Good to go? Okay, we'll go ahead and get started. Welcome, everybody. Thank you for joining us on the first day of Needham's 27th Annual Growth Conference. My name is Quinn Bolton, and I'm the semiconductor analyst for Needham. It's my pleasure to host this fireside chat with Ambarella. The company is a leading supplier of low-power, deep neural network AI vision processors for edge applications, including automobiles, security cameras, and robotics. Ambarella's low-power SoCs offer high-resolution video compression, advanced image and radar processing, and powerful deep neural network processing to enable intelligent perception, fusion, and planning. Joining me from the company is Louis Gerhardy, who is head of corporate development and investor relations, with a direct report into CEO Fermi Wang. John Young, CFO, will also be here in the audience, but the fireside chat will be myself and Louis. Louis, thank you for joining us. Yeah, thank you, Quinn. Since we may have some folks in the audience and on the webcast that are new to the company, maybe Louis could start us just with a high-level overview of Ambarella and sort of the company's products and product strategy. Yeah, thanks again. Nice to see you twice in a week, so CES as well, which I'm sure we'll talk about. But Ambarella has a 20-year history, and the first 15 years or so, it was focused on perception. What does that mean? It means collecting data through the lens of a camera for us, and doing so at the edge of the market, which is another very important distinguishing feature as we move on the discussion to AI. And so we're able to collect data very efficiently through a camera lens in difficult environmental conditions: too much light, not enough light, foggy, dust storm, whatever, and do so in applications that frequently had to run off of a battery, so very low-power type characteristics. And so that was our heritage, and we initially applied that competency to human-viewing applications. About six years ago, we started generating revenue from AI computer vision. What that meant is that we took that data collection engine we always did and integrated into it a proprietary AI inference processor. When those two things are combined into a single chip called system on a chip, you have the data collection engine and the AI inference processor, which enables machines to partially or fully autonomously make decisions. That's our story. What's happened now is that the AI computer vision is now 70% of our revenue. It is our growth driver. The legacy perception-only human-viewing business is 30% of revenue, and that's in a state of decline. Combined, we're delivering growth as now the AI part of the story is a majority of our revenue. What's happening within the AI business is, I mentioned it started out with computer vision, which uses an AI type of network called CNNs. Now that technology within the computer vision chip is being upgraded so that our newest products support transformer AI networks, which enables more advanced things such as vision language models, LLMs, and Gen AI in general. As we go through this transition, what it's meant for the company financially is that our average selling prices have been increasing. Five, six years ago, our blended ASP for the company was $6. Today, it's $12-$13. Our newer products, most of which aren't even generating revenue yet, started around $20, can go up to $75. In the case of auto, we have SoCs that can command up to $400. So very much AI at the edge is driving our revenue, and that's where we're at. Perfect. CES is always a big show for the company. Maybe start with kind of highlights from your perspective of CES. You introduced a new product. My favorite was your demos of the VLMs, where you can search a video stream with words. So find woman in red coat, and it would find it. But spend a minute just talking about CES and some of the highlights for the company at CES. Yeah, CES for us is a major investment. It primarily targets customers and partners. We probably had more than 150 customer partner meetings, and then on top of that, 20 investor groups, and really, this year was about showcasing new products. In fact, almost all of our demos came from products that are not generating revenue yet, so the SoC is real, and based on the different applications we serve, whether it's an IoT application like enterprise security or a consumer application like portable video cameras or an enterprise application like wearables or access control or an auto ADAS and autonomy, what we are showing across all of those applications is the ability for our SoCs to very efficiently support these new advanced AI networks I was referring to, the ability to efficiently do transformer processing and transformer AI networks. That was the essence of our messaging this year, highlighting to customers the wide variety of advanced AI features that we can support in these products, which command the above-average selling prices. Great. The automotive and IoT markets were pretty challenged for most companies in 2024. Ambarella was able to navigate this environment better than most of your peers. Sort of what do you attribute that better performance to last year? We do see the challenges that you hear many companies talking about: lackluster real global GDP, or obviously geopolitical issues, weakness in the auto EV market. But what's really stood out for us has been our new products. And again, coming back to the ones that carry the above-average selling price. And I wouldn't say every customer right now is using these new products, but given they command a higher average selling price, and we have a handful of customers in a variety of different vertical applications using them, that new product success has been enough to drive our premium revenue growth this year, and also to give us the confidence to be comfortable with where the consensus is, and to put some numbers around it. For our fiscal year that ends in two weeks, we have a January-ending quarter. Street has us growing 23% year over year. For the next year, basically calendar 2025, Street has us growing 16%. We're comfortable with that growth. It is being led by our new products that we see occurring in waves. We like to use the ocean analogies, but there are three waves of new product cycles that we're either experiencing or we expect to experience. Then there are, say, some swells in the ocean, things we've been investing in that we haven't precisely identified when they come to shore in the form of a wave or how large that wave might be. The three waves would be CV5, which is a new product cycle that's well underway now. Think of that as being roughly 20% of our revenue this year. CV5 commands an above-average selling price, and we're selling that SoC. For example, it's going into auto application with Rivian. CV5 is going to Motorola Solutions for enterprise security. CV5 is going to Insta360 for a portable consumer application, and CV5 is going to HP Poly for enterprise video conferencing, and I'm mentioning this because one SoC is able to serve a lot of different verticals given the programmability we have on top, so that's wave one. It's underway. We expect it to continue to grow next year. Wave two would be a product called CV7 family. There's multiple chips in that family, and this would be the first SoC family that is capable of supporting these advanced AI networks, meaning they need the transformer capability, and that revenue from the CV7 family actually starts this quarter for the first time, and we expect it to be a material contributor next year, and then the third wave is, I think, pretty well understood. This would be for CV3, which is a central domain controller for the ADAS and autonomy domain in vehicles. This is a high ASP product, sells for $100-$400 per chip, and revenue from this wave we expect to commence in calendar year 2026, and there's legitimate questions: how big will that be in 2027 and 2028? But some of the, beyond those first three waves, some of the swells in the ocean, if you will, where we haven't articulated when they come to shore, would be software stack development for like L2 plus to L4 cars. We've been investing in that for years. Radar perception software, which relates to an acquisition we did of a company called Oculii. The N1 processor, which will enable us to move from the edge endpoints into the edge infrastructure. Then a custom ASIC, semi-custom ASIC business model, similar to the type of model you see Broadcom and Marvell implement, but we'd apply it based on our core competency with AI acceleration and our perception data collection capability. Those are four additional waves that hopefully come to shore, and we can talk about in the not-too-distant future. Perfect. You talked about the growing blended ASP over the last several years as these waves kick in. But unit growth, I think the last couple of years has been declining. Do you see in sort of your outlook for fiscal 2026, are we starting to see units stabilize to potentially even starting to grow next year so that you have both unit growth and ASP growth heading forward for the business? Yeah. Yeah, that's exactly right. We've been posting increases in our average selling price given the influence of the more expensive AI processors, but our units had been declining, and that had offset some of our growth and limited our overall revenue performance. And the reason for that was that the legacy business, which was just the pure human-viewing perception business, is in a state of decline. And during the inventory correction the industry experienced, it faced a significant decline where calendar 2023, which is our fiscal 2024, basically declined 50% that year and represented a majority of the pain we experienced in the inventory correction. We're now at a point where those legacy products, the human-viewing products, are only 30% of revenue. And now that we're through the inventory correction, the rate of decline there, think of it maybe negative 10%, negative 20% CAGR. It's more predictable, and it's a smaller number. Therefore, the unit growth we've been experiencing in the AI business should begin to dominate the story and allow our units to begin to grow while the ASP should continue to increase. And rising units and rising ASP is hopefully a growth nirvana for us. Perfect. Before we get to some questions on the automotive business, we're asking sort of all of our companies to the extent the Trump administration increases tariffs on Chinese imports, would there be any direct impact on your business, or do you mostly export chips to China that then sort of get consumed and re-imported into the U.S.? So there'd really only be an indirect. Yeah, it would be indirect, and it's difficult to measure, but our customers in some cases may be subject to that. We have seen a lot of most chip companies, their ship to China is a lot greater than what is actually consumed in China because so much is exported, as you said. But we have seen more and more manufacturing move to Southeast Asia out of China. For example, Thailand, India, Vietnam are picking up manufacturing business, or even Mexico. So still, though, the built in China but export, it is a big figure. And so therefore, some of our customers may be subject to it. It's indirect. And we've tried to accommodate this risk and many other geopolitical risks by being conservative with our guidance and managing things very carefully. Perfect. Turning to the automotive business, the company provides its automotive funnel once a year. In your update from last November, you announced the six-year automotive funnel decreased by 8% from $2.4 billion the year prior to $2.2 billion this year. Can you talk about some of the puts and takes you saw in the funnel? What sort of moving out, what came in? Yeah. Yeah, so our funnel is unique in that in several regards. One, it's only six years, so it's not an indefinite period of time. And the second factor is we probability weight it. So what does that mean? Say a project lifetime value might be $500 million. If we think we have a 33% chance of winning it, we only put in $155 million for it. So those are two differences in our funnel. But what caused our automotive funnel to decline from $2.4-$2.2 for the next six years? And at the highest level, it's two things. One is probably not surprising to many is that overall forecast, volume forecast from our customers for the next six years decreased from the prior year. The auto industry is experiencing some weakness, particularly with EVs. And so those dynamics manifested themselves in the lower volume forecast from our customers. That was factor one. Factor two is that L2+, in particular the autonomy side of our automotive story, the penetration of L2+ into the number of cars produced globally has been below our expectations as well as many other companies' expectations. We think there's two reasons for that. One, there's been challenges for OEMs and some Tier 1s in terms of getting that complex L2+ software stack working. The second dynamic that caused push-outs and delay was that the price points our customers are telling us for L2+ need to be much lower than where, say, some OEMs were thinking a couple of years ago, which caused plans to be reset, which led to some program delays and cancellations. Given our six-year funnel, if a program start is pushed out one or two years, then the one or two years that fall out of year six in our funnel, we have to subtract out of our funnel. Those were the two dynamics: lower volume forecast, and then the challenges with L2+ penetration were the two factors that caused our automotive funnel to decline. We'll update that again in November of this year. Can you talk about sort of the price points the OEMs are now looking for for Level 2+ and how you're responding with members of your CV3 family to address those lower price points? I think at CES we talked about competitor Mobileye sort of introducing very recently their surround ADAS solution, also sort of reflecting this move in the market to sort of lower cost Level 2+ type solutions. Right. So yes, the premium L2 plus market, which think of like the Tesla FSD and where that is today, we can support that market with a product called CV3-655. However, as Quinn referenced, we see a lot of activity kind of in the price tier below that, where from a wholesale basis, the hardware, software, and the entire ADCU, maybe the target price point is $1,000. We've always had on our roadmap, it hasn't changed, a part number called CV3-8635, which we think is very well suited for that market segment. So we've always had this in our roadmap. It's an integral part of it. And we do have 655 for the premium market, but you're correct, Mobileye's solution for the 8635, they would call surround ADAS. Got it. And you talked about sort of delays in penetration of L2 plus. What would your current expectation be now for, say, L2 plus penetration by 2030 or some year out in the future if you've got a specific year that you guys cite? I'll tell you what they were and also what they are now. But if you went, excuse me, went back to our presentations two, three years ago, probably would have expected L2+ to be in 15%, maybe more of the new cars produced this year. So say this year 90 million cars are produced globally, including trucks, then that would mean 13.5 million cars would have L2+. The figure as we finish the current year, maybe it's around 5%. And half of that belongs to Tesla, and most of the other half belongs to Chinese OEMs, which are attacking us from lower price points. So that number, that the whole thing has just been delayed and pushed out several years, hopefully not more. Yeah. You just mentioned China. That was my next question. Kind of how do you see the China market for Level 2+ and how is the company positioned with some of the leading Chinese OEMs? China market, many different vectors, as you can imagine, and they all move super fast. You have the autonomy market, and you've all seen the Pony.ai and WeRide for ultimate level of autonomy L4. And then there's the mainstream market. Think of that as like L2 plus as we were talking about. And even within L2 plus, does the vehicle support urban L2 plus or just highway L2 plus? And that requires, even within L2 plus, that requires different chips. And then you overlay on that the need for domestic silicon, the requirement in some cases to use domestic silicon. And then overlay on that, the growing export opportunity in China. So where do we fit in all this? Where do we come out? Well, first of all, all of our business today in China is ADAS, so it's not related to this autonomy domain. We're in e-mirrors, we're in drive recorders and black boxes, we're in driver monitoring and occupant monitoring systems. And that's good business, and that's 15%-20%, five-year revenue CAGR type story. But for the CV3 business, which is the central domain controller for L2+ and higher levels of autonomy, we think we're well positioned, but if you apply a couple of filters, you can very quickly get to what's important to us there. And those filters would be, do we have a technology advantage over the local suppliers? And the second filter is, is the OEM in China focused on the export market, which would give us an advantage over the local silicon suppliers? I think an example of what I'm talking about would be this last quarter. We announced our first commercial program with Aptiv, which is a global tier one, but it was for Smart in China, which is an OEM that many of you may know was an auto OEM that came out of Swatch in Europe, but it's now 50/50 Geely and Mercedes-Benz ownership, but half of their cars are exported. And so the fact that they already have such a significant export business and they needed the advanced technology was a factor that likely contributed to our success in that program versus, say, a domestic silicon supplier. Coming back to sort of the level two plus CV3 opportunity, I think in the past you've talked about courting a number of large Western OEMs, that their level two plus programs could be in the range of $500-$1 billion over that pipeline period or over a five- to six-year period. Is that still the opportunity you see for some of those high-volume programs, or has this moved to sort of lower cost solutions, maybe trimmed expectations for what dollar value might win in a large L2 plus program? Yeah, no, those figures are still valid even for the lower cost markets. And hopefully in the next year we'll have an update, but for sure when we reduced our funnel, part of it was delays in terms of decision-making for some of those programs. And I guess there's always a risk that they get delayed again, but we're hoping to have more of an update in the next year, maybe next six months in terms of some of these programs. But I think you're right, and what you implied is that if there is one of these big opportunities that lands, it would most likely be for the CV3-8635, which is the lower cost L2 plus type solution. Got it. Last question for me on automotive. Ambarella has always had a price performance or performance per watt leadership on the hardware side of things, but you're going up against competitors with less efficient hardware, NVIDIA, Qualcomm that have pretty strong software solutions. Talk about your software solution stack, how you're going to market in partnership with partners like Conti and Bosch to provide software as part of your solution over time. Yeah. For Ambarella overall in all of our markets, IoT and auto, we've never provided the application software. Our customers have done that. They've wanted to do it. It's a way for them to capture more value, but maybe even more, most importantly, differentiate their solutions relative to their competitors. At L2 plus to L4, we did acquire in 2015 a software stack company doing L4 and robotics. Our approach is very different relative to, say, Mobileye in that we're an open platform, meaning if a customer is interested in using our full stack, that's great. If they don't want to use any of our software, we'll provide them with a software development kit to access all these resources on our silicon, and they can do the full stack themselves. Or somewhere in between, where if a customer wants to use certain IP modules in our software stack, we'll license just those IP modules, and we'll help them integrate it into their own stack. And so where we've started our first commercial project is a software stack relationship with Continental, which is the top 10 automotive tier ones of the world. And we've announced a commercial trucking platform where Continental and Ambarella, in addition to us providing our CV3-8685, our high-end chip, we're also collaborating on the software stack for the fallback blade in an L4 commercial truck. And so that's an example of us providing some of our software stack IP into a partner who can then sell to multiple commercial truck OEMs globally. But like right next to it at CES, we had a demo of Kodiak Robotics. In fact, they had their semi truck in our parking lot. And in that case, they're doing the full software stack themselves on CV3. Next to them, we had a third-party software stack company called Plus.ai, which is talking to OEMs in Europe, Japan, U.S. about winning business on CV3, but they do all of the software. Right next to that was a Chinese Tier 1 called Hyperview that had just won a major proof of concept project for a commercial trucking company, a Western commercial trucking company. But again, in this case, all of the software came from Hyperview using CV3. And right next to that, you had the Smart Aptiv demo where, again, it's using Aptiv's software on the CV2 chip. So it's not just talk, it's like customers can use a variety of different software stack IP modules from us, or they can use their own software stack. And our software development kit, this is really important, is so robust, it allows these customers to very efficiently access our chip and create their own software. And that's really when you start talking about these complex domain controllers, having that compiler and software tool that not only is super efficient and proven and constantly being upgraded, but that can accommodate these new networks. It's really super important for the overall value proposition. Wanted to spend a few minutes on the IoT side of the business, the security camera business. Certainly had some ups and downs last year, but now seems to be on a better growth footing. Can you talk about some of the, what's your outlook for the security camera side of the business? What are some of the biggest drivers in that side of the business? Yeah. So if you read the Gartner reports on AI at the edge and endpoints, like the first market to really demand this edge AI was enterprise-class security. Because think of it, these companies like Motorola Solutions or Bosch or Canon, Axis, Panasonic, they have these cameras, these passive sensors that are collecting massive amounts of data. And to do video analytics, it's much more efficient if you can do it in the camera versus sending all the data into a server. And so our core market historically has been in this enterprise security market. So that as AI first came to the edge in the enterprise security market, we incubated all of this technology there, and we continued to introduce our new products like CV7 and the transformer network capable chips into that market. That is a growth market where penetration of AI into the installed base is still maybe 15% or so. There's more growth there. Importantly, what we're starting to see now is that from this ground zero for AI at the edge, now we're starting to see non-physical security camera applications begin to adopt AI. At CES, those demos include access control, it includes wearables, it includes moving into the edge infrastructure with products like the N1, it includes portable video cameras, which surprisingly, not all customers, but some have a huge appetite for AI in these portable devices that operate off a battery. Enterprise video conferencing is another market. It's all incubated at the edge in this market you asked about enterprise security. I think it would be very difficult to have a big success in the edge if you didn't start there. Yeah. I wanted to move to a couple of financial questions. You've talked about your two-nanometer development program that's partially being funded by a lead customer. Could you give us the timeline sort of for your two-nanometer development? Is that NRE amount meaningful, or does it just help defray some of those development costs? And then is the lead customer sort of an IoT application or an auto application? Yeah, so you think about this custom ASIC or semi-custom ASIC business model. It provides access to our underlying technology development, whether you're talking about the perception engine or the AI accelerator in particular, which is increasingly becoming a larger and larger part of our value proposition. Such a business model would enable customers to maybe design something slightly different, something maybe proprietary where they might have special rights to. We have announced we have our first customer in this area, but we have not yet articulated it as a wave and said when it approaches the shore and generates revenue. What we've said about it is that we would expect to tape out the product, the two-nanometer chip, at the end of this year, and then it would need to be in the fab for six or seven months. We'd get it back, and if it's working well, then we'd be able to start to generate revenue after another turn in the fab. So you're looking at calendar 2027 maybe for first revenue. So as you can tell, we have a lot of execution we have to do along the way. And as the business case develops, as we execute further, then we'll be able to articulate more with more precision about the timing, and maybe we'll call it wave four at some point. Got it. Got it. And just thinking about OpEx, as you look in the calendar 2025, what are you seeing in terms of wage inflation, merit, variable comp? How does that sort of factor into your outlook for expenses this year or fiscal 2026? Overall, what we're trying to do is really focus on getting return on the significant investment we've made. Our R&D is way above what the typical semiconductor company spends. And so we feel we made a bulk of that investment. There's still some OpEx growth, some R&D growth to occur, but for the year that's going to end in two weeks, consensus has us growing OpEx 8%. It seems reasonable. That's with 23% revenue growth. And then for this next year, consensus has OpEx growing another 8% to about $211 million. So yeah, that's another 8%. And that's with consensus revenue growth of about 16%. So super focused in controlling that number, the OpEx and R&D is expected to grow. But things like two-nanometer that can be very expensive, that's already in the run rate OpEx guidance. Right, so you amortize the development cost. Yeah. We're a couple of quarters into doing that. Yep. I've got one more question just about cash priority. What do you focus? What are the highest priorities for use of cash? And then we'll have a couple of minutes for the audience Q&A. In the past, what we've spent it on is it's really remarkable. Fermi and team since the IPO in 2012, I think they raised $30-$40 million. I think you were involved, actually, one of the few. We never did a secondary. We never issued debt, no convertibles. All of this huge premium R&D investment has been completely funded by our operations. We've repurchased around several hundred million of stock in the past, not recently. We've done the acquisition of Oculii for $300 million of cash. All of those things have been done from the cash from operations. Currently, last quarter, I think we're around $220 million of net cash, of course, no financial debt. We're comfortable with that. Fermi's got a record. This is important. You can tell this is important to him. Even though we've had this huge investment and we've been through so many cycles in the last 15 years, he's had positive free cash flow for 15 consecutive years, despite all of the things that have happened. Right? This year was pretty close. And we'll see how it plays out. But our quarter ends in a couple of weeks. And you can imagine his goal is to have a 16th year. 16th year. Yeah. Got it. I think we've got a minute or two for questions from the audience. If anyone has a question. I have one for the staff. Hi. so I'm just wondering, how, if at all, does that change the complexion of the market? How has it changed the complexion, and how might it change the complexion? Yeah, that's a good question. The question was about Chinese OEMs or even Tier 1s that develop their own chips that compete with us. When I was describing China, I mentioned all these different vectors and dynamics, and they're all moving fast. That's one I forgot to mention. So thank you for bringing it up. But I'd say there's five, maybe six Chinese OEMs that are trying to develop their own SoC that would compete with our CV3. And until they understand, until they get the silicon back from the fab and ascertain their performance and which cars they can and can't put their silicon into, they're not going to consider our CV3 as any sort of long-term solution. So we'll just have to see the results of when they get the silicon back. But there is a scenario where maybe their own silicon doesn't fit into as many models as maybe they anticipated, which would then create opportunities for us or for some of the merchant silicon providers in China. But yeah, it is an important development to watch because until they get the silicon back, they're not really going to want to have serious discussions with us. And we also would want to be careful with them because if we're telling them all of our secrets in our chip, then they might just try to put it in their product. So dialogues with some of those players have been limited to date, but we'll wait and see what their own chip results are like, and then maybe it becomes an opportunity. Okay. Well, it looks like we're out of time. Louis, thank you very much for joining us at the Needham conference. Thank you. Yep. Appreciate it. Thank you.