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

All right. Thank you, everybody. And I'll quickly read the safe harbor, which I know you've heard now 75 times. Please note that all important disclosures, including personal holdings disclosures and Morgan Stanley disclosures, appear on the Morgan Stanley public website or at the registration desk. So I'm very happy to have with us today the CEO of Ambarella, Fermi Wang. The company just reported last night. This is our top small cap idea in semis. So Fermi, thanks for being here, and I know you had to pull up your earnings date to get here. So I appreciate that. Maybe if we could just start a little bit with the history of this computer vision effort. I feel like people -- there's kind of a view that your consumer business rolled off and you needed something to fill the gap and so you started investing in this. This is something you've been talking about since before the IPO. Obviously, it takes a long time to turn to revenue, but you've really gotten to that point now where it's starting to contribute. You've got a dozen design wins every quarter. So maybe just kind of walk us through the -- you were originally a video processing company. You're starting to add analytics to that and become a computer vision company and the path that you've taken to now have leadership products in computer.

Right. So when we started this company 16 years ago, we were trying to build the best silicon for video processing. And we were very successful in the first 10 years, and we went IPO in 2012. And at the IPO time, that's the first time we start hearing our customers saying, "Hey, capturing video is great. But moving forward, I need to analyze the video in real time." And that's when we start looking at different solutions to add what we call video analytics at that time. Now we changed the term to computer vision, but it's -- for me, it's very similar. Technology in 2012 for the computer vision was not the same technology we are talking about today. Today, we are talking about AI-based, neural network-based. In 2012, it's still traditional CV technology. So we kind of are saying that that's not the right thing to do. But until 2012, we -- first time we start seeing papers published on neural network start showing great momentum and great results. That's where we start saying, "Hey, that's an area we need to start investing." And we -- since then, we start investing heavily. And because our company approach is always -- we talk about the algorithm first. We want to understand what kind of computer vision algorithm we can build, then we optimize our silicon architecture for that. So with that, it took us a long time to get there, a lot of money invested. However, just like Joe just mentioned, we start selling our chip last year and we have a great momentum. We announced that in this earning call, we have 7 professional IP security camera guys in production with us, and we have a decent momentum in the consumer IP camera side. Also, we are also -- CV design wins in different automotive segments. And we talk about, in last year, we have more than 100 customer buying samples engineering development kit or bring our boards from us for CV products, not video products, for CV products. So I really think that all of this are -- is a proof that we have -- now we have momentum on our CV technology, but also prove that we have a competitive solution and also some technology differentiation for our customers. And that's where we will continue to focus on. Our traditional consumer business, camera business going to continue to go down a little bit, but I think that CV revenue will become significant for us. We talked about 10% total revenue from CV this year. And hopefully that -- when we deliver that, that will be a base for our future growth.

Great. So I think you've talked about the sort of multiple waves of adoption starting with professional surveillance. Professional surveillance today I think is around 40% of your total revenue. As we've talked to your customers, this seems like it's really kind of transformational in terms of the way they're looking at their products, the capabilities that you can embed in these cameras. So -- and you talked about 7 of the 10 customers, and I think that's sort of a 2x-plus average selling price boost when you start to ship CV to them. So can you just talk a little bit about what this means to the professional surveillance market? What kind of capabilities you're enabling? And again, you've talked to the adoption, 7 out of 10, but what's the pace within those customers of how quickly their cameras transition to AI?

Right. So I think first, the first customer who adopted our technology in professional security camera is Motorola. They acquired Avigilon. And since then, we start seeing -- because they are aggressively pushing this market with this CV technology, we start seeing everybody else start become more aggressive. And we -- the -- at the beginning, the application is still surveillance, face detection, face -- object detection, so that you could provide more surveillance function. At the edge, you don't need to streaming a video back to a car to do those surveillance function anymore. But however, we start seeing that this kind of CV in professional security camera enabling different type of enterprise applications. Amazon Go is talking about using a camera, this kind of technology, to enable surveillance in stores. And we're talking about people start using this for the -- to their access control or even payment system for enterprise. So we start seeing -- using a same technology into a different location by changing the neural network and the different type of services, you enable totally different new applications. That I think is the biggest momentum that we're counting on. And I think professional security camera will grow from a surveillance application to different type of applications moving forward, but leveraging the same investment that we are putting in.

So I guess a lot of the activity that you've seen in surveillance like in the last 2, 3 years, there's been a lot of AI already. And I feel like there's either you're sending a video stream back to the cloud where you're doing the analytics in cloud or you have a box kind of at the edge taking 12 video feeds that's moving directly into the camera. And I mean how pervasive is that? Is that -- why would -- to me, why do you need the box anymore, if I can do that?

Right. In fact, I will say, 2 years ago when people start talking of CV, people talking about at the edge, you put a box, put 12 camera in so you can leverage. But very quickly, we start seeing that architecture kind of disappearing. The biggest reason is you still need -- the camera still need to do some real-time applications. The box delay is still causing some delay problems for you to perform those real-time applications, not to mention that you still need to have bandwidth and storage on those boxes to apply those information. So from a cost point of view, from delay point of view, from the real-time application point of view that the -- all of the CV function sitting on a camera at the edge is the best solution to solve all the problem. And we see the trend is also -- that we are seeing is clearly indicating that, that architecture moving forward.

So as you think about that market, obviously, China is a big piece of it. You have 2 big Chinese customers that are probably still the minority of your professional surveillance, but big, close to half. What's the status of those big Chinese customers both from the standpoint of your existing base business and then really, the prospects for computer vision designs to get released to production?

Let's talk about the traditional business with those 2 company first. I think that in the last earning call, we talked about that because they are worried about entity list, they put in roughly $10 million inventory from fiscal year '21 to fiscal year '20. And based on the current ordering pattern, we start seeing one of them become weaker. So that we -- our guess is that customer start depleting their inventory. But I think majority of that $10 million is still sitting there and that we probably take time sometime this year to go through. And we continue to see they're ordering too from us. So that from a video product point of view, I think it's the inventory that we need to worry about. But however, the business will continue as is. On the computer vision side, we have been talking about that we have engineering activity with them, for both of them. And they have been -- we have been investing engineer resource to support and to do all of the development. However, the 2 uncertainties that they are worrying about, one is, the entity list -- that just yesterday, people start talking about again that dropping from 25% to 10% diminished rules. Although that's talking of Huawei, but I think Hikvision now are definitely worried that will apply to them. The second thing they worry about is the new export control the U.S. government is working on. And those 2 things definitely is the cloud of their decision. I think at this point, I can say that it's up to them. If they want to introduce product, we are ready. Our product is mature. We show that we have the other -- several other customer in production. So we are ready to support them, but it's up to them to decide whether they want to roll out the products now.

I mean we've had conversations with those customers. And what's interesting about that is I mean there seems to be a lot of enthusiasm for the Ambarella road map. And there's also a fair amount of trepidation about reliance on HiSilicon because when you talk about those restrictions, they're obviously aimed directly at HiSilicon. So it seems like it's -- I can see why they would have anxiety dealing with Ambarella. I could also see why they would have anxiety dealing with HiSilicon, which, by the way, is also a competitor at the camera side. So how does that trade-off play out? And I mean it seems inevitable that if you have leadership capability that, at least, in their cameras for export, that you'll be a factor.

Right. So I think that the -- for example, in the last 3 months, when the diminished rule -- 10% of diminished rule we're talking start diminishing. They swing one way. When they come back, they swing another way. So it's up to them. Really, we have no control on that. However, on the export business side, I really think that in U.S., both company has been impacted by U.S. government rules. So I think as for -- at the -- their actual business also got a little impacted because of that. On top of that, I think the -- you know that all of our Chinese companies are trying to develop a non-U.S. supply chain because of the concerns. So that's another thing that we are worrying about whether they will use other suppliers to do their actual business. That's something we need to continue to watch.

Yes. Okay. So I guess just a frame in this professional surveillance. I mean you guys talked last night kind of new information that you'll be above 10% computer vision this year. And if I think of that as being predominantly in this first wave of professional surveillance and I think about 20%, 25% of your revenue being Western camera companies, implies a fairly heavy attach rate with those customers as I mean if you're 10% for the year and you're low single digits now, you're obviously north of 10% in the back half of the year. It seems a pretty big tailwind to this surveillance business.

Well, first of all, I think when we give a 10% guidance there, you are right. The big part of it is professional IP cam. But also, we -- in the earning call, we talked about we also believe some of our consumer IP and customer who are going production this year, and then we'll add to that 10% total revenues. So that's a combination of those 2, but big portion is the professional IP cam.

So it's a pretty big tailwind. And obviously, there's uncertainty on the rest of it that's why you guided the full year to be up without a boundary but...

I think the key is not the unit number, it's really about ASP. The ASP growth is significant for us.

Yes. And talking to your customers, this is the main focus that they have as well in this business. So...

That's right.

Okay. And then I guess the next wave you started talking about was consumer surveillance. That seems like a market where just as a user of your customers' products really having some AI capability to stop my phone from alerting me every time a squirrel walks by, you can see the utility of it. What's the pace of adoption there? And is it harder to get to those kinds of price points and things like that?

Right. Today, all of the consumer IP cam with alarm system. They are using cloud to do AI -- to do computer vision. We believe that we will be the, well, first one to really enable edge AI for this market. And that we talk about there is a design wins we would go after, and we believe that some of them will go into production later this year. And that's why we have some revenue by -- in Q4 this year. So the momentum is strong. But also, the -- another thing I want to emphasize is consumer IP cam in the past is really about doorbells, indoor cameras, surveillance applications. But suddenly, we start seeing other possible applications. That access control we talk about on the home side can be your car come approaching your garage, your garage door opens. So it's not surveillance anymore. It's really become more application that we can enable with the customer and that, therefore, expand our serviceable market. Not to mention that, for example, the thing -- the access control technology with MYCS is pretty much a consumer IP camera. However, that can be used for payment system because of the access control we add to it. So suddenly, you can see that our traditional design that for home monitoring go to a different application and enable different market. That's what we are seeing, and that's why we're excited about that we're going to see more and more new application. Although small at beginning, but some of that has momentum to become million units market for us.

And last night, you guided for the April quarter quite a bit better than our expectation, better than seasonal, and you talked about this home security as being the reason. Can you talk a little bit about that strength?

Yes. So first of all, the strength is now at just 1 customer. It's really have a broad base of customer, that all of them doing quite well. Of course, led by our largest customer, which is Ring, in that market. And we believe the momentum is strong in that market and we have a good market share there. With the CV momentum, we hope that and we believe that we should be able to maintain our market share in the new market. So overall, we have a good feeling about this market as a whole, not to mention that the majority of the guidance raised in Q1 is because of the -- this market.

And you mentioned that this isn't just surveillance anymore, either professional or home. And you look at some of these applications, and I think you mentioned Amazon Go, but like the retail concept stores that are sort of cashier-less, like they really are -- I mean we've had conversations with implementers, and the focus is really vision like, yes, RFID is part of it and other technologies are important. But vision, they sort of see as the foolproof piece that when you -- obviously, in a store with people in it, you have people monitoring things like the ability to detect shoplifting in progress, those kinds of things, the ability to detect what's being put in your shopping cart, all that is vision-based. Who are the -- I guess how do you approach a market like that? Who are the partners that you work with on something like that?

Right. So the go-to-market strategy for us at this point is of course that we have our own marketing and sales going up to the customer. But however, we start seeing that all of the new applications you talk about, not just the checkout counters. In fact, we have a customer talking about having a robot working between aisles in a Safeway. And they can do the autonomous inventory check. Because they use high-quality cameras, look at all of the products on the shelf, they can do a inventory check automatically. And that's a product enabled by our technology and our customer definitely working on. And our go-to-market approach in the past was counting on our own marketing sales. But we found out that a lot of our partners, particularly on the algorithm side, we talk about many of them in our earning call this time, they enable new applications. Each one, for example, the checkout counters, license plate identifications. They bring the customers, their customers that are looking for complete camera implementation with those neural network. And that we believe that working with the software partners that have a unique neural network to enable new applications, that's another way that we can work with -- find out more of this kind of potential customers.

Great. So obviously, the surveillance market is going to do the heavy lifting from the earnings standpoint in the next couple of years, but a lot of the long-term enthusiasm comes from automotive. So maybe we could shift to that a little bit. And right now, you're about 20% automotive. I think you're sort of crossing over the majority of that being OEM-type wins aftermarket. So maybe start with that, your existing base of automotive revenue. It seems like there's a pretty good growth path before we get to computer vision. So can you talk a little bit about that?

Right. So I think that our current automotive business, majority come from recorders, but shifting from aftermarket to the OEM base. In fact, in the past, we've talked about a lot of Japanese and Chinese OEMs now moving to Korea. They're going to have a recorder in their car and as a standard offering. So that business continues to be strong, and we start seeing more and more people asking for recorders. The reason for that is simple. I think the driver -- in the past in China or Russia is really the driver trying to protect themselves. But now with ADAS become more and more popular in those cars and people need to prove that -- the carmaker need to understand whether the driver is really watching the street or paying attention to while they're driving. So a recorder has become an important tool to document when there's an accident, who's -- whether the driver is paying attention, what's the condition of the car at that time. So I think the recorder become -- I think that's the main driver for the recorder business for now. But however, just like you said, moving forward, a CV type of application is where we focus on. The short term, we -- in the last -- starting last quarter to this quarter, we start talking about we have design win with the fleet management companies doing not only recorder, but also CV type of functions. So for outside ADAS, inside monitoring the drivers so they can provide service to understand exactly whether the driver performed to the way they defined. So that's definitely a short-term business that we talk about and we already have a -- not only design win, we're probably going to start seeing revenue the end of this year from the CV-based fleet management products. Then we also talk about that finally, we start seeing some momentum on the electronic mirrors, and also that's going to give us some design win momentum, which also CV-based because the -- I think one reason that people start thinking about electronic mirror make sense because combining CV, you can start -- add more values. For example, you can do blind spot detection on your e-mirrors, which is definitely not only nice, but also adding more safety to the drivers. So those 2 things are short-term CV-based momentum for us, including driver -- also driver monitoring system. But longer term, I think most important thing for us from a business development point of view and other design win point of view, as a company that getting design wins from ADAS, consumer car, ADAS is important for us. That's something we are working hard on. And also that we believe we start seeing a lot of momentum from the OEM side, talking about Level 2+ or Level 3. Basically, it's an ECU design with more than ADAS function, but help you to control the car better. So I think this 2 market, ADAS and Level 2+ with the ECU design, those are 2 things that we are focusing our business development and the development side of.

Yes. I want to talk about that. But maybe first, just from -- just understanding the revenue driver, it seems like surround view, this sort of overhead -- sort of artificially created overhead view is a pretty popular option now with cars?

Yes.

Shopping for a car myself. It's an expensive option, but it's very common. And then the e-mirror thing, we got excited 3 years ago talking to one of your customers about the possibility and it's sort of not intuitive, but there's a fair amount of computation involved in these things. And there's a lot of discretionary wind resistance that you take away by getting rid of side view mirrors and things like that. But it's taken a while still. And I guess we should -- there's probably a little bit of a note of caution about overall ADAS, and autonomous adoption takes a long time in cars. But when I start thinking about tens of millions of cars and $10 to $20 ASPs for e-mirrors, like it seems like it's a pretty big TAM still.

Well, it's just -- 20% of total revenue today from auto is really about $50 million, right? Any OEM design win can generate $10 million , $20 million. So it could be a significant growth, just even get a 1 design win on the e-mirror side, and not to mention that probably ADAS side. So I think that's the reason we continue to focus on this market.

Great. And then shifting to the ADAS. Again, maybe a little bit of a history for people. You got -- I mean GM Cruise was here in the session before us. A few months before GM acquired Cruise, you guys acquired VisLab, which was a company that had been doing autonomous cars in 1998, semi-autonomous cars in 1998. And so the long history of software development in this space and systems knowledge, and you bought that before those assets got very valuable like just a year later. So for a company that's relatively small and relatively new, you have a fairly rich history of analytic development around this kind of -- everything from ADAS all the way to autonomous. Can you talk a little bit about that?

Yes. Well, the reason we acquired VisLab is again the -- our algorithm-first approach. If we -- our belief is, if we don't believe, we don't understand the real application, the real algorithm required in this space, we should not get into that business. And working with VisLab, that's exactly what they offer to us. Because they have 20 years autonomous driving experience from an algorithm point of view and from a software point of view. By working with them, by acquiring them, it boost our understanding on this business especially on the engineering side. And then finally, we can optimize our silicon based on this knowledge. And that's what we did for the CV2 and CV2FS that we announced just at CES. And not only that, I think when you have a silicon, the biggest problem is how you convince OEM or Tier 1 says, "This chip can perform to the way that you say it can, right?" If you only have silicon, you cannot prove it. So what we found out there is, with VisLab software, we can implement a complete system with our chip. With the VisLab software, with the improved AI capability we add to it, we can demo a Level 4 car, which we did at the CES this year, and with not only autonomous driving at daytime, but also at night with our technology. All the hardware and software in that car is ours. So I think we use that acquisition, the talent as well as the technology we acquire to prove not only we have silicon, but also we know how to use this silicon, implement assistance to deliver the performance our customer wants. I think that's the point we want to strike at this point. But in the future, we definitely need to plan to say, "Can we leverage this to get design wins? How do we monetize the software?" This is definitely a challenge that we need to continue to consider and find a way to make it become a reality.

Great. So the path to ADAS into autonomous is, you've said it's going to take a couple of years. You've never shied away from how long that's going to take in your automotive growth. In the near term, it's going to be driven by the other stuff that we talked about. But I guess some of these applications, when I think about driver monitoring or actually, the logistics win that you announced last quarter with Mercedes, which is more like for delivery vehicles. Like that's less -- that's not forward-facing ADAS kind of life determinant stuff. That's stuff that is additive to safety, could probably be adopted much quicker. And in fact, with the in-car monitoring, it's mandated to be adopted fairly quickly. So can you talk a little bit about those opportunities?

Well, yes. In fact, there are -- just like security camera market, in automotive that -- we start seeing just a camera with a CV start enabling many, many different applications. We talk about the CoROS, which is Mercedes-Benz using our camera to do assistant check-in, check-out for the packages, delivery vans. And that definitely is an interesting application. We didn't even know that until we talk to Mercedes-Benz for the ADAS. And this group come to us and say, "Okay. I can use your technology with different reasons." And we are -- start seeing a few more things about that. Of course, those applications from a revenue and size point of view is not as big as ADAS. But however, you're definitely seeing that it can be a potentially good business for us for the delivery vans. We're still talking about millions of units out there. So I think that we continue to monitor where our technology can go, can leverage our hardware and software without adding tape-out [ not a new ] chip, adding new software functions. So just all of the application we talk about right now is leveraging silicon portfolio. We talk about send -- software SDK that we offer to a lot of customers. So the leverage is there, and we -- by -- although we talk about enabling a lot of different application today, but the total amount of engineering increase, R&D increase is not significant compared to the amount of serviceable market that we enabled.

Again, you think about the delivery van thing, millions of units. I mean it's pretty high-dollar content for vehicle for you. So I mean this could be really impactful for you as a company if that -- if you start seeing...

We talk about that each van of Mercedes-Benz is going to take 2 CV2s, and we basically give a guidance of CV2 price is around $50. So that's basically the content we're looking at. So it's a pretty healthy margin, a healthy business for us. And also, we believe those smaller segment in automotive, people are willing to pay a higher price because of the smaller volume and had to be customized for them.

And we've talked to our customers in the in-car monitoring stuff, too. And there's a lot of enthusiasm for the Ambarella solution there. Maybe one of the things that I, as a semiconductor analyst, underappreciated initially was the ASIL B certification with a couple of the new chip solutions. It seems like that's a fairly challenging threshold to cross for you guys.

Yes.

What does that enable for you?

So first of all, ASIL B doesn't help us to improve our CPU performance or CV performance or other video performance. It's really about follow a safety standard defined by the automotive industry. Without that, people won't even touch you for any safety-related products. ADAS is definitely one of them. So -- but ASIL B, the complicating part is not the technology side. It's really the process you need to follow, making sure you document everything, making your design -- you have safety concerns in there. So if there's arrow in your chip, you know you can immediately pull the arrow and find a way to mitigate the risk. It's really about safety. It's not about technology. So although it's just engineering practice, but the amount of resource and the money we need to put in to enable this function is significant. And we're happy that not only it goes through the process, I think we are confident that we're going to pass that ASIL B certification set up by the automotive industry.

Right. So I intentionally got pretty deep into this before we started talking about forward-facing ADAS and autonomous because I think people view it as this kind of binary around those things, a lot of opportunity before you get to those things. In terms of forward-facing, can you talk a little bit about what you're doing there, the HELLA Aglaia relationship and the opportunities for you to potentially displace Mobileye and some of those applications?

Right. So here, this is a market we definitely need a software partner. Although we have our own algorithm, but to satisfy Europe NCAP standards that we believe we need to work with a partner. We talk about HELLA Aglaia, which has the algorithms running on other platforms to deliver ADAS function in the past. And the reason they choose us because not only our performance, but our power consumption that we can deliver with all the performance requirement. So the combination of the powerful silicon platform was already approved by the automotive industry, together, that will probably a fast way to -- for us to cater the market, and we're going to continue working with HELLA Aglaia in the market. But also, in this time, we also talk about other Asian type of ADAS software partners like StradVision in Korea. So -- and the SenseTime in China. So all of that, you can see, our strategy for ADAS is simple, we know -- although we have our own algorithm, we are not trying to push it. We are trying to work with the industry partners that they already have a proven ADAS algorithm that can be used immediately.

And it seems to me that there's a fair amount of opportunities that you probably can't talk about yet because in this space, people -- Mobileye is obviously incumbent and nobody's going to announce anything until they have something to replace them with. But it seems like there's a fair amount of awareness with you guys that maybe wasn't here 12 months ago with those types of customers.

I think that's true. I think that with our business development activity, a plus, that we continue to show our performance, especially for people who spend time to put their algorithm onto our chip and look at the performance and power number they can generate, that's where we get the most tractions. We really need to have a -- helping customer to go through that stage, then suddenly, people realize what we are offering others compared to our competitors. So it's a process. But we definitely think that in the last 12 months, we get -- we improved a lot. On top of that, today, with the functional safety chip, the CV2FS, that we announced is really also -- we're going to start showing the documentation that we pass, the certification that we'll add, continue to add confidence to our customer that we are not only will be able to provide a technology they can use, but also we can be verified as a safety supplier for the automotive customers.

Great. And then thinking about higher levels of autonomy, Level 3, 4, 5. I guess the enthusiasm from investors has maybe waned a little bit for that because it's going to take a while, it's becoming clear. At the same time, there's still a healthy amount of development activity, a lot of it in China. And you've talked about your role there. And I think you've sort of alluded to wins that maybe you can't announce yet. But can you just tell us what you're doing in those kinds of markets? And it seems like it's largely complementary to what guys like in video are doing at least as we sit here today?

So in addition to ADAS, we are also focused on this we call Level 2 plus or Level 3, Level 3 application. So basically, instead of just 1 camera facing forward, there are -- our OEM customers start putting multiple camera in the radars around the car and do sensor fusion to add more safety function into the car. We call this Level 2 plus, Level 3. And the difference is those chips go sitting at ECU instead a camera module sitting on a windshield. So with that approach, we already start talking to -- not only talking to customers, but also showing people how our current chip can play a role in there. I think the biggest advantage for us is that our CV performance as well as the power consumption help us to getting there. But I think Level 2 or Level 3 -- Level 2 plus and Level 3 is definitely a new market everybody investing on. When I say everybody, means OEMs, and we try to prove to them that our chip and our road map can help and serve them very well. In terms of Level 4, Level 5, I really think that will take longer than people expected. And people continue to invest, but I think the real project is -- the number of the real project on the Level 4, level 5 is much less than the Level 3 level.

Great. So I just have one more big picture question, and then we can open up to the audience. This is more just generally about how you're thinking about the company at this time. People pick on the stock because it's expensive on 2020 earnings. And so I think it's important to put into context how you're thinking about this year or even next year. You're investing a lot. I mean you could be earning a lot more if you weren't investing a lot. So how do you prioritize earning the money this year versus investing and obviously these are long-tail investments that you're making. Just think about the -- can you tell us about the trade-offs that you see there?

So starting 3, 4 years ago, when we started investing CV, we truly believed this company if -- was -- a significant value of what we have right now is because of CV market. And to be successful in our traditional surveillance-plus car, automotive business obviously important. But we do see there are many, many new applications enabled by CV market. So when we define the company moving forward, CV it is. There's -- really, we think that we will not ship any camera without CV function in the next 2, 3 years. Video product, by itself, become traditional and the value is getting lower and lower. So we will -- if you want to -- if you -- the way that Ambarella's management looking at this is, we are totally commit to CV, and the focus is not only just enable the CV technology, but also new applications and new markets that we can enable with that. However, at the same time, we also believe that having a healthy financial model is important. In fact, we have been cash flow positive since the IPO, and we want to keep that. So the trade-off between this 2 is as long as we can continue to have a cash flow positive, we will continue to invest heavily into CV direction. One way to invest heavily is to move more resource around the traditional CV -- video products into CV areas and invest heavily on advanced technology. For example, we talk about that we're going to table our first 5-nanometer chip this year and the investment on the R&D and the tape out fee and the car tools is significant. So that's how we look at the company. CV technology and CV application are our future.

Great. Now let me see if there's questions from the audience. Keep going. Anyone? Maybe on that note of cash, I mean you have a pretty good cash balance. I think that's been comforting as you've sort of navigated a bunch of factors in the last 18 months. But what's your thinking about the right level of cash and sort of buybacks versus acquisitions versus other activity?

I think the priority for us -- when I look at cash, there are 2 main priorities. One is, we need to help keep a very healthy cash position for 2 reason. One is with the current market situation, cash is the king. We need to make sure that we have enough to continue to maintain our operation, which I think we can. The second thing is, with the current market situation, I think the acquisition market might become available to us because of the valuation. Probably, I hope we'll become more reasonable on the private sector so that we can have a more acquisition target that we can look at. So I think there's 2 things that we're really looking at how we use cash. But at the same time, we still have a buyback plan in place and which is available to us all the way until to June. And in the next 4 meeting, we'll talk about whether we want to extend that.

Great. Okay. And then maybe just one of the newer opportunities you've been talking about, robotics. Can you give us a view on how Ambarella could position themselves in that?

I think robotics is one application I truly believe -- personally, I truly believe in long term, when I say long term, it's 5 to 10 years. This is going to be a huge market. Look at all the manufacturing side, people are talking about the problem we are dealing with today, the manufacturings, the home automation. There are quite a few things what can be addressed by robotics. So that's why I'm excited about this technology and that we are building a platform that can enable people to start developing that. However, we also realize that although the design cycle is not as long as automotive, but it's long. The design -- for example, some of the design win we are talking about is 18- to 24-month development cycle because it's very complicated system. So our investment on the robotic is purely based on belief that we believe the market will be there and we're going to be, and we have some several important design win activities that we're engaging with. But the design cycle is longer. The revenue is probably longer than our traditional surveillance market.

And the state of computer vision in that space seems pretty limited. Like talking to some of the incumbent robot companies doing cobots and things like that, they're really not taking advantage of the advantages of machine learning and AI and all the stuff that that's given us in terms of solving these vision-based problems, it seems like we're really at a very early stage of it all.

I think you're right. But however, I think when -- from the customer we talk to, they all realize that the trend is vision becoming important. But it's not just using neural network to do object detection. It's really about how to measure the distance, how to -- decide how your movement -- basically, our experience with autonomous driving can be used in -- the robot that you need require to movement and not -- and also, you have to do slam. You have to understand this 3D environment. There a lot of vision technology you need to add -- need to be added to this current robotic system so that it become more powerful, more -- providing more function. And that -- I think that's why I'm excited about this opportunity.

Okay. Last question, I think, I guess we talked about the history of this. You guys were talking about computer vision certainly before I recognized how important it was going to become. Like what -- are you surprised how important it's become? I mean machine learning seems like it's really been an accelerant to a lot of these activities. And obviously, the enthusiasm for it in video is pretty well understood. That's mostly training. Like they're mostly creating these databases, creating these algorithms that are going to be deployed at the edge and it's going to need a solution like yours.

Well, in 2012, when we start looking at this technology, yes, I'm definitely surprised how fast we get to this point. But I really think that's because of the neural network development that the global institution put in and the sharing technology enabled this growth so fast. But because of this -- the growth that we're seeing on the machine learning in the last several years, I have to believe that the progress will continue to accelerate and continue to get more -- in providing more functions in this industry. So I'm pretty optimistic about this opportunity.

Great. We'll wrap it up there, Fermi. Thank you very much.

Thank you.