Good morning, everyone. Thank you for joining. My name is C.J. Muse. I follow semiconductors, semiconductor equipment industries here at Cantor. Very pleased to be hosting Ambarella Inc. And we have Louis Gerhardy, VP of Corporate Development. Welcome. Yeah, thank you, C.J. Thank you for having us at your first conference. Yeah, yeah. Great to have you. Maybe just to start, it might be helpful given you just reported a couple of weeks back, maybe a brief recap of the key takeaways. Yeah, at a high level, Ambarella, the last five years has been an AI at the edge story. A major takeaway from our last earnings call, and it was our fiscal fourth quarter, is that we exited the year with AI Edge, almost three quarters of our total revenue. Let me explain what that means, because for a lot of companies, if they're in an AI application, that counts as AI revenue. For us, that 75% or so of our total revenue exiting the year that was from AI. W e're actually doing that deep learning processing in our own proprietary AI accelerator at the Edge. It's not like we're sitting next to a GPU and therefore we say we're doing AI. We're actually doing the deep learning in all of these applications we're in. Financially, our business has been and remains strong. For Q4, our revenue was 8% above consensus expectation. For Q1, our guidance at the midpoint was 13% above consensus. For the year, consensus ended up revising their revenue for us up about 4%. The business is strong, and we're starting to see our edge AI processors beginning to move into more and more markets. I'll pause there. Perfect, perfect. Maybe just to take a step back, for those that are less familiar with Ambarella, could you perhaps give it kind of an overview of the transformation of the company towards AI computer vision? Sure. Ambarella has been in business for more than 20 years. In the first 15 years, we established a reputation in camera applications where in front of us in a camera application sits a CMOS image sensor from, say, a Sony or an Onsemi. We are the digital chip that sits behind it. For the first 15 years, we collected the information coming out of the CMOS image sensor, and we would make it a very high-quality image with very low power in very difficult environmental conditions, which would range from you are in a dark tunnel and you are looking at brightness or vice versa, or it is snowing or it is foggy or the sun is shining into the camera or it is too dark. We had a reputation in portable devices, again, frequently operating off a battery, to get the best image possible. What has changed now is we have leveraged that capability, which you can think of as now ingesting data through the lens of a camera. We integrated into it a proprietary AI accelerator, which ingests the data coming in through the lens of a camera. The AI accelerator makes the decisions, which allows the electronic system to be partially or completely autonomous. That is the transition we have been through. Again, exiting the year, almost 75% of our revenue was coming from these AI SoCs for the edge, where the chip is doing both the camera perception and the AI processing. We serve a variety of different IoT markets, both auto as well as just the very broad IoT space. Maybe just to level set kind of where we are, if you think about the 75% in 2024 on an actual basis, how would you kind of apportion the different kind of sub-end markets within IoT and auto? I'll answer that question for our overall revenue, because we don't typically break it out just for AI. It's going to be very similar ratios. For the IoT business, that's about 70% of our revenue, and auto is 30%. Within the auto business, there are two different buckets. One would be, think of like IoT type applications, for example, telematics or driver monitoring or electronic mirrors or some front ADAS or personalization and access control features in a car. That business I just described represents almost all of our auto revenue today, and it's on a really nice growth trajectory. The other part of our auto business is very new, and it will command much higher average selling prices. This is what we call the CV3 platform, which is there to serve L2+ to L4 applications where our chips are actually navigating the vehicle. That business, we said the first revenue would come in calendar 2026, starting in commercial vehicle applications. We continue to believe that's the case. Switching gears over to our IoT business, that's about 70% of our revenue. The largest part of our IoT business would be what's called security. Within security, there's physical security. Within the physical security market, there's the enterprise part, which is the largest for us. There's public security, and then there's also the home type security business, all with very different dynamics. The other 60% or so of our IoT business is coming from a lot of different markets. It's IoT, so it's difficult to hang your hat on any one. This would include things like body-worn cameras, wearables. It would include portable video cameras. It would include access control. It would include enterprise video conferencing. It include some VR business. There's a lot of areas in our IoT business that for the first time are implementing Edge AI to bring new utility to their markets. That's what's so encouraging, because our foray into AI actually was in the enterprise security market. In fact, that's where AI originated in the Edge. Now we're seeing signs that it's beginning to move into some of these other IoT markets as well. Excellent. Before maybe going into the various end markets, we'd love to hit on technology kind of perspective. You spoke around AI inference at the Edge, very low power, and how that's accelerating. Can you kind of speak to your competitive differentiation there? Sure. Yeah, what we face at the Edge is so different than, say, what's in a hyperscaler or a data center and server. We have a very specialized AI accelerator architecture that's very different than a GPU in the way it works. To create that architecture, we have a fundamentally different approach than what a lot of our peers or competitors have taken. That is where before we started to lay out transistors, we took a look at the type of AI algorithms we thought our customers would need to execute on our chip. We created an architecture optimized, purpose-built for those types of algorithms. It's this approach that differentiates us. The reason we've been successful, by the way, we've shipped cumulatively 30 million Edge AI processors to date. We're on our third generation of AI accelerator. The reason that we've been so successful with the AI accelerator is that ultimately, with this unique approach we call algorithm first, it yields a superior performance per watt per dollar. That's what's required to be successful at the edge. To put it in perspective, our blended ASP just for our AI products today, the 70% of our revenue, is about $20. We all know how much a GPU costs. It's a very, very different type of business. A little bit more. Right. All right. I'm going to go out of order here, because I want to dig deeper into what you just said. I think on the last call, you talked about how DeepSeq has a net positive to Ambarella as open source reasoning models and enable enhanced intelligence decision making. H ow would a consumer kind of implement DeepSeq onto your device? The consumer wouldn't, but whoever's hardware system they're buying, they would have it. To put our AI business in perspective, there are waves of technology. Right now, of the 70% of our revenue that's AI, it's all based on CNN type networks for computer vision functions. There are many different ones, but it ranges from face detection to line counting to attendance taking to license plate reading. There is a detection and classification type nature to what these CNN networks are doing. That's all of our AI revenue today, and it's a growing part. Our latest generation of chips, and I'll just mention the family CV7, N1, and CV3, all have our third generation AI accelerator in them, which can do more advanced AI networks where transformer networks are the enabling technology behind this new wave of technology. This is where Gen AI comes into the story or reasoning models, for example. At this time, we're just demonstrating. In fact, I gave my first demo last week of DeepSeq 1.5 billion parameter model, running in a chip that ranges from, we have two different families, 1.5 watts to 5 watts, doing 1.5 billion parameters. We'll be showing that to customers for the first time at ISC West, which is a major physical security and access control conference coming up. By the way, if any of you would like to see that demo, I can help you get into our suite there to see it. Just let me know. We're at the stage for these advanced AI networks, including reasoning models, where customers are going to see the capabilities. What are the capabilities here? It's improved contextual awareness that can make the system come to better conclusions o r it's more complex reasoning logic, where you break a problem down into smaller pieces and it yields a better conclusion o r diagnostic type features. All of these things weren't possible with the generation that represents 70% of our revenue today, which is computer vision. Exactly what applications will be the killer applications, it's really hard for us to say, because we're providing these super efficient advanced processing resources, and our customers will be the ones that create the applications. Our customers' customers will be the one that ultimately decides what's going to be most successful. It's got a lot more power to it, a lot more utility, a lot more efficiency than the current generation. Maybe focusing on the technology side for the third-gen accelerator, is that what will be kind of good enough for implementing these large language models at the edge o r is that kind of the first iteration you're going to need to build a new generation? Just maybe to better understand what that accelerator looks like, I mean, is that just a chip with gobs of SRAM? How would you kind of define it? Yeah, this algorithm first approach to our architecture has yielded what's needed at the Edge is extreme levels of efficiency. What it ultimately yields, fast forward, is that we don't have to go on and off chip a lot to access just commodity DRAM. We don't use high bandwidth memory. This is one of the underlying reasons why we're so efficient. Embedding large amounts of SRAM on the chip would be very expensive, because SRAM is a very expensive form of memory. That's not the way we do it either. It's really in the architecture and how the information flows within our chip and doesn't have to go on and off the chip to DRAM so much. That's the underlying advantage that we offer. I think I might have missed the first part of your question, though, if you could remember. In terms of your third gen accelerator, is that what you see as the go-to-market solution o r will there be a next generation that would really deliver the goods for a DeepSeq or other type of model at the edge? Yeah. Our third generation, which just started to generate revenue in Q4 for the first time, and it commands a much higher average selling price on average than the current generation of chips. That is something we should talk about later is the ASP ladder we're on. We can very efficiently execute these transformer models and reasoning networks on our chip. For sure, there will be a fourth generation of our AI accelerator architecture. We just haven't provided much information on when that will happen. What node is the third gen on? The third generation is all 5 nanometer. The last product on the third generation is 2 nanometer. That is the custom ASIC project that we're working on. The second and subsequent 2 nanometer chips will most likely be on that fourth generation of technology. Got you. Makes sense. Maybe looking at a segment basis, f or automotive, on your last earnings call, you discussed attaining final round, but losing the bid for CV3 ADAS. You said not a technology issue, not a price issue. W here is that particular customer on your scale? Is that going to be a recurring challenge? How can you kind of mitigate that? Does your partnership with Bosch and Continental kind of provide cover? Within that entire construct, how might your go-to-market strategy evolve? Yeah. Recall our auto business is two categories. The category that represents almost all of our auto revenue today would be for, say, smaller systems within a car, not navigating the vehicle, but doing features like driver monitoring or electronic mirrors or front ADAS or telematics, AI video telematics. That business isn't subject to some of these incumbency and scale issues like the second part of our auto story is. The second part of our auto story revolves around the CV3 platform. These are chips that navigate the car in scenarios that range from L2+ to L4 levels of autonomy. You're correct, we did say in our last earnings call, there was a project that would, the first full- year of revenue would have been calendar year 2028. We did not win it after being a finalist in that process. I think I understand why you'd say scale. I think before you talk about scale, it's lack of incumbency in some of those markets, which led to a scale issue. What we understand is there were other financial considerations that led to us not winning that opportunity. There is a variety of tactics and measures we can take to address that going forward. We are in the process of discussing those now. Obviously, aligning yourself with powerful tier ones like Conti and Bosch is certainly part of the strategy. Continuing the discussions and developing other tier one relationships is also going to be very important. Is there a framework for thinking about kind of your core underlying auto business growth, CAGR, and then on top of that, how to think about CV3 and ADAS? It's kind of gravy. Yeah. I think for the core business, which would be the smaller systems, that's almost all of the 30% of our revenue that was auto last year. That's about $80 million. Think of that as having a CAGR, and I'm going to give you a big range here, and I'll explain why, 15% to 25% like five-year revenue CAGR. The reason that range is so broad is because many of these applications we serve in this core auto business, the penetration rate of the technology into cars produced is quite low. Take, for example, electronic mirrors. Maybe 2% of all the cars produced last year had these electronic mirrors. If that penetration rate rises faster, then maybe that part of the business can grow closer to the 25%. If that penetration rate stagnates or does not grow very fast, then that will be the factor that caused it to come in at the low end of the range. It is a portfolio of different automotive systems. One that is quite strong for us and doing well would be AI video telematics. We announced, I think, Q3 of last year, Samsara became a customer. That by itself is just a great opportunity. As AI finds its way into telematics, we are helping power that. On the CV3 platform, it is much more difficult to put a five-year CAGR on that business. What I can do is talk about revenue commencing in that market in the commercial truck market. For example, it has been announced that our chips are used in the fallback system at Aurora in their L4 trucks. A company called Kodiak, which is in the same sort of market, is using us. We have several other wins. Our wins in that market are in the commercial vehicle market, which tends to be lower volume but higher value. Until we land, say, a portfolio of wins, including passenger vehicle wins, it will be difficult to kind of put a CAGR on that autonomy type business. Yeah. Makes sense. From a potential win perspective, Tesla's internal, I'm assuming China is going to go domestic. W hen you look at the overall kind of market for automobiles, are there kind of two or three, I don't want to call it low-hanging fruit, but two to three kind of customers that you're focused on? What's the best way to kind of think about the prospects of truly driving ADAS and passenger vehicles? Yeah. For the core auto business, there's lots of opportunities out there. For the autonomy side of the business, there's fewer opportunities. The ones that exist are very large, given the ASP is higher. I'd say for China, it's 15% of our auto business today. It's 15% to 20% of it. For, say, our opportunity, we describe something called a funnel of opportunity that covers a six-year period. China is about 15% of that too. You're right, though. The domestic silicon, when it's good enough, it'll get the business. There's a number of companies that are doing their own or trying to do their own custom chip. Until they understand how these chips work when they come out of the fab, it'll be hard for us to land that business there. The two filters I'd apply on the China opportunity is, one, do we have a technology advantage? Is this performance per watt per dollar that we offer really that compelling to them? The second filter is, are they exporting vehicles? Because if they export, the opportunity for us to win it is much higher than if it's just a pure domestic business. When you apply those filters, that's how you kind of reduce the opportunity. It's 15% of our auto, 15% to 20% of our auto revenue today, and maybe 15% of our funnel that we describe. It's highly filtered in those two ways. Yep. And i n terms of the core auto growing 15% to 25%, what is embedded in there in terms of the mixed shift, CV3, CV5, CV7? There would be no CV3 in that core growth, right? Because that's in the autonomy bucket. The core business would be products such as the CV2 family, which includes CV5 and also CV7 family. Those chips have an average selling price in this category anywhere from $10 to $75. Currently, our automotive ASP is close to the corporate average of $13, right? As we move the market to these newer chips that bring along with them, more efficiency and more advanced AI network capabilities, that's what should drive not only revenue growth, but the higher ASP in those markets. Perfect. Maybe moving to IoT and the Edge. We talked earlier about kind of some of the key end market drivers. I want to kind of narrow in on greater than 30% year-over-year growth last year. Can you kind of speak to your outlook this year? I think you guided mid- to- high teens growth. A re there changes to the underlying growth drivers within that outlook? Yeah. For Fiscal 2025, we just completed at the end of January, our revenues grew 26% for the year. The IoT business grew above that. The auto business grew, but grew less than that. You look at Fiscal 2026, which is basically calendar 2025, as you said, at the midpoint, we expect to grow 17%. Again, auto will grow, but less than the corporate average. IoT, maybe it grows around 20%. I think it is really important when you think about Ambarella and understand, why is Ambarella leading the Edge AI market and s hipped 30 million units? Nobody that I am aware of has shipped anywhere close to that many deep learning AI accelerators into the edge. The answer is that we had this video processing heritage, and then we integrated the AI accelerator. From an application point of view, you look at what Gartner says or these third-party research firms. What they say is that AI at the edge originated in the enterprise security camera market. That was the first edge market that began to put AI in the camera at the Edge and t hat was a market for us. We were early to see this. Now we've been shipping edge AI chips, albeit different generations. We're on our third now for five years. What's beginning to happen is that those similar chips and technology that we were so successful in enterprise security world are now finding homes in non-security applications, other IoT markets. That's why we're where we are today as having the leadership position. We had the incumbency in video processing. We developed a super efficient AI accelerator. We had a strong presence in enterprise. Now we're beginning to reach into all of these other markets. When you think about that growth outlook, and I know you talked about 17% for the company, presumably IoT is 20% plus something. How much of that is kind of units versus kind of pushing product up to CV5 and CV 7? Yeah. Yeah. Yeah. That's super important. Because with our chip now integrating the AI accelerator, which is becoming a larger and larger portion of our total chip surface area and dominating it in many cases, what has happened is that our selling price for our AI chips is much higher than it was for our legacy products. Even within AI, the newer generations are more expensive than the prior generations just because there's so much more utility in these advanced networks. For the last five years, our average selling price went from mid single digits to this last year we just finished, it was about $13. Our newest products, I think I mentioned earlier, the price is from $20 to $200. We've got the ASP ladder in there. We just need to execute with the design wins. That ASP story has been with us for some time. However, one of the challenges is that our units had been declining because some legacy business was going away. You had in the last four of the last five years, you had units declining, and you had ASP rising. Last year that we just concluded was the first year where you had ASP rising, units rising. We are expecting this year the same phenomena to happen. The reason that our unit growth is now being able to start to break out is that our legacy products is now only 30% of revenue. It is not like that anchor around our neck in terms of the unit growth. We are now in a situation where unit growth and ASP growth can both happen. Of course, that is a key to having premium revenue growth. Is there a framework for thinking about which end markets and/or geographies are driving that unit growth? I wouldn't think about it really, geographies. I would just think of it more IoT, where the time to revenue is quicker. Of course, some of the opportunities, especially on the autonomy side for auto, take much longer to get into revenue. Not surprisingly, physical security is one. Telematics has been very good for us. We're starting to see both EMIR's and driver monitoring beginning to grow. There is a bucket of what I'd call green shoots, where wearables or body-worn cameras are starting to use more and more AI and interested in our third-generation AI technology. Access control is a really interesting market for us. We announced ASSA ABLOY last quarter. If you're not familiar with them, they do for enterprises, ingress, egress from doors and into the office or into your Starbucks or coffee shop, the automatic doors. AI can do all kinds of interesting things in those markets. Portable video cameras are using AI not only to improve the image quality, but to do auto editing and things like gesture control. Everything has started for us in physical security. Now you have these green shoots in other markets. It is really difficult to say which one of those green shoots is going to grow into the big redwood tree and which one will not ever be that big. There are a number of irons in the fire there. Makes sense. Maybe going to the financial model and thinking about the top line guide. Y ou did highlight kind of worries around tariffs, I think more centered around auto. I f that were off the table, what kind of upside could you see? Yeah. We did bake into our guidance of 17% growth for this year at the midpoint, some implications from tariffs. Actually, we see it in both auto and IoT. It is with a very select group of customers. I think the farther east you go, farther east from California, the profile of that company potentially pulling in business from the second half of this year to the first half rises. We have asked most of these customers, is this happening? Are you building inventory? They all deny it. We have taken the precaution given some unseasonably strong orders that we have seen in the first half to just be more cautious with the second half. Even with that taken out of our guidance, we are still going to grow 17% this year. Your question was, did you give two sets of guidance what it would be? Of course, we did not give two sets of guidance. I can try to help you. What some investors have done is say, okay, you did $84 million of revenue in Q4. If I apply normal sequential seasonality, like take a five-year or ten-year average of our business to each of the four quarters this year, what revenue number would that yield? That would suggest, based on that type of analysis, maybe there is $12 million or so of conservatism baked into the midpoint of the guidance we provided. Makes sense. Maybe on gross margins, I think you got it 61.8% for the April quarter. How should we think about kind of the incremental gross margins over the next kind of one years, three years, five years, particularly as we kind of bring in the higher ASP solutions? Yeah. Our gross margin target range is 59% to 62%. If you go back five or six years ago, our gross margins were 58%. We were looking to move up into the range. At that time, five years or six years ago, 30%-40% of our revenue was coming from three customers in China. For geopolitical reasons, most of that business has gone away. It is almost zero now. When that happened, our gross margins went from 58% to 64%. All of a sudden, we were on the other side of our long-term gross margin range. What is happening now as non-China customers, we start to build some concentrations with them. It is causing our gross margins to come down into the range. We still believe the 59% to 62% is the right range to be in. Without that explanation, I can understand why it's a question. We think 60%-61% gross margins for our type of businesses at the Edge is still a very good figure to have. Got you. Makes sense. As customer concentration increases, that will kind of mitigate some of the ASP increase. D o you have a long-term kind of model, again, one, three, five basis in terms of how to think about operating margins over time? Just the long-term model we've put forth, it's been out there for some time, is high- teens, five-year revenue CAGR, gross margins 59%-62%. The non-GAAP operating margin target is 30%. Right now, we're around break-even in terms of operating margins. The key to getting that positive operating leverage is the revenue growth. That's our target is to drive control OPEX and drive premium revenue growth to get that positive operating leverage. Is there a notional value that you're thinking about where you could hit that 30%? I mean, at this point in time, there is. It always depends on how long does it take to get there. Yeah, we do not really go into that type of guidance. Perfect. I think we've run out of time. Thank you very much. Really appreciate it. Yeah. Thank you. Thanks so much. Yeah. Thank you. Yeah. Appreciate it.