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

[ Welcome to ] the afternoon session. Glad you could join us. I'm Vivek Arya. I cover semiconductor, semi-cap equipment at BofA, and really delighted to have the team from Ambarella join us, Fermi Wang, the CEO; and Brian White, the CFO of Ambarella. And what I thought would be helpful, Fermi, is for a few people who might not be aware of Ambarella, right, and your products and your kind of technology moat, would be very useful to give us an overview before we get into the nitty-gritty of macro and short-term trends.

Absolutely. Thank you. Ambarella is a semiconductor company focused on, originally, video processing. In fact, that's how we started the business 18 years ago. And at the beginning, we're focused on to produce high-quality, low efficient -- low-power consumption silicon solution for capturing video. But throughout the years that we went public in 2012 and at that year, we knew that capturing video is great, but it's not enough. We need to add actual functions, particularly analyze video on the fly, basically, means supplying AI technology to those video data we collect. And that's when we started working on the computer vision technology. And throughout the years, in the last few years, we started ramping up our second-generation CV product lines, I would say, 2 years ago, and then you can start seeing our transition from the video-only solution to the computer vision solution in the last 2 years, that's the technology side. And the other important thing is at the IPO time, we we're focused on more consumer-like products. Now we are focusing on industrial and automotive. And I think that transition is basically done. And this year, I think majority of our revenue comes from this industry.

Got it. Absolutely. So for me, it's interesting that the semiconductor industry as a whole is sounding very strong about all the secular trends. But investors appear to be very concerned about the cycle. What part of your portfolio would you kind of classify as more cyclical? What part would you classify as more secular? And what are you kind of broadly seeing in terms of the demand trends right now?

Right. So I think the -- while there's a lot of huge up-cycle for the semiconductor a couple of years, but also I think that one of the biggest problems for all of us is the supply chain issues. And we -- in this quarter, we definitely talked about how this Shanghai lockdown impacted our business. But however, I think that the computer vision technology we developed for security -- for the IoT business as well as automotive, the impact, the momentum is still there, intact. And we believe that moving forward it is getting stronger because not only by -- that is not only our belief, but also supported by our design win activity, the multiple customer taking our CV product into production. Last year, computer vision was 25% of revenue. This year, it's going to be 45%. So I think that momentum is there, and we believe will continue.

Got it. So you mentioned China lockdowns, right, that have created headwinds, right, not just for Ambarella, but for the entire semiconductor industry. What's your kind of real time read on where we are? Because one day, we hear a story things that are opening up, next we hear a story cases are back. So what's your kind of real time read on...

We're just a few days from the lockdown. So everybody is expecting the impact from the Shanghai lockdown will recover, but the question is how fast. And our view is, at least by talking to our customers, we believe it's going to be a gradual recovery instead of sudden boom that everything go back to normal. I think that's -- just look at the logistics, look at how the labor need to go back to the factory. Some of the working force has come back to their original place, and now they need to come back to Shanghai. So I think there is -- it will take a period of time to get -- fully recover from that. But even with that problem resolved, I think that the kitting issue -- the shortage of supply like microcontroller for automotive, for Wi-Fi chip, PMIC chip shortage, that still persists and still there. I don't see that -- the situation won't get any better until the end of the year. So I think while we solved one problem, but I think the supply chain problem will persist.

Got it. So you mentioned, right, that on PMIC and Wi-Fi, there are supply issues. I'm curious, though, what is your visibility into your component inventory at your customers?

Right. So I think we have some visibility because we talk to customers. However, we don't have a -- to their number of units, right? So however, what we saw is we start talking about kitting issues starting, I would say, Q3 last year. And then we started talking about maybe for customers who cannot get other components, they might have taken some extra inventory from us waiting for the components they need. And the Shanghai lockdown basically really amplifying that problem. In fact, our forecast for Q2 dropped quickly starting April 1. That's exactly when the Shanghai lockdown started. And when people expect that the lockdown will take longer, you start seeing the revenue forecast decline. So I believe that, that really reflects the inventory problem. However, there's one more point I want to point out, that we allow our customers to push out their order probably easier than other semiconductor companies. In fact, we don't believe in that forcing customers to take extra inventory if they don't need it. It's a good practice. So in fact, starting April, we allowed a lot of customers pushing out their orders from Q2 to Q3, even Q4. And that's another reason I think that we probably saw this problem earlier than other companies.

Got it. So for me, your baseline plan, that assumes that there is a recovery from the lockdowns in the following quarter? Or does it persist till the end of the year?

Well, I think it's probably a little longer than just with total recovery next quarter. Remain to be seen, but I do believe that it will take more than a quarter to recover back to normal.

I see. And what have you seen -- so I understand that maybe the delivery of components side, et cetera, is obviously gated by what they can actually do with it. But what about all the design engagements, right? What is happening to them during the lockdown?

I think design activity continues. And in fact, that our customers continue to engage new design win with us. And I don't think that slowed that. However, there is a caveat I want to mention. In fact, we do see that because of the shortage of Wi-Fi chip some of our customers moved their resource to replace the component with other component for their old products. So they do take away some engineering resource, but however the momentum of new -- new design wins on the CV side, I think, is very strong, and I'm still confident that is going to continue to be a growth area -- a strong growth area for us.

Got it. The second thing I wanted to ask you at a high level before we get into the product side is, a quarter or so ago, there were concerns about availability of components, right, from your foundry on 14-nanometer wafers from Samsung. Since that time, we have seen some of Samsung's other foundry customers move to DSM. What's your level of confidence that Samsung will be able to support you in the 5-nanometer cycle?

Well, from a supply point of view, I don't think there's a problem because we knew that we're going to ramp up our first 5-nanometer chip in production second half of this year. And we already see the delivery plan. So I'm pretty confident about that. I think the problem that people are talking about is the yield problem, right? So the 5-nanometer yield problem. So I think, first of all, the way we deal with Samsung is the yield is not our problem. In fact, the yield is covered by Samsung. The second thing is, in fact, we have gone through many process node with Samsung in the past 15 years. And at the beginning of every process node, they have kind of different levels of yield promises like any other process, but however, after going through this process a few times with Samsung, I know -- and with the current evidence, I believe that 5-nanometer yield can be resolved. Hopefully, sometime next year that we don't have to worry about this anymore. So from that two point of view, I think I feel comfortable that 5-nanometer delivery should not be a problem for us.

So you're only getting good, you're not having to pay for, right, the entire wafer?

Exactly.

But the fact that if they face yield issues, right, conceptually, then doesn't it in one way or another reflect in the cost structure to you?

Yes. So if their yield is like 30%, yes, that's an obvious economic problem in mass production. But I think the current yield problem we're seeing is not as real as that. And we think that if any chip it can go -- at the end, go to 90%, I think -- or 95%, the yield, we are not there yet, but we are -- I think it's not -- we are not far away from that. So that's why I continue to see improvement on the yield side and hopefully, that we can totally resolve that problem with Samsung by next year.

I see. Does this motivate you to diversify foundry sources in the future? And how long is that process? Or is that not something that you are focused on at this point?

We are, I would say, focused, but we always evaluate our options. In fact, we talked to TSMC or Intel for the possible process node. The question is always whether price support, whether these 2 things can meet our requirements. So the next node, for example, we can now afford to have 2 suppliers in one process node. For example, for 5-nanometer, we're coming to Samsung and that's it, right? When we go to 4-nanometer or even 3-nanometer later, we need to make a decision whether that -- what's -- who's our best supplier. And that's a process we need to go through at that time. And it's going to be very expensive because when you're switching from one foundry to another foundry, it's not just dollar size, but also the -- your whole team needs to refocus on a process at a different SDK that you deal with. So it's a huge investment on both engineering capability as well as the resources.

I see. I'm curious, what do you think of Intel as a foundry?

We have talked to Intel. In fact, Intel is trying to be our foundry for a while, right? And we talked to them several years ago. We recently talked to them. I think that I believe that their technology, at least at the 5-nanometer side, they are not there yet, but we are waiting to see whether they can deliver a process node that can be -- that is equivalent or not equivalent just close to the -- what we are using with Samsung or maybe even better with Samsung, but we haven't seen that yet.

I see. And Fermi, just given in this day and age, it seems like supply assurance has become such a big deal for customers that it is not about how much money they want to spend, it's whether there will actually be supply, right, to support their -- their business needs. So you think in your discussion with customers, are they taking a deeper look on your supply chain and your ability to guarantee that amount of supply.

Yes, absolutely. In fact, particularly for automotive customers, right? They know that any project they engage with us, we need to commit 10-year supply, and they need to also talk to our foundry to make sure that they commit to that. So I think that for those customers, we are -- now we are engaging with the customer. But as soon -- before they come in to us, we need to have our foundry talk to our key auto customers for them to feel comfortable that the foundry is in the loop, and they are committed to support them, too. So I think that, yes, every customer, all the customers are very focused on our supply chain issue. So I think, particularly for auto, and that's an area we need to make sure our key customers feel comfortable with our choices.

Got it. So auto is obviously one of your most exciting growth areas. What is the right way to kind of position Ambarella in the competitive landscape? Because when we think about auto, we think a lot about the established players, NXP and TI and ADI and others. But on the compute side, Qualcomm is becoming bigger, NVIDIA, right? I mean, Mobileye, right? So where does Ambarella kind of fit into this competitive landscape?

Right. So it really depends on which product line you're talking about. In fact, if you look at our automotive application that we are trying to address with our solutions, there is a portion, which is like e-mails, recorders, which we compete with the TIs and the Renesas type of supplier. But when you move to ADAS and Level 2+ car, we are competing with NVIDIA and Qualcomm and the Mobileye. So I think that, however, it doesn't matter which -- it's really about different performance level that for different solutions. But I think the differentiation for us is always we try to deliver better performance and better video quality at a much lower power consumption than our competitor. And that's critically important not only for IoT customer, but also for automotive, especially now a lot of automotive opportunity moved to EVs. And so with that kind of requirement, your battery life and your power consumption is critically important, especially for the Level 2+ our power consumption can be 100 watts below. I'm talking about the sum, the total sum power can be 100 watt below our competitor, which is very meaningful in terms of not just cost, but also the battery life, right? So I think from that point of view, that's another way we differentiate ourselves. But we can go deeper in terms of other technology. For example, on the Level 2+ today, another important factor is how do you provide a scalable architecture that can address from the low end to the high end markets for your customers? And that is not only architecture, has to be software -- SDKs. And to offer that kind of scalability, you have to have a unique -- a very unique silicon architecture designed for this application particularly. And that's why you can -- at the high end, you can scale -- if you design a high-end chip, you can always scale down to a -- for the single camera or 2- to 3-camera solution. That scalability is not easy, and I think we are one of the few who can really do that.

Got it. So it's basically the performance at much lower power. That's the main. We have seen Qualcomm make a number of acquisitions, try to get a broader footprint. Mobileye, I think, is planning to go public, I imagine that at some point. When customers try to select their suppliers, do you think they look for best of breed like yourself? Or are they looking for companies -- again, given the supply dynamics of the industry, that they're looking for companies who can bring a breadth of product and are much bigger. Like does scale help your competitors more?

Yes, they do. But however, sometimes, quite auto, our customer, OEM, the Tier 1s, they all realize the biggest competitor for them is not traditional car companies. It's Tesla, right? If you cannot compete with Tesla then you have a problem. So technology matters. That's why I'm trying to tell a customer that, don't settle down with a bigger company, even they have a second best technology. It's really about why -- whether you can have a technology that's different -- that competes with your biggest competitor, Tesla. And we believe we have a unique offering there that we can help our customers to do that. So scale always matters in semiconductor. But however, in this particular case, performance matters more, in my opinion.

Right. What is so difficult about -- Qualcomm makes processors, right, for smartphones. I imagine they are very low power also. So what difficulty should they face in making processors for a car?

So for both NVIDIA and the Qualcomm, their approach is different than ours. Their approach is using the existing silicon architecture. NVIDIA using GPUs. In fact, if you look at their OEM chips, a bunch GPUs in there. And then if you look at Qualcomm is using processors, in fact, their -- the Snapdragon is there. We take a totally different approach. We look at this application and try to design unique silicon architecture to address this problem and this problem only. And by doing that, we can deliver a much smaller dye, much higher performance and much lower power consumption. That's how you can achieve that. And I think too for NVIDIA or Qualcomm to compete with us on that level, they can do it if they change their architecture. But that means they have to do a brand new silicon architecture and change their software structure. So I think that's a lot of change for them to do. So from that point of view, I think we are offering a really application-specific solution that is most efficient for the solution, and that is not just video only anymore. If you look at how we acquired a radar company recently called Oculii. The way we think we can integrate radar technology and offer a differentiation, I think, is very important because to get better performance on autonomous driving, the perception is one of the most important area. And we think that all we can do with Oculii is not only that Oculii provide a really outstanding algorithm, but we can do sensor fusion at a low data level, which very few people can do, and that will improve the accuracy and the performance. So that is something we can add to our differentiation level against our competitors.

Got it. Late last year, and I think even at your Analyst Day earlier this year, I believe you almost like triple the size of your pipeline. So what drove that -- such a large expansion. And the one place where I wanted to maybe push you a little bit is how much of that pipeline is based on what one might call, let's say, new age, for lack of a better phrase, younger, newer automotive companies who may be great in recognizing good technology but may be constrained just because they may not have access to batteries or other components, right? That is also a constrained business. So how -- what drove the expansion in the pipeline? And then how solid is that pipeline?

Right. So maybe we should go back to how we built out this pipeline first, right? We announced this first pipeline 2 years ago. At that time, we announced $600 million of pipeline. In the pipeline, there are 2 categories. One category is called, ONE, which is a business that got customarily committed to us. We know it's happening, and we're doing engineering work. The only question is how much volume they're going to deliver at the end by whole. So that's one area. Of the $600 million, that's $400 million. The second category we call it design-in activities. Basically, you have RFQ you're bidding on in there, you have to decide what's your chance to win and what's the potential of the volume that you have to. So we have a way to minimize the potential by having multiple -- multiplied -- multiplier to go in there to control the risk. So for example, there's an RFQ award stack at $1 billion for 6 years with a multiplier, at the end you probably only have $200 million in the design pipeline. So with this, that's how we step up the first year. And last year, we gave a lot of update on the pipeline, which is $1.8 billion. Out of that, $700 million is one business, and $1.1 billion is design-in activity. And to answer your second question is, in fact, if you look at the first $600 million, all of the ADAS and Level 2-related business in that $600 million are new company like Rivian, like Motional, the customer that we announced. And however, the $1.8 billion, we always start seeing them -- we add some of the traditional OEM Tier 1 that we have RFQ that we're bidding out and potentially have revenue in the year '26, '27. So that's the combination of how we do that. And the expansion from the $600 million to $1.8 billion, I think, is really about that we start bidding on the Level 2+ opportunities.

I see. And when you engage with these customers and you ask them for a similar level of forecast, what is the variability around units and pricing and so forth? Because you're giving a very specific estimate of this, "This is my pipeline." So I imagine there is a level of specificity around units and pricing.

Yes. Absolutely. In fact, we have to plug in the unit number they gave us and our price. But however, you will need all those to have a judgment, right? Because the judgment is, for example, there's a price attrition from year-to-year model. But more importantly is the volume that the customer gave you, what's your judgment on the -- at the end, what's the true volume you're going to get, right? So there are multiple levels of multiply we put in to control the risk, to minimize the risk of the model.

Have any of those customers given you any kind of prepayment or funding to ensure supply? Because I imagine that they will also want to have as much supply, back to supply assurance.

So most of the project we kick off were NREs, but that's more of our engineering support. From a supply point of view, what we did is we ask them to give us a 12 months running forecast and 6 months POs. And they have to commit to the 6 months PO. So that's basically what we're doing. And then with that, so far, we have been -- we haven't -- we don't have any delinquency onto our customer on the automotive side and as well as that they continue to get -- to meet their commitment on the 6 months of appeals.

Got it. And the last one there for me is that if you were to look at what you would have projected 3 years ago versus the actuals, what's been the kind of the report card on how well your identification of how this pipeline has kind of stood the test of time. I realize it's a relatively younger business so maybe you don't have as much statistics on it.

Well, in fact, that 3 years ago, we have not even started shipping our CV2 products, right? So maybe let me report on the CV2 report card, first. I think our CV2 revenue grew faster than we expected. In fact, I think 3 years ago, we talked about Wave 1, Wave 2 and Wave 3 revenue forecast. Wave 1 is for enterprise security camera, wave 2 is the consumer security camera, wave 3 is automotive, the 3 CV wave revenue. And we achieved the goal that we set out earlier than all 3 waves. That's for CV2. So if you ask me how we compare to 3 years ago, I have to say that we grew our CV revenue faster than we expected. And also we get more design win -- also we got more design wins on the CV side than we expected. So I really think that the major difference between those 3 years is at that time, we tell people we'll have a computer vision chip, people say, "Okay, show me the result." Today it's, "Great. You know, you proved to me CV2. Prove to me that CV3 will be everything as good as you told me." So I think that's where we are at right now.

I see. How sensitive are the trends to automotive production? Because that's another place where we have seen a lot of disruption, which has nothing to do with your product, just has to do with the dynamics of the auto industry.

Right. So you're talking about supply point of view?

That's right.

From a supply point of view, we don't have -- because we really focus on -- when we deal with automotive customer, we definitely have a long-term contract with them, and we have commitments. So we focus very much on the delivery to automotive customers. And I can say that we don't have any delinquent. Even through the last 2 years when there's such difficult supply problems, we don't have any problem with delivering our promise to our customer. In fact, the biggest problem for us is always that the microcontroller shortage that is well known in the industry really have continued to surprise to us that even some of our biggest customers cannot get enough supply to that, and that impact our revenue on the auto side. So I think from a supply point of view to our customer based on our chip, I don't think there's any surprises there. It's really about how fast we can really get past this -- the current supply -- semiconductor supply problem. And hopefully, after that, we can have more visibility and a more predictable outcome on our automotive revenues.

I see. Fermi, do you think that as parts of semiconductor industry demand slow down, PCs and phones, obviously, where you have no exposure. But as this slow down, do you think that frees up capacity on the trailing edge for what you do need that -- so that supply becomes less of an issue? Or it's too early to have that kind of wishful thinking?

You thought that with the PC and the cellphone slowdown, some of the capacity should free up. But if you talk to all of the foundry, including Samsung that we talked all the time, they didn't recognize that. They still tell us it's tight. Try to -- they tell us that they still have full capacity. So there -- I don't know how to close the loop between this two. I think sooner or later, we're going to face the reality that either we'll continue to have supply problem or we could have a slowdown like everybody thinks that if PC and the cell phone continue to be weak, I just don't see how these other demand can fill the foundry capacity. So from -- based on my conversation with our supplier, I don't think the situation recovered, but I have doubts on that.

Right. That has been surprising. Or is it just that maybe those things are not interchangeable as quickly, right? Maybe they are just on different nodes.

If that's the case, then in 6 months, we should see a major change because the data is less than 6 months right now, right? So if that's the case, in fact, that's really scary because that means, it to have seen major -- become -- switch from a supply problem to a demand problem on the foundry and that might not be good for semiconductor business also. For us that have been in the industry long enough, I still remember in 2000 when TSMC had this huge shortage when they reconnect on the strategic change, how much of that impact the semiconductor business, and I hate to see the same thing happen again.

Got it. You mentioned Oculii. A very, very interesting transaction. Maybe walk us through what drove you to that asset? How has the integration been so far? I think you've had like $3 million, $4 million of -- so it's still early in its life cycle. How do you see the progress there?

Well, the thought process, in fact, when start -- when decide that we need to do domain control like CV3, we look at, say, is there any other sensor modality we need to integrate into CV3? We know that video we have worked well. But is there any other sensor modality we need to do? And very quickly, we decided that we need to do radar because when we look at all the applications we're doing in addition to just automotive, even IoT space and robotic space, the second most powerful sensor is radar sensor. And on top of that, when we start looking at all the radar sensors out there 3 years ago, it became very clear that radar will provide similar performance and resolution and LiDAR. In fact, I think we are just a few years away to see for the image radar to deliver LiDAR performance. And based on that, we start looking at, okay, can we -- if that's true, how can we do to provide better differentiation or better offering to our customer. That's when we start doing AI simulation and quickly, we understand that today, all of the sensor fusion is happening on the [ objectless ] level. Video, radar, LiDAR generated, they're all objectless. And you do sensor fusion on that level. And without simulation, we believe that the drive level to do sensor fusion is on the raw data level. For us to achieve that, we have to acquire a radar technology that we can integrate to our chip so we can do a low-level sensor fusion. And that's when we start looking at acquiring a radar company. And the reason we acquired Oculii is they have a very similar approach to the problem. We talk about that our approach to the problem, we go to the algorithm first. We only try to solve video processing or computer vision problem. The first thing we do is try to understand the algorithm. And from after we understand algorithm, we try to find a way to optimize the performance and the power efficiency with that architecture, right? And surprisingly, we found out Oculii taking exactly the same approach. They are not a semiconductor company. They're a software algorithm company. Their algorithm really try to optimize the performance based on lower, much cheaper sensor, much lower number of antenna to achieve similar performance. I think that -- we appreciate that approach, and we believe that's the best match. And that's why we acquired Oculii at the time. And now 6 months past after acquisition. And I think Oculii is every bit that we expected, and we are integrating their software into our CV3. I believe that very soon, we will be able to demo that -- the power of doing low-level sensor fusion between a video radar, which very few people can do.

Got it. Now maybe just to kind of push you on the competitive landscape there. So an NXP or TI or Infineon, right, others who are in the radar market already. What is to prevent them from just bundling software like this with their product?

Well, I think the -- yes, they can. The problem is the most important differentiation is, we believe -- for example, we look at the mobile radar solution. It takes 1,000 antenna to achieve certain level of performance and accuracy. We believe we can achieve similar performance with 1 order of magnitude with less antenna and -- from sensors. That's a huge bond saving. On top of that, with that savings on the antenna that reduce the bandwidth requirement to take the sensor data signal into the main chip, and with that savings, we can integrate multiple radar into sensor with less bandwidth requirement. That really enable the sensor fusion I talk about because if we're to talk sensor fusion, you cannot just take in 1 radar sensor, you need to take all the radar sensors around the car, usually 5, 6 of them. And with the Oculii architecture, they give us advantage, not only to reduce the bond, but also enable a feature that's very hard to do from the system integration point of view. And that's why we think we have a unique opportunity to defend -- to compete with those guys. Of course, NXP and Infineon are very, very big on the radar side. But however, we believe that we don't focus on the -- each radar, we focus on the domain controller. When you do your Level 2+, you cannot have a domain controller only to video, the domain control should process all of the sensor data, including video, including radar. From that point of view, I think that architecturally, that we should have a domain controller to integrate all the video data and the radar data into 1 place, which is not the same architecture that ASP and all other people are trying to propose right now.

I see. Do you think having -- so you mentioned different modalities, so video and radar, to the most important modalities. Do you think you have now adequate coverage so that your domain controller is -- so basically, I should call it sensor fusion, that is enough, that it covers enough sensors? Or do you think you need to add to this portfolio to become more relevant to customers?

I think we will be able to demo CV3 with all of the sensor modality we talk about. And we don't -- of course, in the future, there will be new things coming up we will consider, but with the current generation, I think CV3 will be able to integrate all the sensor modality you need -- our customer needs.

Got it. Maybe if I could bring Brian into the conversation. So Brian, maybe on the gross margin side, I think Ambarella continues to surprise on the upside when it comes to gross margin, even the -- still well above your 59% to 62% target. Do you think as auto grows as a percentage of mix, you get back into target? Or there is a case to be made you're structurally going to be better than your target margins?

Yes. If you look at the gross margin trends of the company, it's been a very positive story with rising gross margins for the last few years from the high 50s to now 63.5% for last year, and we did 64% last quarter. So we've reached a point where we're well above that long-term model that was put out there, 59% to 62%. What that model comprehends and allows for is potentially going after very large revenue opportunities in the future. So it's less a function of change in mix than it is dramatic revenue growth and the realization that we'll be competing for very large opportunities. And so we need to contemplate that in that long-term model to provide for it. In the near term, I would expect our gross margins to stay in the same range that we're delivering today.

Got it. And is pricing playing a role? Because we have seen across the semiconductor industry that suppliers have been very successful in passing along rising foundry cost. So are you anticipating you will be able to continue to do that? Or do you think as there is enough foundry capacity available that customers will ask you to lower pricing anyway. Like, are these kind of pricing levels, are they going to be stickier over time?

Yes. So you've seen our gross margin be strong recently and through the supply-constrained environment. So we've been able to pass through any challenges from a cost perspective and preserve our margin. And it would be our intention to continue to do that in the future. So I don't see pressure on the gross margin associated with rising input costs that are beyond what we could pass through.

Got it. One of the trends that, Brian, we have seen is many companies that recently changed boundaries had to go and put up a lot of prepayments for capacity. Is that something you contemplate doing? And I asked that because historically, Ambarella has had a very high cash balance, right? And then when you bought Oculii we saw that come down. So it's very comfortable, right? So I'm not -- it's $200 million, it's very comfortable. But does it give you enough cushion to have to -- what if you have to prepay for capacity? Or you don't have to do that so it's not an issue?

Yes, it's an unknown as to whether we have to do that in the future. It hasn't been something that we've had to do in the past. So we'll evaluate that in the future. But we have more than adequate liquidity in the company. So we certainly are carrying excess cash today that could be used for that purpose, that could be used for other purposes as well.

Got it. And the last one there is M&A. For me, how do you think about Ambarella 3 to 5 years out, right? Is it executing on -- you have very interesting exciting growth opportunities on just an organic basis and then integrating Oculii right on top of it. Does that provide you enough in terms of IP to look out 5 years from now? Or do you think there are other pieces that could be required over time?

Well, the technology innovation happens every year. I think 5 years down the road, I can imagine, there are some new technology popping up, and we need to either grow organically or find a way to acquire. And my guess is with the innovation that happens around us, we probably need to look at acquisition more than just trying to do everything ourselves. So probably I would like to point out, we have done only 2 acquisitions throughout the company life. Oculii is radar processing. And when we acquired VisLab, which is a complete software stack for autonomous driving 7 years ago, a lot of people are scratching their heads and says, "Why a semiconductor company want to acquire a software stack?" But now you see Qualcomm acquired Arriver 5 years later paying like 50x more than we paid for VisLab. So I think what we need to do is really -- one thing we -- I think we -- our strength is because we focus on perception, focus on domain controller right now, we know -- we try to understand the -- try to predict what's needed for our future. We acquired VisLab 7 years ago to prepare for CV3 that will happen in the near future. We really grew that technology for 7 years. We grew in that 7 years, although I love our VisLab employee, but they contribute 0 to the revenue for now. But -- we are very happy to do it because if today, we want to acquire VisLab type of software, it will cost us the amount of money that we cannot afford them probably. So what we are trying to say is we need to continue to look at what's our long-term strategy in terms of technology and business and identify the void that we need to fill, and that's usually how we come to the target for acquisition. And that applied to VisLab, also applied to Oculii. But also, another thing I want to point out, if the current market situation continues for another 6, 12 months, I would like to believe that there will be many targets that are available at a more reasonable price to us. We shall opportunistically continue to look for a match for us, and that's something that Brian and I will continue to watch.

Got it. It's interesting, one of your other competitors on the GPU side, they have spoken about actually getting into a recurring revenue and software and subscription relationship with Mercedes. Do you ever foresee Ambarella entertaining something like that? Or will you have to really build up your capability to ever think about actually being able to charge separately for your software?

In fact, we have done the charging -- we have charged software. In fact, one of the reasons we can have 60% gross margin is we charge the software. But the way we charge is, we -- for each chip, we create 20 different part numbers, each part number associated with the software. So that's how we charge software in the last 18 years. But that business model probably need to be upgraded because suddenly -- in the past, there's a video software you need to upgrade, you have a radar software you need to upgrade. You have other software you can upgrade. So we have to find a way to generate software revenue in the future, but reoccurring software is a total different business. I think many semiconductor tried before. None of them been successful. I won't -- I have -- I can bet that Ambarella is not going to be the first one to make it successful. The hardware -- software revenue definitely is something that we need to pay attention to.

Right. And just maybe the last one. So clearly, the company is going through kind of the lower part of the growth cycle with all the supply constraints that you mentioned. When should we expect the company to start getting back to year-on-year growth? Is it something that can happen this year? Do we have to wait for next year? Because obviously, compares are also tougher from prior period. So when should we start to expect a reacceleration to year-on-year growth?

In fact, this year, even with today's number, we still grew from last year, not as big as what we hoped for at the beginning of the year, but it still grew. But with our momentum of CV and momentum that we built on automotive, we believe that we will continue to see growth, especially I think the biggest growth is not -- is from the ASP growth, right? We talk about our video processor to -- CV2 processor, our ASP basically doubled. So with that momentum on the ASP side that we can continue to drive our growth. And where we have unit growth associated with that, that will be even better for us moving forward. So for example, we also talked about CV3 ASP will be much higher than CV2, and that should continue to drive our growth in Ambarella.

Right. Terrific. Thank you so much, Fermi. Thank you, Brian. I really appreciate it.

Thank you, Vivek. Thank you very much.