+ car, like Rivian, is using multiple of our chip to implement the video perception function in the current truck. So that's the current revenue generation from automotive. However, moving forward, just like you said, the total TAM here on the right-hand side is our SAM growth. You can see that in the fiscal year '23, our total TAM is roughly $4 billion, and it will go to roughly $10 billion in fiscal year '28. Majority of that growth, which is dark area -- dark bars, are automotive. The biggest growth will be 2 things. One is the ADAS market. The other one is Level 2+ cars. And that's where we're going to focus on our efforts.

Got it. Great. And Fermi, I know one thing that's unusual for the Ambarella story as a semiconductor company is that your overall average selling price has been rising the last several years. And I understand that you expect your ASP to continue to rise, especially in the mobility market. Could you talk a little bit about what's causing that? And are you seeing -- are the supply chain dynamics allowing you to raise prices? Are you expanding the utility and function of your products by application, which is allowing you to increase more your pricing and charge more?

Right. I think our ASP growth is not because we raised the price on our current silicon solution. It's really about we add more functions to the silicon so we can charge higher price for the new solutions. The biggest example is when we talk about computer -- sorry, our solution for human viewing and transfer to the computer viewing. And that transformation increase our ASP. For example, I would say, 3 years ago, we sold only the -- here, right here. You can see that the -- if you see the table on the left-hand side, the yellow portion is our video processor revenues, and the green portion is our CV revenue. And the ASP difference between CV solution versus a video-only solution, CV solution is roughly twice higher at a similar spec. But because we added the CV function, so we can charge double the ASP. And with that, the different ASP, our company-wide ASP grew more than 35% in the last 3 years. And we expect that the ASP growth will continue in the near future because when we start selling silicon into the ADAS, the ASP will continue to go up because the functions of -- the content of our chip continue to increase. Therefore, we can charge more, which also probably [ exists at a pace ] similar to our competitive price. But when we move to Level 2+, the price will increase further. So I think that the trend we are seeing is that because we continue to selling more content in our silicon to the new applications, that will allow us to continue to maintain higher ASP -- in fact, increase ASP in the future.

Got it. That makes sense. And Fermi, while your [ content ] is going up, you're also bringing significant cost savings to OEMs. How does that work for Ambarella's ASP to go up but the total cost of the advanced cars come down at the same time?

Yes. For example -- let me use our customer, Rivian, as an example. We are selling multiple chip to Rivian today, and all the chips that we are selling to them only serve video perception function today. But when we move to a domain controller [ tied to ] like CV3 that we announced a few months ago, that chip integrated not only to video perception but also the -- all of the radar processing and all other sensor processing. And on top of that, we increased -- we integrated all of the high-level function like sensor fusion to path planning, which is currently running on a totally different CPU chip on a Rivian car. So you can see that while we generate [ north of ] $100 of ASP from Rivian, by integrating all the functions into the -- our domain controller chip, that helps our customers not only have more efficient implementation of the [ board ], which reduce the power to the power total DRAM cost and the total [ board ] sizes, but also reduce the number of silicon required to implement their function. I think that is a huge -- even regardless of our selling price savings, I think that, by itself, is a huge savings to our customer.

Got it. great. And I think the tip of the spear for your ASP story is your new product called CV3, which is an ADAS central domain controller. I know in the summer, you completed a significant demo and marketing push with your OEM and Tier 1 supplier customers, which we understand was quite successful. Could you please share an update on how this solution has been received by your customer base and any key developments stemming from that marketing effort?

Right. So we started talking about CV3 probably, I would say, 12 months before the silicon was available. And because I think CV3 presented really a very optimized solution compared to our competitors, and in those presentations before we have silicon, we [ claimed ] certain performance and power efficiency. And I think the first few months that we focused on to deliver those messages that now we -- the real silicon deliver the exact performance number and power efficient number that we promised in our presentation and we can demo that real time. And with that demo, what that means is really that our power efficiency of our current silicon is multiple times better than our competitors. And based on that, we are talking -- start talking to customer how to use CV3 family of chips. By the way, CV3 is not one chip. We're going to generate a family of CV3 chip, which will go from high performance to lower performance and to serve different applications. And now we are talking to customer how to leverage this CV3 family of chip to build a product family that our customer wants, right? So for example, the CV3 mid, that's the first ASIL chip that we're going to produce for our customers. That will be our high-end -- toward to the mid or high-end chip. And also, we are talking about to other lower-end chip that this is a family of chip that will help our customer to determine their product road map. So right now, I think that we have done all those things I just mentioned above. And what we need to do is really helping our customer to evaluate not only just the neural network based on our implementation, but also putting our customers' neural network to our system and to prove that their neural network can achieve similar performance boost as well as the power efficiency that we talk about. And we are working with the customer on that. As soon as that is finalized, I think we should expect fewer POC as well as design wins for our customers to kick off. And that's where we -- then we're going to have more engineering support engagement after that.

Got it. Great. And maybe, Fermi, we can double click a little bit on the power efficiency that you're mentioning. And I understand that your SoCs are very energy efficient relative to other competitors in the market. Could you just provide a little more detail on that? And maybe we can start with just kind of quantifying the watts differentiation between your products and competitors' products in the market?

Yes. So when you talk about chip, you need to talk about power efficiency and power saving. [ You have to prove power ] performance because you really have to talk about performance and power efficiency -- power consumption at the same time. So basically, what we did is we put multiple public domain neural network that our -- that was available both in our competitor's platform and our platform. And for example, the people -- a lot of people are using Orange. So we can -- and those number are widely available on Orange. So what we do is we compare the, first of all, the performance, right? And with our silicon, we can show people that with all those public neural network, our chip is running 3 to 4x faster than Orange chip. And at the same time, we also show them the power consumption. I think Orange's power consumption is [ probably ] well-known out there. And the chip that we demoed in our road show a few months ago, the power is running around 50 watts. So that gap that we have a more -- higher performance at a much lower power consumption, that's what we demo to our customer and that's where -- why people are really interested in our solution today.

Got it. And then, Fermi, how does that translate into battery performance? Because of the energy efficiency, you can help an OEM either enable longer range or reduce battery weight. Could you just help kind of frame how that energy efficiency translates into battery performance?

Yes. So when you talk about -- the power difference we just talked about is only for our SoC. But if you look at the total [ bond ], power saving is even bigger because we require less DRAM bandwidth as well as DRAM. And also, we don't -- because you require less power, so your power dissipation solution, it doesn't need to be as fancy as others. So if you look at the total [ bond ] saving as well as the power efficiency saving, not only just for the chip but the total system implementation, we are talking about significant power difference. And when we talk to our customer, we establish the fact that every watt difference, right, represents roughly $2 of cost of a battery. That's one way to look at the -- how much saving you can do. And also, you can convert that to the mileage you're talking about, but different car has different mileage. But even from a mileage point of view, we are talking about probably 10% to 12% savings just purely from this standpoint of view. So I think that those power efficiency is very important for the EV cars. And I also want to make an argument that all the EV car will have ADAS or Level 2+ function and that people are expecting the EV car going to become higher percentage of penetration in the near future, which also means that Level 2+ car will become popular, too. So for those kind of EV car, battery life as well as the battery cost is quite important, and I think we have -- we are in a very good position to competing there.

Got it. Great. And Fermi, could you just frame what the ASP is of your CV3 SoC?