Ambarella, Inc.

Thank you for standing by and welcome to Umbrella's second quarter fiscal year 2024 earnings conference call. At this time, all participants are in listen-only mode. After the speaker's presentation, there will be a question and answer session. We ask that you please limit yourself to one question and one follow-up. You may get back in the queue as time allows. To remove yourself from the queue, simply press star 11 again. As a reminder, today's program is being recorded. And now I'd like to introduce your host for today's program, Lewis Gerhardy, Vice President, Corporate Development. Please go ahead.

Thank you, Jonathan. Good afternoon, and thank you for joining our second quarter fiscal year 2024 financial results conference call. On the call with me today is Dr. Fermi Wong, President and CEO, and Brian White, CFO. The primary purpose of today's call is to provide you with information regarding the results for our second quarter, fiscal year 2024. The discussion today and the responses to your questions will contain forward-looking statements regarding our projected financial results, financial prospects, market growth, and demand for our solutions, among other things. These statements are based on our currently available information and subject to risk Uncertainties and assumptions, should any of these risks or uncertainties materialize, or should our assumptions prove to be incorrect, our actual results could differ materially from these forward-looking statements. We're under no obligation to update these statements. These risks, uncertainties, and assumptions, as well as other information on potential risk factors that could affect our financial results, are more fully described in the documents we file with the SEC. Access to our second quarter fiscal 2024 results press release, transcripts, historical results, SEC filings, and a replay of today's call can be found on the investor relations page of our website. The content of today's call as well as the materials posted on our websites are Ambarella's property and cannot be reproduced or transcribed without our prior written consent. Fermi will now provide a business update for the quarter. Brian will review the financial results and outlook, and then we'll be available for your questions.

Thank you, Louis, and good afternoon, everyone. Our fiscal Q2 revenue was approximately flat sequentially and consistent with our guidance. Our AI business goes sequentially and year over year, while our video processor business was down sequentially and down about 50% from a year ago. Our blended ASP in Q2 revenue was above $12 and is on track to grow about 20% year-over-year, thanks to the richer mix of AI SoCs, highlighting the value of our emerging AI inference business. Our mid- to long-term growth outlook for the AI inference process of business remains positive. However, the near-term environment is very challenging for our overall business. Customers are now more aggressively reducing their inventory. and we are now seeing some pockets of a weak in-market demand, which complicates our customers' ongoing inventory reduction efforts. Given this, we have reduced our second half outlook. We are not expecting a recovery in calendar 2023, but we do anticipate our customers' inventory will normalize by the end of the year and set us up for return to growth in calendar 2024. We continue to expand our position in the rapidly evolving AI inference processor market. Cumulatively, we have shipped more than 17 million AI inference processors into device endpoints for IoT and automotive applications. And we are now expanding our AI inference processor reach into vehicle autonomy. As announced on the last earning call, we continue to evaluate the AI inference accelerator market opportunity. I will now summarize the status of our three major SOC product families, video processors, CV2, and the CV3. First, video processors are human viewing are expected to be about 40% of total revenue this year, down from 55% last year, and they typically command a single digit ASP. For several years, we have been prioritizing our limited resource on AI technology and products. And for this reason, we anticipate our video processor revenue to continue to contract. However, the revenue impact from the video processor contraction in fiscal year 2025 is anticipated to be significantly lower than what we are experiencing this year. Second, our CV2 family of SOCs established umbrellas in the AI-influenced market, and these SOCs are expected to approach 60% of our total revenue in fiscal 2024, up from 45% last year. This family of AI-influenced SOC commands an ASP close to $20 and serves computer vision applications for auto and IoT. CV2 remains an important growth market for umbrella in the mid to long term. Third, our CV3 family SOC first began to sample a year ago. Based on our third generation of AI inference technology, this SOC target more challenging AI inference workload, such as partial or complete mobile system autonomy. The CV3 family SOC range from $50 to more than $400 per SOC, and the autonomous driving software stack optimized to run on CV3 can add hundreds of dollars per unit of incremental software value. The AI inference processor embedding our CV3 SOC is a starting point for our evaluation of the Gen AI acceleration market. In the last quarter, we began to port Meta's LAMA2 to the CV3 AD high, and we expect to have chatbot demos available later this year. We will provide updates on our continuing evaluation and encourage to see generative AI opportunities emerging on both the server, and the device side of the market. I will now summarize representative customer activity in the quarter. Design activity in the enterprise security camera market remains robust at major customer worldwide. Motorola introduced its H5A multi-sensor camera based on our CVE-2 AI SoC. The camera offers up to 360 degree view utilizing to full image sensors with up to 32 megapixel resolution and AI analytics. Access, a unit of Canon, announced the 2 megapixel M4215 cameras and the 4K M4218 cameras, both based on our CV25 AI inference SOCs, targeting indoor surveillance applications. Japanese market leader iPro announced the expansion of its Rapid PTZ X series and the S series with 16 new models based on our CV25 and CV22 AI inference processors. DynaColor introduces Model Q next-generation multi-directional camera using our CV5 AI processor to support four 5-megapixel sensors. And in South Korea, Hanwha launched three new bi-spectrum AI cameras based on our CV2 AI SoC. These cameras provide 4K video and thermal view simultaneously for the rapid detection and the classification of vehicles or intruders. I will now talk about representative customer activity in the automotive market. In our May 30 early call, I mentioned the positive feedback we received at the Shanghai Auto Show for our CV72AQ AI inference processor, a derivative of the CV3 family of SOCs. During Q2, I visited Tier 1 in China, and I am pleased to report multiple Tier 1 wins for Level 2 Plus applications. We expect some of these Tier 1 projects to commence production in the second half of the calendar year 2025. We are pleased to announce our first CV5 win in a passenger vehicle. We expect this win to enter production in the next 12 months. In this application, the CV5 will support AI-influenced processing for multiple cameras. Additionally, in July, GAC Motor in China unveiled its Hyper-GT intelligence group, including an L2 plus ADAS intelligent driver assistance system based on our CV22AQ. And recently, the Chinese government passed a new policy allowing camera monitoring system, CMS, to replace conventional left and the right side mirrors. The policy also covers interior rear mirrors with the CMS-enabled models being legal beginning in July 2023. This CMS system represents a significant opportunity for umbrella the CV2 family of AI inference processors. During the quarter, BAIC, one of the largest automotive OEMs in China, began selling SUVs equipped with a CMS system based on our CV22AQ. In the automotive aftermarket, Toyota introduced its wireless backup camera system for trailers based on our H32AQ video processor. The camera will be an option for Toyota's model year 2024 Sequoia and Tundra trucks. Canopy, the startup resulting from Ford and ADT's 2022 joint venture, introduced its first product, the Canopy Pickup Cam. Based on our CV25 AI-influenced processor, the camera provides a full HD recording, 180-degree field of view, person detection, and the reaching detection for the back of a pickup. And in June, action camera maker Insta360 announced its Go3 camera, a lightweight but powerful 2K camera that utilizes our H22 video processor. This representative engagement indicates a healthy pace of continuing customer design activity for AI-influenced processors. Our investment strategy is aligned with the anticipated market demand for more sophisticated software-intensive AI-influenced applications. In the last three years, thanks to the CV2 family, we have demonstrated the ability to capture more value per win as customer demand migrate to AI from video processors. Looking forward, we believe our newer products, such as the CV5, CV72, and the CV3, are well positioned to support the increasingly sophisticated AI inference workloads our customers are anticipating. As this new product ramps and as we also capture more software value, we anticipate our blended ASP will continue to rise. While actively managing expenses through the current market turmoil, we will continue to drive our strategic R&D investment to fully realize the AI-influenced market opportunities we have discussed today. With that, Brian will now discuss the Q2 results and the outlook in more detail.

Thanks, Farmi. I'll review the financial highlights for the second quarter of fiscal year 2024 and provide a financial outlook for our third quarter ending October 31, 2023. I'll be discussing non-GAAP results and ask that you refer to today's press release for a detailed reconciliation of GAAP to non-GAAP results. For non-GAAP reporting, we have eliminated stock-based compensation expense and acquisition-related costs adjusted for the impact of taxes. For fiscal Q2, revenue was $62.1 million, in line with the midpoint of our prior guidance range, flat to the prior quarter and down 23% year-over-year. Sequentially, IOT revenue was up slightly while automotive revenue was down slightly. Non-GAAP gross margin for fiscal Q2 was 64.6% at the high end of our prior guidance range. Non-GAAP operating expense was $46 million, below our prior guidance range of $48 to $50 million, driven by continued expense management and the timing of spending between quarters. We remain on track to our internal product development milestones. Q2 net interest and other income was $700,000, and our non-GAAP tax provision was $800,000. We reported a non-GAAP net loss of $6 million, or $0.15 loss per diluted share, equal to the prior quarter. Now I'll turn to our balance sheet and cash flow. Fiscal Q2 cash and marketable securities decreased $10.9 million to $216.5 million. DSO was relatively flat at 45 days, while inventory declined from 151 to 147 days, down 6.5 million from the prior quarter. Cash used in operations was 6.8 million, and capital expenditures for tangible and intangible assets were 5.4 million. Free cash flow defined as cash from operations less capex was minus 20% of revenue for the quarter and positive 4% on a trailing 12-month basis. We had two logistics and ODM companies represent 10% or more of our revenue in Q2. WT Microelectronics, a fulfillment partner in Taiwan that ships to multiple customers in Asia, came in at 54% of revenue. Shikoni and ODM, who manufactures for multiple IoT customers, was 14% of revenue. I'll now discuss the outlook for the third quarter of fiscal year 2024. The near-term revenue outlook is challenging. Customer inventory management actions have accelerated, and pockets of end-demand softening have appeared. Considering these factors, we estimate that our fiscal Q3 revenue will decline to approximately $50 million, plus or minus 4%, driven by our IoT and market. At this time, we anticipate that this revenue range could continue into our fiscal Q4 and with sequential revenue growth resuming in our fiscal Q1. We expect fiscal Q3 non-GAAP gross margin to be in the range of 62 to 64%. We expect non-GAAP OpEx in the third quarter to be in the range of 46 to 49 million, with the increase compared to Q2 driven by higher R&D tied to new product development activities. We estimate net interest income to be approximately $1 million, our non-GAAP tax expense to be approximately $700,000, and our diluted share count to be approximately 40.1 million shares. Ambarella will be participating in Evercore Semiconductor Conference on September 6th, the Asia Investor Conference on September 12th, hosted by NASDAQ, the Morgan Stanley Bank of America Future Car Series on September 28th, and the Mobility Conference hosted by UBS on October 2nd. please contact us for more details. Thank you for joining our call today. And with that, I'll turn the call over to the operator for questions.

Certainly. And as a reminder, ladies and gentlemen, please limit yourselves to one question and one follow-up. One moment for our first question. And our first question comes from the line of Christopher Rowland from Seshquahanna. Your question, please.

Hey, guys, thanks for the question. So if you guys could highlight a little bit more the pockets of weaker end demand that you're seeing there. And then maybe you can put this in terms of under shipment by end market. Last quarter, you talked about, as an example, a customer that was under shipping by 30%. Has that undershipment changed at that customer and have new customers join that kind of level of undershipment as well? Thanks.

Right. So, first of all, in terms of the pockets of weak in market demand, I think there are two areas. One is on the, of course, in China, we generally think that the China market is weaker than the other market. But I think more specifically, there is our home IoT business, we do see some weakness. And we were expecting some demands for the year-end sales. But we don't see that. So we expect to see, we are expecting a weaker market in consumer IoT. Those are two areas that cause an allowance. In terms of the, you said that from the particular customer last quarter, we quote about the on the shipment. In fact, that customer come back with even further push-out demand, and so that our shipment to them will be even lower than the 30% we talked about last quarter. And I don't think there's people adding to the list, but it's just really a lot of customer already on the list, but we see a new wave of push-out, the cancellation, coming to the end of quarter and all the way going to August. That's where we talk about.

Understood. Perhaps also the revision for next quarter was pretty significant. Can you talk about what kind of backlog coverage you have going into each of these quarters, what kind of visibility you have, what kind of turns you typically need in each of these quarters? And then just kind of broader picture, I think you guys kind of thought maybe July would be the last of all of this. It now looks like the weakness is going to continue through January. I couldn't quite sense it. I don't know, was January going to be flat? Was it going to be up a little, down a little? Any of your thoughts there? And, you know, why, you know, do we have confidence that that is indeed going to be the bottom and we'll bounce from there? Thank you.

Hey, Chris, this is Brian. In terms of backlog coverage coming into any quarter for the company, we typically come in with the quarter in backlog and don't really rely on turns to make the forecasts that we provide. That remains the case. As we give guidance for Q3, Q3 is covered with backlog. As we look forward into Q4, you asked about whether that's flat or up or down. In the prepared remarks, we talked about an expectation that our range of revenue in fiscal Q4 will likely remain in a similar range as Q3. And again, that's based on additional clarity that we've obtained as we've moved through this cyclical correction. Certainly, we're much farther through that correction at this point. We've seen some new information. We've incorporated that into our current outlooks. We think we have greater confidence in our ability to you know, get our heads around how this is shaping up at this point. We think that there's an opportunity for sequential revenue growth as we move into our fiscal Q1 for 25. But that's what we see at this point in time.

Thanks, Brian.

Appreciate it. Thank you. One moment for our next question. And our next question. comes from the line of Quinn Bolton from Needham and Company. Your question, please.

Hey, guys. Thanks for taking my question. I guess first for me, you mentioned some of the opportunities for the CV72S in China, where I think you mentioned having secured several level 2 plus 8S winds at the ramp, I think you said, towards the end of 2025. But just wondering if you could give us any more color? What's the magnitude, potentially, of the lifetime revenue of those wins? Are they significant? Or are they sort of smaller projects? And then I've got to follow up. Thank you.

Thank you. Right. So I talk about the several design with a Chinese tier one for this level two plus application. And the application is very specific, which is smart ADAS plus parking for the Chinese market. And we believe this is going to be the next high-volume opportunity in China to replace the current single-camera ADAS market. I think the price is right and the feature is right, and we also have major OEMs looking at evaluating the products at this point. So I think we think this could be a... as big as current ADAS market given time. Of course, that when we're ramping up the revenue in early, in the second half of 2025, the volume will be just ramping up. But I think when that peak, I think it can be a significant volume for the Chinese market point of view.

Great. Thank you, Fermi. And then the second question, it sounds like you guys continue to make progress on the data center or enterprise AI inferencing application. I think mentioned you ported the MetaLava model to the CV3 chip. Wondering as you look out, you know, kind of what are the next steps to, you know, for that project? Where are you on the software and platform development? And are you still sort of thinking of this as, you know, probably an 18 to 24-month time to revenue? Is that the right timeframe to be thinking about potentially for revenue from this opportunity?

Yes. So first of all, we start porting LAMA2 as soon as it becomes available. And I think we made great progress on that. We are planning to demo this LAMA2 in a chatbot demo to our customer sometime in the coming quarter. And also, I think that we continue to believe that our current CV3 represents not only cost-effective and power-efficient, but also performance-wise is capable to compete with A100. So from that point of view, we continue to develop. But I think you hear one important task is continue to build out the software infrastructure to support customers. That's definitely another area we are ramping up the resource. We talk about we only carve out a resource from the current employee pool to support the effort. But when we move forward, when we start engaging customers, provide customer support, we probably need to increase HACOM. But that's the second phase after we start engaging customers with our chatbot demo. In terms of the revenue, last quarter we talked about this is more like a 24-month cycle. I continue to believe that's the case.

Perfect. Thank you, Fermi.

Thank you. One moment for our next question. And our next question comes from the line of Joe Moore from Morgan Stanley. Your question, please.

Great. Thank you. Can you talk about how you're thinking about spending? You know, your OpEx is at this point close to your revenue. You know, I understand revenues are at a temporary low point. Just how are you thinking about that sort of the balance between the importance of the revenue pipeline versus the sort of near-term cash burn?

So, Joe, I think while you can – I think you'll notice that we are – We are definitely trying to control expense. You can see that our Q2 OPEX come below the guidance. And that's the direction we're going to continue to look at where we can cut, where we can save. But, however, we still want to continue to invest on our strategy directions, namely a series 3 architecture as well as you know, for the OTO and IOT, IOT, OTO, and LLM. But those are the three big pieces of the investment area. But things fall below the line. We need to look at whether we have a resource to support it. So while we continue to managing carefully about expense, we do not want to sacrifice our strategy directions.

Great. Thank you. And then in terms of the video processing process, market that you talked about you know we've obviously recognized that there was going to be a replacement cycle from even from video processing to to computer vision uh but do you think you're losing share in the segments the sort of legacy markets um that that aren't moving to cv or is that part of why the numbers are challenged here i think like what i said in the consumer ip can majority of the product today is really focused on the low-end the close

cost-competitive solutions. And that's where we really don't spend a lot of money to invest on. As you know, our investment strategy is always focused on the areas where it can continue to demand higher AI performance. And for the consumer IoT side, that we haven't focused on low-end SoC. Rome have definitely heard us. I think that's where we have the biggest risk in terms of losing market share.

Thank you.

Thank you. One moment for our next question. And our next question comes from the line of Matt Ramsey from TD Cowen. Your question, please.

Thank you. Good afternoon, guys. Forgive me for this question for me, but maybe I read a little bit too much into sort of the tone of your earnings release and some of your commentary. With all the, I mean, Joe asked the question just now about your investments and how you're going to focus them. And what I'm trying to get at is maybe one step, maybe one level of distraction above that is like the focus of the company like changed a lot now. I mean, it seems like there's a maybe de-emphasis of your camera business and now a shift toward investment in hardware and probably more software to support inference inclusive of some data center applications. So I guess, should we be taking some of the tone here over the last tonight and maybe last quarter's call as well as a big shift in direction of where you and your team are focused? And I guess the second part of the question is, has there been a change at all in the focus of the company on the automotive end market? Changes in expectation of revenue, timing of revenue, investments, et cetera. Like I said, maybe you can, if I'm reading things that aren't there, certainly tell me, but I think it's an important question to address. Right.

First of all, I think we continue to commit to, you know, the IoT market. I think that this is, when you say camera, I think you meant IoT market. I think that's one area we need to continue to focus and continue to provide solutions. After all, we think that we have differentiated technology as well as a big customer base, and we need to continue, we will continue to provide the solutions to our customers. So I think that is one area I will never, you know, say we take our eyes off the ball. And given that, I think we want to continue to invest in auto, but when you say changing auto strategy, one thing that I will say what will happen in the last six months is we believe that the China market will give us earlier and shorter term revenue than the other market, we definitely move more focus our resource to Chinese market for the CV72 and CV3. And I think that's the area we believe can give us faster to the revenue. But that doesn't mean we don't focus on our US and Europe or other market that we can get our CV3 design in. But obviously, those designing will take a longer time to go to revenue. So I think it's really a focus on the short-term revenue versus longer-term opportunities. But I would say for LLM, one thing is it's become very clear, even our current market, like even security camera, when we talk to our professional security camera customers, they all start thinking about how LLM will impact their business, how to use LLM to integrate multiple cameras into the services for the service they provide to their customers. And also that automotive guys also start thinking bigger, bigger transformer model, which is the performance getting higher and leverage Huawei investment LLM model. So I really think that although that LLM start with you know, on the server side, which is definitely an interesting area for us. But just in the last three months, when we start talking to our existing customer, it becomes very clear that ALM is also on the roadmap for all our existing customers. So ALM becomes a roadmap for us. It's not just an opportunity we can, you know, choose to invest or not. I hope this has clarified your question.

No, thank you, Fermi, for all the color there. I appreciate it. I guess as my follow-up question, you mentioned there's going to be different phases of your new investment around inference and LLM. Have you guys thought about sizing some of those investment areas? I mean, what number of people or number of dollars that you're shifting internal resources? And then if it goes well, like what kind of, magnitude of investment are you guys considering given where the P and L is right now? Thanks.

Right now. So for the current phase, we talk about, we only carve out a team that we using our current resource, which is under our payroll already to do that. But obviously when we start ramping up, we probably need to duplicate a similar size of team to support the LLM. So we are talking about, you know, anywhere from 60, 80 people total to support, a limited engagement for the customer. Like, our goal is not going to all the possible customers, like we said last time. We need to prove the concept and with several customers that really want to have a second source for LLM on the server side. Also, several customers in our current customer base that can use LLM for their roadmap. So I think with the site, to limit the scope And we think that we can fund this activity with our P&Ls.

Got it. Jump back in the queue. Thank you very much. Thank you. One moment for our next question.

And our next question comes from the line of Ross Seymour from Deutsche Bank. Your question, please.

Hi, guys. Thanks for asking the question. I wanted to ask a near-term question, and then I'll have a follow-up. On the near-term side, I guess there's kind of two parts, so forgive me for that. But for the guide for the third quarter for me, Brian, you mentioned that I think you said the majority of the weakness would be in the IoT side, not auto. I just wanted to get some color on that. And then the second part of the first question here is going to be the fiscal year basis, I think Joe kind of asked it earlier, but it looks to me like your non-CV revenues are going to be down about 50% year over year. How much of that, if you can guess, do you think is share that's just gone and it's the low-end stuff you mentioned versus just the cyclical dynamic of inventory burn and some of that will snap back?

Right. So, first of all, I think that for the showtime side, we definitely believe that it's a –

I think the question is related to... Yeah, I think the first part of your question, Ross, was confirmation that the weakness that we see in revenue for fiscal Q3 is driven by the IoT side of the equation, and that is correct. We're seeing, obviously, some significant rebalancing of inventory and orders across our businesses. But in terms of what's driving this leg down in revenue between Q2 and Q3, it appears to essentially be all IoT at this point.

Right. On the second part is really whether that the video processor side, I think that the video processor side is definitely part of the weakness we talk about in the consumer IP can side. And so that really depends on the inventory situation, how fast they can rebound to it. But I don't think we can snap back to the original level. We believe that some of our video processor business will be replaced by our low-end CV chip, too. So I think that we talk about next year, our video processor business will continue to go down, but not in the same scales that we saw this year.

And then I guess as my follow-up, and forgive me, I guess there are kind of three here. If we look at the growth for fiscal year 25, you know, you said that the video process will still go down. What do you think are the key growth drivers in your CV business that we should look forward to, you know, either when design winds kick in and inventories burn, so cyclically or secularly? Just roughly speaking, what do you see as the biggest tailwind to offset that video process or headwinds?

So for the CV side, I definitely believe that both when inventory burns through as well as our new project like CV5 and CV72 production will help us to get tailwinds to boost our IoT business. But even our auto side, we believe that there's new design winds can help us to get more revenue growth for next year. So I think that... From the professional IP can side, it's become very clear that we think that our market is still – we hold our market share very well. And as soon as we get the inventory cleaned, we should be able to see a rebound on the current design wins. And plus, with the CV5 and CV72 design wins we talk about, that should help. Thank you.

Thank you. One moment for our next question. And our next question comes from the line of Tristan Guerra from Baird. Your question, please.

Hi, this is Todd for Tristan. Thanks for taking the questions. You touched on the IoT side of the business, but maybe on the auto side, have you seen any step-downs in order patterns from automotive ODMs as they take inventory control measures in the past few months? And do you expect any further order reductions before year-end in auto?

Right, so we did some small customer that's reducing their inventory, but it's not as bad as the IoT side. I think most of the time, I think our automotive customers continue to take the part of the plant. Although we see some weakness, but it's not, like I said, IoT is really the main problem we're dealing with right now.

Okay, great. And then for my follow-up, How different are the potential engagements for CV3 given the long-term nature and software platform cost of developing ADAS solutions versus the traditional segments that you play in, including data recorders?

Right. So obviously the CV3 design will take much longer than our Recoder in the past. Recoder in the past was like 18 months, even its auto grades, 18 months of design cycles. Here in outside China, we're talking about four years. That's the reason we decided to really focus on our CB72 and CB3 opportunity in China first, and so that we can address the time to market, time to revenue issue. So for that, we talk about CB72 most likely will have a revenue second half 2025 calendar year. And I even think CB3, We have a design wing with a tier one that we might generate revenue in fiscal, sorry, currently in 2026. And that is definitely shorter than any other areas that we're seeing.

Okay, great. Thanks for the questions.

Thank you. One moment for our next question. And our next question comes from the line of Torres Vonberg from Stifel. Your question, please.

Yes, thank you. Fermi, just back to sort of the resources and the opportunities and specifically thinking about CV3, the leverage of the software, because, you know, we're talking about OIDO, right, which is an edge device, but we're also talking about AI accelerators at the core. So just trying to understand how much portability you have with your current investments Um, so that, you know, you don't have to go through a completely new investment cycle, if you will.

Right. So we don't plan to have a brand new investment cycle with our current resource. We only build up our software stack in a, in a, in a way we are ready to demo a brand new software stack. That's a hundred percent AI based. Very soon. And that's what we talk about in the past, right? We talk about that. We, we, we are working with the county. our software stack, but in parallel, we're doing our own software development. With our current resource, with the tool acquisition, both Ocula as well as Palma VizLab, we are at a stage that's ready to demo our next generation of software stack. So from the development point of view, software side, I think we definitely not only prove that we have enough resource, but we're finishing the work to some extent. So now the issue is with LLM. Also, we funded our first phase of LLM investment based on our current resource and leveraging a lot of investment we put into CB3 software already. But for the second phase that we are waiting to see whether the result of first phase of engagement with customers, Even that ramping up is not going to be a lot more than what we have today. We definitely will be ready to talk to investors when we are talking about ramping up. From that point of view, you can think that even for the silicon side, for LLM, like I said, this is going to be our next phase of of a CV3 roadmap. So we have to build a next generation roadmap for the silicon side, and this will leverage 100% of existing VLSI team. So any new development or new investment cycle for LLM is a software team to support customer. I hope that clarify the question.

Yeah, no, absolutely. And I also wanted to follow up on, I think it was a previous question about, you know, LLM or AI accelerator milestones. So, you know, what kind of milestones as investors should we be looking at here? I obviously understand the timing part of it, but, you know, what are some of the more specific milestones that we should be keeping an eye on?

I think the first important milestone is that we demo our chatbot demos to our customers, and it will happen sometime coming quarter. So I think that's important because with the demo, it also shows the performance, how we compete compared to our competitors, and all those information will become open to our customers. I think that's probably the biggest near-term milestone. Very helpful. Thank you, Fermi. Thank you.

Thank you. One moment for our next question. And our next question. comes from the line of Kevin Cassidy from Rosenblatt. Your question, please.

Yeah, thanks for taking my question. My question is also along the lines of the software stack. With the CD72 in China, how much of the software stack you developed, is there a component of that? Can you sell that software stack too?

Yeah, so there are two. I think our strategy, software strategy inside China, outside China are different. Inside China, we are counting on our software partners because to really collect data and train the data in China is problematic for us. So we are counting on working with our Chinese software partners to deliver CV-72. Last quarter, we talked about we already identified multiple software partners, and they are porting aggressively their software to our CV-72 hardware platform. and ready to demo to OEMs in this quarter. So I think from the China side, we know exactly what to do with CV3. With outside China, with CV3, we talk about our collaboration with county, but more importantly, we want to demo our own software stack, which we don't plan to bundle 100%, but we definitely think this is an important software important technology that we can help our customer to leverage what we have developed. This software stack, we'll talk about more next time, but however, it's really, we think, one of the very few software stacks is 100% AI-based. And we can show the performance and the functions that are close to the BSD level. I think this is definitely one thing we need to talk about, not only technology, but also our business model when we are ready sometime in the near future.

Kevin, I'll just add, next week, September 5th to 8th, Continental will be demonstrating our joint software stack on CV3 at the IAA show. public demonstration if anyone's in the area to check it out.

Okay, that's great to know. We'll look for that. Maybe along those lines of demos, your work with Compete, is there any update of how many OEMs you're talking to and any progress at all?

We announced one design week last time, and this time we will continue to engage multiple OEMs. with potential collaboration with county, but also independent that we also talk to OEM directly. So we continue to have engagement, multiple engagement with OEMs at this point.

Okay, thank you.

Thank you. One moment for our next question. And our next question comes from the line of Suji Da Silva from Roth. Your question, please.

Hi, Fermi. Hi, Brian. So thanks for giving us the mix of revenue for the year roughly between the products, video, CV2, CV3. Can you give us a sense of what the mix is either currently or kind of when it normalizes between auto and IoT? Because it sounds like those two categories are having different trends right now. I think it would be helpful to understand kind of where each one sits in your revenue today and maybe in a year or so when things normalize.

Yeah, sure, Suji. So if you go back to last year, Automotive was about a quarter and IoT was about three quarters. Given the relative stability of automotive this year versus IoT, which has been much more volatile to the downside, that mix is looking more like 70% IoT and 30% automotive for the current year. Obviously, we said that the size of our SAM they're pursuing over a multi-year time period is much more levered to automotive where, you know, it would be the inverse of that relationship where long-term we would expect automotive to be up 70% and IoT about 30% as we move out, you know, several years in the future as we get traction with CB3 and some of the other automotive solutions.

Okay. Thanks, Brian. That's very helpful. And then just trying to reconcile, guys, the large pipeline number you've been giving the last several quarters versus the inventory correction here. Is there a timeframe in which some of that pipeline starts to convert and meaningfully contribute? I imagine that that process would be independent of the inventory perturbations that are happening right now. Just correct me if that's wrong, but that's not pushed out or pulled in in any way because of what's going on right now. Thanks.

So I think that you're talking about the funnel number we talked about. So for the very near term, for example, if you talk about the funnel for this year, Definitely there's some impact for the inventory, but in general, I don't think the current inventory correction should have any impact to the funnel because it's really based on design wins and also the probability and the volume of the design win. So we'll be ready to talk about this number in November this year.

Can you just give us an idea of what years those start to kind of come in, the elbow of those, like how many years away that is?

Well, our funnel, Suji, is six years, and Given the time it takes to land some of these winds, and particularly with CB3, you know, it was back-end loaded in the, definitely in the latter half of those six years. Okay, great. Thanks, Lou.

Thank you. One moment for our next question. And our next question comes from the line of Gary Mobley from Wells Fargo. Your question, please.

Hey, guys. Thanks for taking my question. If I'm not mistaken, there's about 20 customers that have really moved the needle for your overall business. Have you had an opportunity to review those top 20 customers and where they stand with respect to inventory balances, whether healthy or not, and to give us a sense of the irer of the 20% sequential revenue decline expected for the third quarter of How many customers are driving that down, or is it isolated in just one or two?

I think that in general for IoT, all of the top 20 customers are having inventory correction problems. But auto might not be as bad. Some of the auto customers have an inventory problem, but our top auto guys, some of them may not. So if you really separate the application looking at IoT, any of our top customers, all of them have inventory problems. Maybe at a different degree.

I'm sorry, go ahead.

I'm just saying that maybe they have different degrees of inventory, but all of them have some significant inventories.

Okay, thank you. As a follow-up, I wanted to ask about your relationships with Bosch and some other China tier one partners for CV3. Where do those stand relative to how Continental's moving along?

So I think in China, we focus on, you know, working with the Chinese tier ones as a first priority and also with Conti and Bosch. So I think that those are the priorities we have. And we definitely have a, continue to have some RFQ projects from OEMs to those tier ones. CV72, because this is really, the silicon is not ASO, but it's a system ASO solution. This is definitely a Chinese tier one play. For Bosch and Conti, we'll focus on the CV3 levels of a solution to Chinese market right now. And we definitely have multiple discussion with Bosch and Conti on this.

Thanks, Fermi. Thank you. This does conclude the question and answer session of today's program. I'd like to hand the program back to Dr. Fermi-Wong for any further remarks.

Thank you, everybody, for joining us today. I'm looking forward to talk to you soon. Thank you.

Thank you, ladies and gentlemen, for your participation in today's conference. This does conclude the program. You may now disconnect. Good day.