Welc ome back, everybody. I'm Joe Moore from Morgan Stanley. I think that was loud. Morgan Stanley's semiconductor analyst. Very happy to have with me here today the CEO of Ambarella, Fermi Wang. Thanks, Fermi.
Yes.
So maybe we just jump right in. You reported earnings last week, had a very good quarter, both the revenue and the outlook. Can you give us a recap and tell us where you're sitting here today?
Right. Thank you, Joe. I think for the last few quarters we have been doing quite well since we come out of the inventory correction. And with this quarter in earnings, we provide guidance that this year we're going to grow anywhere between 15%-19%, midpoint 17%. I think that's very healthy growth. And both markets, our IoT and auto, both are going to grow this year over last year. And the driver is really about our flagship products. Both CV5 and CV7 are ramping up quickly. CV5, last year, we said we delivered more than a million units of the shipment. And then this year we want to continue our growth on that. CV7 is our new chip for IoT that starts shipping in Q4 for three key customers. And this is a chip we can talk later about.
This chip is capable of running from the traditional CN Neural Network to the CLIP to the DeepSeek type of reasoning model. And we should reserve time to talk about that. But overall, I think financially we're doing quite well.
Great. So I wonder if maybe we could start the conversation a little bit more long-term about the auto market. And you guys have had amazing technology leadership now for a number of years. And I'm still telling people, you look at some of the demos from your OEM, Tier 1 partners mostly, I should say, really impressive capabilities that show like 12 cameras and sensor fusion and all these advanced things. And we're still in a world where not a lot of that stuff's happening, at least outside of China and maybe Tesla. Can you just talk about what you're seeing in the investment in ADAS now? I mean, people can't ignore the FSD having moved the world forward for Tesla. Can't ignore that China's doing a lot of innovative stuff and will eventually sell cars here. Is that stimulating investment in ADAS?
What form is that investment taking?
Right, so first of all, I think you are right that the FSD, as well as BYD's announcement just a few weeks ago, really continue to push this Level 2+ market moving forward, which is consistent with what we have been saying for the last two, three years. Two, three years ago, we started setting up the goal that we want to build a family of chips of CV3, targeting from the low end of Level 2+ all the way to Level 3, Level 4, and we did that. We have the silicon available. We assemble to a customer. We have our software stack that runs on our car, which integrates the software stack and also end-to-end AI, plus 4D imaging radar that we acquired from Oculii, so from an engineering point of view, we do deliver a complete roadmap to our customer.
In fact, more importantly, in CES, we demo on the CV3 high-end chip. We integrate a large language model for the reasoning purpose on the running in parallel with our software stack in a car. So overall, I think technology-wise, I think we look at this market three years ago. We have delivered the product family we think is important, and which is definitely what the market requires right now.
It still seems like the investment shifts a little bit towards what you're calling L2 plus versus full autonomy.
Right.
I guess, do we have that Level of sensor fusion where we need a dozen cameras or those kinds of sensor inputs? Are you kind of, in a way, moving towards more of a maybe smaller revenue opportunity per car, but one that can get adopted more quickly?
Right. About a year ago, in this conference, we talked about our Western customer stuff focusing on what they call value engineering, basically trying to deliver a high-performance solution at a competitive price, and we do see that this trend becomes continued in the last 12 months and highlighted by one of the projects we're building on, and it's definitely that they try to compete with a Chinese offering that really either highway or city-Level autonomous driving, but with 5-6 cameras with radar, and that's the sweet spot that we're seeing a lot of our OEMs asking for right now.
And maybe talk about that a little bit. So you've been sort of talking about the potential to land a large win with a production OEM. And you said on the call that maybe that's not happening, or at least not happening on the timeframe that you originally thought. Can you talk about that? I guess it's a customer we don't know about that you maybe won and lost. But what does that tell you about the longer-term opportunity? And how does that shift your focus as you kind of think about things going forward? Because that was a win that was still several years away, and now you didn't get it. Does that mean other CV3 opportunities also pivot to different technology Levels? Or just how are you thinking about that?
Right, so first of all, in our conference call, we talk about we lost a major Western OEM. And in fact, we also talk about in that the feedback from the OEM directly is that we think we have better technology and we have a competitive price. And we lost because of some other financial considerations. But given that it's a loss, and when we look at this, we definitely believe we learned a few things. First of all, it confirmed that working with a large Tier 1 does give us the scale we are looking for. Particularly in this particular case, there were probably a dozen Tier 1 with all kinds of different silicon solutions in there, and we were, in the last round bid, us versus incumbent. That shows that we do have the best technology as well as the competitive price to compete.
So that's a confirmation that we think that we have. The second thing is that throughout the process, we think that working with a large Tier 1 is the right strategy because without that, we cannot get there. So what we need to do, I think, first of all, we're still building RFQs that we talked about last year in our final discussion. Last year, we talked about still building dozens of RFQs. And they are still alive. And most of them are still in that same target category about value engineering. So I think we still continue to work on those cases. But of course, we need to be more sensitive about other financial considerations that our competitors might play. But that's something our Tier 1 and us need to continue to do better. But throughout the confirmation, I think we are confident with our technology and products.
It's our plan to continue to push this technology and product into the automotive space.
I mean, that has to be incredibly frustrating to sort of hear that your technology is better and cheaper and didn't get the socket. And I guess the question that people ask me, and hopefully this doesn't seem like a hostile question, but your three biggest competitors, Qualcomm, NVIDIA, and the Mobileye segment of Intel, those are the three largest semiconductor companies in the world. You guys are small. And you've had leadership technology now for a number of years. How do you get past that? How do you get people to really say, obviously, Bosch and Continental were huge wins for you guys that sort of said, this is the best technology for us going forward. How do you take that to the next Level and kind of get the OEMs to get comfortable with those ancillary things?
Right. So in fact, all the three companies you mentioned are in this design win we talked about, and we really show, like I said, on the engineering side and technology side that we are there, now the issue is with the financial consideration. You can say relate to scale, but I think at the end, it's really about business negotiation happening at the end, so I think we understand the scale problem. We need to get scale, but working, like I said, working with Bosch and Conti continues to be the direction we work on, but at the same time, when we negotiate next year, we need to be very sensitive about how to upsell our technology to make sure that our bond differences, the advantage of our bond, can cover whatever that the advantage that other competitors can offer, so it's difficult.
And it's also frustrating, like you said. But however, we are committed to continue this path.
Yeah. OK. And how much of this you've had so much emphasis in the last really three or four years on EV. And you're dealing with Western OEMs that probably have somewhat limited scope for innovation. And a lot of the innovation was going in that direction. And it seems like, obviously, that's a great trend for you because you have such power-efficient products. But it seems at this point that innovation needs to come back. It doesn't seem like we can just say that we don't need that kind of capability to move forward when, again, you have this competition from Tesla, competition from China that's coming.
Right. So I agree with you on what you said. I think that the key learning is this: with the technology, it's not enough. So we need to make sure that in the next process bid that we work with our Tier 1 to make sure that we don't make the same mistake of being sold over by just financial considerations.
Yeah. And you guys have a lot of great OEM relationships. I know about it. I mean, you've been working with some of the biggest OEMs in the world, with Oculii and with some of these other technologies. And I know you have those conversations every day. So maybe we could talk about the funnel a little bit, the Automotive funnel. Over $2 billion for a company doing less than $100 million a year for a six-year funnel. Pretty impressive potential ramp there. You've tweaked that down a little bit each of the last few years, which I actually like because I think it shows that it's actually an authentic process of going. We know the activity slowed down. So I feel like you're being rigorous about that. But how do you think about that funnel?
It doesn't seem like this loss is necessarily that big of a piece of it. If you end up in a world where there's a lot more singles and doubles and less home run kinds of wins, can you continue to grow along the lines that your funnel would suggest?
First of all, those definitely take down the funnel somehow because the way we define funnel is that it's a six-year window. So for this particular case, it's only three years of design within that six-year window. And also that the design win total value is discounted by the possible winning factors. So with that, I think there are some discounts there. But I believe that all the other we are all bidding on other projects. Hopefully, they can cover that. But however, you are right. Getting singles and doubles help, but it doesn't solve the problem. We need to get focused on winning a major Western OEM as a first company target. And that has to be our first priority.
What is the lead time around that? I mean, to the extent that you lost one, it's no big deal. There's a lot of other potential ones out there. But it seems like it takes a long time between maybe not when you tell us, but when that sort of engineering work begins and when the revenue.
We're still bidding on a few. And maybe a couple of them will make a decision if they don't delay. So I think there's definitely stuff we're still cooking with our OEM customer. And we are giving demos and, in fact, inviting people to look at our software stack because right now, people are not just looking at the silicon solution anymore. They need to have a total software solution. So we have multiple engagements cooking right now. One thing that is for sure, everybody knows automotive business, the whole market is slow. And also, we start seeing the decision got pushed out because for different reasons. One of the biggest reasons, in fact, that when BYD announced this brand new roadmap, I think a lot of people started scratching their heads and said, should I continue my current RFQ? Or should I target differently and compete with BYD?
So all those kinds of things can delay. But our commitment is that we'll work with the OEMs that have RFQ out there and trying to get design win.
And who do you guys feel when you're bidding for these? Can you talk about the competition a little bit? Who you deal with? Because it feels like Qualcomm is very much on kind of an L2 integration with infotainment. They're not really trying to add a whole bunch of advanced features. And Mobileye, where they are, they sort of have a little bit of a jump from where they are now to autonomy. And maybe the bridge, to me, is a little tricky to figure out. And then NVIDIA is really autonomy. I mean, NVIDIA is not really doing much for that. So it seems like you guys do something that, other than what Tesla is doing, your chip, to me, looks very similar to theirs.
There's really not a lot of other people doing AI inference at the edge at low power for automotive applications in the way that you do it.
I agree, and I think that from a competition point of view, NVIDIA is really targeting the high-end because of the graphics chip, and then Qualcomm, like you said, is trying to take advantage of the infotainment system and bundle everything together. Mobileye is a black box, and they are incumbents for many design wins and are trying to sell that, so we have a unique opportunity there, but like you said, the key is that when we deal with those big incumbents, that we need to be careful about not being overwhelmed by the financials, so however, we still believe, especially through this competition, that our strength is in our AI programmability plus our video processing capabilities.
Yeah. Great. And then you've mentioned a couple of times the kind of reasoning element of this. That seems like a relatively recent innovation. Can you talk about I think of that as being intuitively obvious in the language domain? But how does that work in your domain?
Reasoning models.
Yeah, reasoning models.
First of all, just until a month ago, I thought the reasoning model can only happen at 100 TOPS or 200 TOPS chips. Just last week, we demoed the DeepSeek 1.5 billion parameter reasoning model on a two-watt chip of our chip called CV75. That just shows you how fast the technology changed and how fast that the very complicated model can come down to the edge AI and change the application. You can imagine in the last month that conversation has been centralized. How does this application, how does this combination of a reasoning model on a very power-efficient silicon, what kind of application you can enable at the edge device or edge infrastructure? Those two sides are both locations we're talking about. Edge device is quick, the quickest way to get the revenue. The application is just amazing.
For example, for all the security camera guys in the past, with the CNN-type neural net now, you know there's a person, you know there's a car, but you don't know what's the intention. Right now, with the reasoning model running on a CV75 2-watt chip, suddenly you can start guessing. The model can tell you what's the intention of the people stopping there in your gate or parking your car, close your door, or delivering a package. The reasoning model starts telling you the environment as well as the intention of the object. That is not possible in any security camera in the past. That's just one example. Think about how other possible applications that you can install at the edge device and using a reasoning model to understand what's really going on. It's not just how many objects around you.
It's what's really going on around you. That, I think, opened the door, but I think that's just, for example, we already shipped CV72 in customer, and the first wave of customer CV72 to run CN type network later this year. They're going to run CLIP type network on CV72, and next year, I think they're going to run a reasoning model on CV72. So they can leverage the same hardware and the same infrastructure to replace that with a different model. They can enable different applications. That, I think, is a powerful message to the industry that people can leverage their R&D expense for different applications. And that's just a 2-5watt chip. And the other possible application is this building alone probably has, I don't know, a few hundred cameras. None of them is capable of running CLIP or reasoning model. How do you enable that?
Not to replace the camera, but to install some edge servers that are sitting in some closet that can run those CLIP or reasoning model locally. And that's how we view that edge server can really quickly upgrade the current infrastructure for different applications. So both markets we're looking at. And we have great products. In fact, if you look at our family of chips, we have CV75, CV72 for IoT camera products. We have an N1 and N1 655 for edge servers. None of them is designed for reasoning model. But however, all of them can run reasoning model. So you can assume that we're going to use those chips to engage new customers and new applications. And our next generation will be optimized for these applications.
I mean, the applications in surveillance seem really clear. What other opportunities in IoT does this open up?
Wow. Good question. I think it's going to open up thousands of different applications, which I'm not probably in the best position to say it. But I think when I talk to enough customers, at the end, give me three months, I think I probably can come back with a report, at least the engagement. I personally think there's tons of possibilities. And just to give you an example, a 7-Eleven clerk put on a wearable camera with reasoning model in there. Suddenly, you can understand exactly what happened in your store. It's not just objects. It's just a customer. How many customers there? What are they doing? What are they doing? It can become a marketing tool for you. So there are multiple things. That's just something that just comes to my mind.
Yeah.
I think there are tons of other possible applications.
I mean, it is amazing because you have all this really cool technology. And the automotive industry is so slow to adopt any of it. And then you start thinking about humanoid robots. You start thinking about the way the world is going. And I know you don't want to get into prematurely hyping those things up. But I mean, the reality is the features that you're able to put in a car that nobody's putting in a car are very useful in military drones, in delivery drones, in all of these applications. And I mean, it seems like, given what's happened in the last 18 months, the company that has the best edge AI capability on the planet, the stock has kind of done nothing because we're worried about cars. And you have all these opportunities ahead of you. And I know you still want to invest in cars.
And there's still a lot of exciting opportunities there. But does that change the way you're thinking about your investments going forward?
I expand our investment strategy now. Automotive is still part of it. But I think when I look at the potential application that we can go, you mentioned drone. We were at drone 10 years ago, that DJI will put on Entity List, stop that. But we know that a drone is really just a flying autonomous driving machine. And all the things that are required to do that, we have the CV3 can do all of that. And we know if there's an opportunity, we will catch on that. So I think that's just one example of a robot. I think drone is a robot that's going to continue to expand our applications. I think robot, in general, I have been saying this for five years now. But robot, in general, is a target market. But I have to say it's very segmented. It's small volume.
We will continue to have a way to engage customers. I don't think we can count on Huawei to drive our revenue anytime soon. I think you mentioned drone. You mentioned other potential edge AI applications. That definitely opened the door for us for other applications. I think this reasoning model is a game changer for us. Hopefully, that in three to six months, I can report back with a potential customer that we can work with.
Great. OK, so pivoting back to automotive, the singles and doubles, I know is not the end goal. But you picked up some nice ones there, and the digital rearview mirror, you've been talking about for a while. You had a win there with NIO last quarter. Can you just talk in general about the progress that you have towards those types of applications?
First of all, I want to point out we have revenue growth in automotive both last year and this year. I think that's rare considering the current market, and that's because.
A couple of tough years for automotive.
Yeah. And that's because, like you said, the bunk, the singles continue to help to have growth. Driver monitors, e-mirrors. In fact, we picked up the Samsara, which is a Telematics for their driver monitors, as well as monitoring the ADAS. All of that is really helping us to have a growth. And the reason we can have that is because it's edge AI growth. In fact, we have a very powerful accelerator, which is common throughout our whole product portfolio. Although we are talking about IoT and auto, but if you look at deep inside, the same AI accelerator is being used in every chip. So one customer, if you use our AI accelerator, the same network can be reused in all the other silicon in our company. That helps us. And also, that helps us to reduce our R&D costs.
With that, the automotive growth is really about having one AI accelerator plus very powerful video processing enables automotive. It takes us a long time to get to the e-mirror or driver monitor system because the market didn't develop quickly enough. As soon as it develops, I think we start showing the.
Yeah. You have the sockets as the markets develop. Yeah. Can you talk about China? China, you guys have very good relationships there. I remember back in the days of surveillance, those two customers in China used to be huge for you. And then there were export control reasons that went away. Within automotive, there's a lot of innovation, obviously. And you've announced Leapmotor. And you have some important wins. But you've also said that it's only, I think, 15% of your funnel. So can you talk about the challenges of that market and the opportunities in that market?
If there is one company hit by the Entity List the most, it's got to be Ambarella. We lost Hikvision, Dahua, DJI.
It was like 35%, yeah, almost 40%.
40% of total revenue lost in just one shot. But however, we are still committed to the Chinese market for different reasons. But also, we understand we cannot because most of the Chinese OEM tier ones are favorable, are picking domestic components as a first priority. We know that. We understand that. However, we do believe that all the customers, if you talk to the Chinese OEM Tier 1 , they all want to do export business. And they realize that selling Chinese components and the software to Western world is a challenge. So we do view that if Chinese OEM tier one want to sell export business, they need to have a diverse supply chain, one for China, one for non-China. And we think as though we are targeting those companies that want to do export business. And we want to sell heavily to them.
And we believe that's the best chance for us to compete with China. I don't think we have a chance to compete in the domestic Chinese market. But on the other side, I think that's open for us.
And why do you think the markets evolve that way? I mean, we saw the same thing in surveillance, where at one point, Hikvision and Dahua, you were mostly in their export cameras. Why aren't the Chinese solutions able to be exported in the same way?
I think regulation is the biggest one. U.S. government already said that in U.S. automotive, no Chinese component or software allowed. So regulation definitely is and also geopolitical situations that continue to be a factor. I think geopolitical is behind everything. If we allow that, we can probably still ship to China for security camera and automotive. So I think this is just a reality we need to deal with.
Yeah. And can you talk about I mentioned your competitors being three big multinationals. But in China, you also have domestic competition. Can you talk about the capabilities of?
Yeah. So for IoT, the biggest competitor is Huawei, HiSilicon. Although they've been banned, but they're still there somehow. So I think we'll still deal with them in China. In automotive, it's Horizon Robotics. I think this, with all of the support they got from the government and domestic suppliers, so I think they're going to be very strong in China. But like I said, I don't view them as competing with them anymore because I'm not competing for domestic product. I'm competing for the export business. I think that's separate.
Yeah. OK. Can you talk about the N1 product family and what applications are promising there?
Yeah. We announced N1 two years ago when the Gen AI started. And throughout the two years, we come to several conclusions. One is trying to compete with NVIDIA at the cloud is a debt. So we focus on edge AI. And here, we think we have a lot of differentiation and strength. Second, the application we just talked about is two years was Gen AI only. Last year, CLIP came up. This year, with a reasoning model. All of them help us to justify powerful AI silicon at edge. And that helps us. And N1 and N1 655 family is designed for edge infrastructure. And that's the purpose of that.
Your guys' approach seems good when the market moves because it's more of a programmable solution. When people can tailor custom silicon for something that takes two years to develop, you have more of a programmable AI solution that can morph as those new workloads come along. Is that fair?
Yes, but however, we are not totally programmable like NVIDIA GPU chip. In fact, think about this. Our biggest differentiation is power efficiency, and you have to hardwire some of the operations inside your silicon to achieve that power efficiency. But at the same time, you need to leave enough programmability, so when there's a new model from CNN to transformer to reasoning model, you can continue to use your silicon. Without risking your silicon, you can address that, so you need programmability. Also, you need to have a hardwired portion of your design so you can get power efficiency. The combination of these two is the difficult part. How do you cut the line between hardwire and programmability? I think that's where our biggest value. We have gone through this many times for video processing, for compression efficiency, for AI performance, now for modeling.
We proved many times that we know how to differentiate, how to offer a solution both power efficient and programmable enough for people to switch, and I think that's the biggest value we have.
Great. So let me ask one more near-term question. And then I'll open it to the audience if there's any. On the call, you sort of talked about some conservatism in the back half. There was some maybe tariff anxiety there, things like that. Can you just talk to is that conservatism on your part? Is that something specific that you're seeing? What drives you to make those statements?
The only thing we see is there's a handful of customers, well, more close to Asia, that have a really different ordering pattern in the Q4 and Q1. That's the only thing that triggered us to start thinking, what happened, and we go back to ask those questions. In fact, ask questions to those handful of customers and all our customers, see whether they are doing any kind of inventory accumulation. The answer is no for everyone. But we believe that we think when we look at the strange ordering pattern, we think it's good for us to put in some conservatism in the second half just in case something happens. There's no proof on anything. But however, even with that conservatism, we still show 17% growth year over year, and if that conservatism is proved to be not true, I think we can show even better.
In my mind, we are presenting a model that has more upside and downside this time.
Great, and then, I mean, as we think about tariffs, the automotive business seems clear. If you're assembling in Mexico, Canada, then there's probably some disruption, but on the IoT side of the business, do you have any of those same kind of geographic sensitivities?
Well, I think so. But however, I just want to—this is public information. Our biggest customer, Motorola, they have both manufacturing in Canada and Mexico. But if you look at their inventory, it's the lowest possible in the last three years. So that just shows you that people are taking this in a different approach. Some people panic. Some people think, let's wait and see. So it cannot be generalized to this. And I do believe there are a few people who say, well, they have a knee-jerk reaction, a.k.a. let's buy some more so they protect themselves. But other people say maybe this is just a short-term policy change. We should wait for the election. So overall, I won't generalize it. I don't see that in our U.S. or Western customers, but more in the Asian customers, we see this kind of strange ordering patterns.
Great. Let me see if we have questions from the audience. One up here. You wait for the mic just to the guests. Thanks. So you spoke about the market for robotics. Can you elaborate more about the time frame for the robotics market? Because that would be a big market, I would think.
Right. So when I talk about robotics, I need to be specific about which kind of robot you're talking about. An autonomous driving car is a robot. A drone is a robot. A manufacturing robot is a robot. The home cleaning machine is a robot. And the humanoid is a robot. I think all of them come to one conclusion: you need a powerful domain controller that can take in multiple sensor inputs and do a sensor fusion. Then you start doing planning and taking actions. This is a highly defined robot. But it can be for different markets. The only true robot market today is automotive. That's why we focus on that. I think drone will become important. But today, it's dominated by DJI. I think there are multiple companies trying to challenge them.
I think that we want to be part of this market when it happens because I think that's a real market today. All the other robotic markets, other than this home cleaning machine, which is very low-end, we don't play. But for other robots, it doesn't matter if it's AMR or manufacturing robots or humanoid. I think it's very segmented and small volume. And it will take time to get there. I don't want to predict what Elon Musk predicted in 2015 that autonomous driving should happen before 2019. And we haven't seen it. It will take time. But I believe it will happen. But you just don't know when. My job is to focus our limited resources on the market that can drive volume. For me, it's Automotive and drone and maybe some of the short-term opportunities.
Following on from that question, could you help us understand would verticalization be the best way to play so you'd have your own hardware and software stack for sensor fusion and the camera structure so that you've got and then what would be the price point? Would it be several hundred dollars for this sort of solution?
For software, you mean?
I'm talking about the robotics solution here, either humanoids and/or cars for that matter.
Right. So in fact, I can use our automotive solution, to give you an example. Our silicon is anywhere between $100-$400. I think that can apply from the low-end robots to the high-end robots. Maybe low-end robots can use even our $50 chip considering what's happening. I think for low-end drone, a $20-$50 chip can work. But if you want to go to really the higher-end humanoid, that could be on the high-end side. But there's a software play too. And the software play, rule of thumb is I think the total software stack value is probably as high as the silicon value. So when I price our autonomous driving software stack, that's basically the rule of thumb I use. Thank you.
Great. Well, I think we'll wrap it there. Fermi, I hope the spirit of this for me is if I'm frustrated, it's that the automotive industry isn't implementing really cool capability that's out there. And I'm just trying to test where you guys are at with all that. But I'm really excited for the technology you bring and for that technology to eventually get used. So thank you very much for your time.
Thank you, Joe. Thank you.
Thank you.
Thank you, guys.