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

Good morning, and welcome, everybody, to UBS TMT Conference. We are very happy to start this morning with Ambarella, a very interesting company, focusing increasingly on computer vision. With us today Fermi Wang, CEO and Founder of the company; Casey Eichler, CFO; and Louis Gerhardy, Head of Corporate Development. And I'm here today with Tim Arcuri, Head of U.S. Summit Research for UBS. And I'm Nick Gaudois, I'm with the APAC tech research team for UBS. With that and with no further ado, let's kick off.

So Fermi, over the past few years, you have invested a lot into computer vision with a CVflow architecture. What are the key building blocks? And how do you manage to achieve differentiation notably behind from an architectural perspective?

Right. So Ambarella is a 16-year-old company. When we started this company, we focused on video processing and video compression. And we started looking at computer vision, I would say, 5, 6 years ago. When we design any architecture for around those technology, we take approach, we call, algorithm first. Meaning that when we try -- always, for example, when we do video processing, we look at all the algorithm we want to implement on the video processing point of view. After we optimize the algorithm, then we -- based on the algorithm, then we decide the best architecture to implement. When I say the best architecture, not only just deliver the quality, but also we want to get the performance at the smallest by size as well as the smallest -- the lowest possible power consumption because we are focusing on age device, power consumption is important. And that's an approach if also applied to computer vision when we started 5, 6 years ago. And at that time, Neural Network-based computer vision just started. And we believe at that time -- computer vision can go back 30 years, but the traditional computer vision algorithm doesn't work in our opinion. And when we look at the Neural Network approach 6 years ago, our CTO, Les Kohn and I, decided this is right direction that we want to invest. And we spend a lot of time and the resource to understand the algorithm that we want to improve on computer vision. And in fact, because of that, we acquired the company called VisLab, which is autonomous driving company which has plenty software algorithm developed for the autonomous driving in the last 20 years. And by looking at those algorithms for autonomous driving, also the computer vision algorithm for security camera or other market that we are interested in, then we came up this CVflow architecture. And the most -- if you ask me, the one most important differentiation is we figure out how to not only deliver the highest possible performance for computer vision processing point of view, but we also figured out how to use the lowest possible power consumption to achieve those performance. And in fact, if you want to ask how we've achieved that, I think the only one magic word is utilization of the hardware, right? If you look at -- a lot of people design general DSP or CPU or GPU to implement computer vision. And if you look at utilization of the hardware, it's very low. And we focus on finding out, architecture deliver multiple acts of utilization of hardware so that you can -- you don't need to drive the high density over hardware, but you still achieve much better the computer vision performance while you can drop your power consumption. So the computer vision performance per watt, we are much, much better than our competitors, which, including all the big names out there.

Great. Thank you for that for me, Fermi. That's very clear. So with that focus on, I guess, efficiency of design and power consumption, you're leveraging this core computer vision algorithms for several applications, including person detection, object classification, analytics. Can you elaborate as to where you see the most traction application-wise for design wins so far?

Right. So in fact, there are 2 markets that we have the biggest attraction right now. One, we call the security camera or some people call this surveillance. But basically, even in that market, there are 2 major different branches. One is we call professional security camera, basically industrial level of a camera that being used in -- on the traffic -- in the traffic, in the government building, in the banks, in the casinos, this kind of industrial level of professional security camera, which we are doing very well in there. And also, the other branch under the security camera is we call Consumer, basically the home security camera or -- so we call smart home camera. So this is like the ring camera, like nests and the -- along Datacom. There are many, many customer out there using our solution today. So this is probably our biggest market, which represent roughly 60% of total revenue today. And the second market, which the TAM is even much bigger than the security camera, is automotive. Although the automotive only accounts for 20% of total revenue, but I think the potential of this market is huge. We will talk about this in detail. I believe, that -- but however, just look at not only just the ADAS market that already become popular. Moving forward, the Level 2+ car that Tesla is doing and also even in the future Level 3, Level 4, I think this is really -- we are talking about another 20, 30 years of development work for computer vision for automotive. This is a huge market. We think it's a multibillion-dollar TAM for Ambarella. So this 2 market are the most important and will generate revenue for us. But in addition to that, I have to believe that, in the future, any robotic application, anything you want to -- using automation for the delivery, for the smart manufacturing, smart city, smart home, anything related to computer vision, you need a smart eye. Then you need a computer vision chip, and that become our market. So I will say that, in 10 years, I won't be surprised that robotic application, which require computer vision technology, will become our biggest revenue driver at that time frame.

Great. Thank you. That's great. Over to you, Tim.

Thanks a lot. So Fermi, you just kind of talked about what the drivers are going forward. If you look at computer vision, revenue was mid-single digit in the first half of this year and sort of marching toward 10% of your mix. And you did just talk about 3 waves of drivers going forward. Can you just -- I just wanted to kind of double click on smart home cameras. And I just wanted to ask you sort of why you're winning there. And then maybe you could also double-click on sort of how quickly you think that the robotics piece that you just talked about, how quickly you think that that's going to become a much larger piece of your revenue. Is that still out there 3 to 5 years? Can you just kind of double click on those 2 things for us?

Okay. So first of all, on the home -- smart home camera, we are winning basically low power because, if you look at the rings, the Dolby type of application or battery-driven camera design in your home, all of them require very low power technology in there. And let's not lead that, but also low power with computer vision moving forward. I think that combination is very unique, and we definitely is one of the best to provide the combination of low power as well as high-quality camera solution. And in fact, also, we were the first to the market. In fact, many years ago, when we worked with a company called Dropcam, which is pretty much the leader of this market and which was acquired by Nest, now become part of Google. So we are the first in the market with a better technology. I think that's the reason we are winning. And we also talk about the second wave, our computer vision revenue is from the smart home camera, and we believe that we're going to start seeing the trend happening next year, particularly you're going to see a meaningful revenue at the end of next year. Back to your second question about robotic. In fact, I would like to argue that ADAS or autonomous driving is a unique type of robots. And so when I say robotic implication, I am trying to say there are many different application. People using very simple processor and for the -- your vacuum cleaner at home today, like iRobot. And there are people trying to using very complicated technology on the manufacturing side to do smart manufacturing. They are all different application. All -- and every different application will require a different combination of performance, power consumption and also video processing technology. And so that's what I'm trying to say is robotic application or market is not one market. It's a multiple different application. All of them require different combination of a price and performance. And so what -- in terms of the starting of the market, I think ADAS already started. Level 2+ will happen in a year. iRobot is there, already shipping some kind of simple vacuum cleaner. You can call it a very simple robotic today. But I can imagine that 10 years down the road, that will be even better, much better. So I think we are really at a -- not even a hockey stick of the market yet. We are really seeing gradually movement of this computer vision into robot market. So we -- I have to say, this is really for early innings of this market. And I have to say, I think that will take at least another 3, 5 years to develop to go to the higher volume market opportunity for us. But I truly believe it will be there.

And when you think about that, there's a lot of much larger companies that are going after that space. How do you plan to ward off competition from much, much larger companies that can invest a lot more money?

Right. So in fact, if you look at the computer vision semiconductor company, I really think we can say NVIDIA is one, and the Intel bought Mobileye as a different one. And everybody else, I think there's nobody has really meaningful computer vision program that we think that they can compete with us in the long run. So let's focus on these 2 things. NVIDIA is using the GPU to compete, and they are -- because they -- really, there a lot of user for the application. They are the first one to move into the market. But however, I think that's also eventually going to become their biggest weakness. I think that for the market that we focus on, let's call the H device -- computer vision for the H device, power consumption is critically important; and ADAS, which is really a very popular auto application. But -- however, if the power consumption of your main silicon go over 4 watts, you don't fit into that the camera anymore. So the power consumption -- anywhere, if the power consumption is a limitation, NVIDIA doesn't have a solution. In fact, they don't have a solution to address this really aged device type of application. So they can probably develop some powerful solution with 100 watts for the level 4, level 5 cars, but they are not really overlapping with us on the -- our target market in terms of smart manufacturing, smart home or even the ADAS type of application in automotive. Mobileye is different. Mobileye is totally focused on ADAS, but they don't focus on anything else. So we'll find a way to compete with ADAS. And -- but however, if you look at all the other computer vision application, we don't see Intel Mobileye at all. So my opinion is, if you look at -- if there's one company that's really looming for computer vision with video processing technology in there, we probably have the most focused technology and also business development direction than any other companies that are there. In terms of...

Got it.

You talked about resource, so let me address that. Looking at our CV2 family, it's a 5 10-nanometer chips in a family from CV2, 22, 25 and 28. Now we have CV2 functional safety chip. Five CV chip or 10-nanometer chip, but we develop with our budget at roughly $100 million R&D CapEx -- R&D expense every year. And I have to say that we are very efficient using our R&D budget to develop a family of chips, even -- I don't see any other company, including NVIDIA and the Mobileye, has this kind of a family of chips that address different performance per watt and performance per dollar.

Got it, got it. Can I ask about ASPs? I think you've talked about computer vision ASPs are roughly twice that of video processing products. Can you just talk about how the computer vision ASPs are going to trend over time? And maybe also share more thoughts around the gross margin profile for CV products versus video processing.

Right. So just like any other silicon, consumers introduce, the price is going to continue to go down from there, right? However, just like I said, there are so many different CV application out there. You can always introduce a different chip -- different CV chip down the road, which will provide different -- maybe higher performance or focus on battery power efficiency or SoC target on particular application. So when you do that, we -- every time we introduce a new chip and have really a targeted market, that really reset the price, the ASP. So what I'm trying to say for -- if you look at just 1 silicon, 1 CV silicon, we introduce the price is high and can go down from there. But every time you introduce a new chip, you basically have a way to reset that ASP curve. And so that you can always have a healthy ASP -- average ASP for the whole company. From a gross margin point of view, we still believe that our long-term gross margin at the 50 -- 58 to -- 59% to 62% is the right gross margin target for us for both video and the computer vision. And the -- however, that really is a combination, right? You can imagine our high-end chip like CV2, higher gross margin, our corporate gross margin range. But our low-end chip, that CV 28, has lower gross margin. So -- but we still believe that our long-term gross margin target is continue to be there for a while.

I'll turn it back to Nick.

So maybe shifting gears and going to the China market. It seems like you've been able to actually sell to Dahua. I think you talked about this in your last conference call, and even possibly Hikvision. May we assume this is coming from the application of the de minimis rule for the total camera solution? And are you assuming revenues continuing for those customers in your base case going forward?

Yes. First of all, I think that for Dahua Solution, we definitely pass the de minimis rules, and we believe we can continue to ship to them, including our CV chip. In fact, we announced we have a CV design wins with Dahua. And also, we talk about, we already start receiving orders, production orders in Q1 from the -- our Chinese customers. So that's all the good things. And -- but however, I want to make sure that when you talk about Hikvision, we didn't talk about computer vision design win from there yet. We are talking about -- they come back to start ordering our video products. And basically, they -- I think they gradually digest the inventory -- the video product inventory they're building 12 months ago and gradually digesting it, we start seeing -- they are coming back to order some of those parts starting Q1 next year.

Right. Right. Okay. Clear. So -- and I mean, this is a very difficult question, but from where you sit, your base case is continuity of those businesses basically for the next 12 months or late into next year?

Yes. Well, I -- we are assuming that the -- we cannot -- first of all, we cannot guess what's the -- how the political situation is going to check. We are assuming that what we are seeing is a status we're going to continue to deal with. And based on that assumption, we are giving the same guidance about next year. For example, we did talk about that we believe our computer vision chip will generate roughly 25% of total revenue next year from 10% this year. So that's basically assumption that there's no more geopolitical situation that will impact our revenue either way.

Right. So that is cool basically. Got you. And I mean, regarding -- the flip side of that situation, obviously, is how do you assess the removed position from HiSilicon, which obviously has been traditionally a key competitor for you, very capable up to now, but of course, it's facing significant issues in deploying designs into foundries, specifically, of course, for Huawei? So how do you look at that going forward? I mean do you think they still meet? Or do you think they get a -- if nothing changes, they may effectively fade down?

First of all, I think one of the reason Hikvision -- Hikvision, not HiSilicon, Hikvision hasn't used our CV chip because I think they continue to build up a huge amount of HiSilicon solution already. So they don't -- they are not eager or they have no urgency to switch their solutions. So although HiSilicon cannot build any more inventory, but I have to believe, they build a huge amount of inventory that they can continue to supply internally also to their key customers. In addition to that, geopolitical situation is hard to predict. On top of that, there's already rumors about Huawei want to spin off that semiconductor division. Again, that's a rumor, we don't know whether that's true or not. So there are many variables that we have to -- we might have to deal with in the long run. But my suggestion to my -- all my colleague is there's a window opportunity that we can win many designs. And that if we secure design and we took our customer into production as quickly as possible, even they found out the way to come back, we already acquire a market share that we probably can keep for a while. And also, if they spin off, it won't be Huawei anymore. It might be a different situation. We might be competing with a totally different competitor. So who knows? The only thing I'm telling my colleague is, today, the window of opportunity is open. There are many customer that we can win, and we are winning a lot of design in China. And that's how we -- our job is to make sure that we can help those customer going to mass production as quickly as possible.

That's quite clear. The -- and one other question on competition in China is there's quite a lot of discussion, as I'm sure you're aware, of those customers, Dahua and Hikvision, going potentially the [ AZ crude ], right? So I mean if you could explain very well of the importance of -- obviously, we -- [ the Viagos ] optimizing design, et cetera. I mean another alternative is you just throw basically [ guessing ] and transistors that are problem, and you end up building some not so efficient ASIC, but you internalize, and you isolate yourself from those dynamics. So how do you consider those risks?

Well, in fact, today, in China, because the government pouring so much money on the semiconductor, if you don't do a semiconductor design, you are not a high-profile Chinese company at all. So for any system company, I don't even need to ask what they are doing. They are all doing it because government is putting -- giving them money to do it. So it's really about how you want to position yourself. I truly believe that the only way we can survive, just like competing with NVIDIA and Intel, we have to be a technology leader. We have to believe that we can continue because our experience on video, our experience on the computer vision, more important, our experience building a very efficient silicon, which is still not an easy job to do. Based on our experience, we can build a much better technology, higher-end mainstream technology chip, that we can continue to be a leader in this market. If we're not the leader of technology, it doesn't matter who's doing that, we're going to lose the market and we have no reason to survive. So we already have plenty of competitors out there. It's NVIDIA, Intel, Renesas or TI. They're all bigger than us. It's really -- the only way I can focus on is to build our technology and innovation, so we can compete. It doesn't matter who they are. Internal solution is always difficult because as soon as they have an internal solution, they're probably going to use that. But however, if we provide a much higher end chip, then we definitely can have a market share on the mainstream higher end, and that would let -- goes to their own chips.

Very clear. Tim, back to you.

Sure. So everyone wants to talk about computer vision, and that's obviously the fastest-growing part of the company. But I wanted to ask a question about video processing. And maybe, how do you see the relative growth rate of that business versus the computer vision business? And maybe, if you could think about the mix in a couple of years and if you sort of look at those relative growth rates, maybe just double click a little bit on the video processing market.

Right. So first of all, I really truly believe that all of the video processing market will move to computer vision in 5 years. All of them. Maybe some of them will have a longer trailing tails, but the -- when we talk to all of our current customers, it doesn't matter security camera or automotive, computer vision is on their road map and is most important, the things they're asking for. So the -- our video product revenue will continue to shrink, where our CV will be taking over. However, I need to point out one thing. That doesn't mean video processing technology is not important. In fact, you make it -- the whole video processing technology become more important because if you cannot process a video better, then you're feeding some garbage video into your computer vision processor. You've got to garbage out. If you cannot process video properly like at any light condition, for example, at the pricing line, had high contrast lighting or even low -- very low light conditions, doesn't matter lighting condition used, you always need to represent the video -- the -- all of the video content in a way that computer vision can apply the algorithm properly. So video processing technology continue to be very, very important, even though all the revenue moved to computer vision. So that's another thing played to our advantage because most company out there, talking about computer vision, they don't have enough video processing technology, in my opinion. If you look at NVIDIA, Intel, they have never done any proper -- appropriate video processing technology. And we think that our video technology, although it's not going to generate video processor revenue anymore, but it will be -- continue to be one of the biggest differentiator for our computer vision chips.

That makes sense. Got it. I wanted to ask you about surveillance. And I just wanted to ask you when you think that computer vision is going to play a bigger role there, whether it's professional surveillance or whether it's consumer surveillance. And can you talk a little bit more specifically about what trends you're seeing in that market and maybe what your market approach there would be that might differ from the other markets that you're talking about.

Right. So for surveillance market, in fact, computer vision is there already. In fact, just people didn't know that. In the past, in fact, last several years, people continue to streaming video from the edge, from the camera to the server. Then you apply computer vision algorithm on those low-stored transmitted video on the server in a non-realtime format. For example, the -- in the Boston bombing situation, that people go back to the video and applying computer vision techniques on those video, and they've traced back to find the potential suspects. So the technology has been there. The major difference that we brought into this is we help people to move that computer vision technology from the server to the edge. Why do you need to do that? Because when you -- every time you want to do that on a server, you need to try to make a video, which will require huge bandwidth on the infrastructure. And you need to store the video, which require huge storage space and the cost. And that you had to run the application on the server side, which is processing cost. So there are multiple infrastructure costs and the processing costs and storage costs related to the server architecture. But more importantly, that architecture can scale you to the -- 1 million camera. But how about 100 million camera? How about 1 billion cameras? There's no way you can continue to scale those costs up. So the easiest way to do that is you push all the computer vision to the edge. Every camera, analyze the video, identify most important thing. So you -- instead of transmitting video back to the server, you transmit only the analyzed result to the server. By doing that, you reduce all the costs and it become much easier problem to scale when you're trying to scale from the -- a million camera to the billion unit camera. And with that, that's why I think -- although we just start talking about computer vision, but computer vision in surveillance market has been there for a long, long time. It's just never been on the edge. In terms of -- I think that -- the use case is clear, right? In fact, for the policemen, they want to track down the license plate of a car of -- or for a smart home, you want to make sure that whoever take your your package from your door, and you can chase back. All those things that become a useful application that can be used in a security camera. We -- and I also believe we're going to start identifying more and more different application on those cameras to help people provide a safer environment.

Yes. Following up on that. So when you're pushing CV to the edge, it seems to be a lot more about the software-hardware integration versus just hardware. If you look at Mobileye, they take really a platform approach and whereas using to partner more with the OEMs. So can you talk about that? As you're trying to push CV to the edge, how important software is? And how your solution is different from maybe a platform approach taken by Mobileye?

Right. So in fact, that's a hugely important question because, software, when we release a video product, we basically provide all the software, maybe even to application level. But however, then all of our OEMs complain that they have no differentiation. But in the CV case, it become even more difficult because everyone, all the OEMS, our customer, like it doesn't matter if it's Nest or Hikvision or Dahua or Vision, they all want to control their own computer vision algorithm, but they don't want to spend time to understand how to do video processing or compression or update the -- all of the baseline software on the camera. So we are providing still a platform, but we focus on an area that our customer doesn't want differentiation. So we're providing a pure video processing SDK. We can easily connect to different sensor product and give them -- always and helping them to do a streaming. We do all the video processing for them, but we leave the computer vision algorithm open. Meaning, we have our own algorithm. We can show them our algorithm, but we help our customer port their algorithm onto our platform easily. And so they can differentiate. They can claim that our algorithm is better than our competitors so that they can charge more money. And with that, they are willing to use our solution, which is totally different than the Mobileye. Mobileye, I have all the things, everything. You just buy from me as a black box. And the biggest complaint you heard from the old auto OEMs is they cannot differentiate and they don't like it. So I think that our flexibility, we know where to draw the line on the software side that our customer, they don't want to put their R&D on the software. And then we do it. For that, our customer really want to put the R&D dollar, and we allow -- we help them, allow them to differentiate. That's our strategy.

Got it. Clear. I'll turn it back to Nick.

Yes. Thank you, Tim. So I guess, looking at some of the emerging opportunities, I just wanted to go back a little bit to robotics you talked about earlier. When do you see the most promising application verticals and use cases in robotics? I mean, obviously, we can think of many for your technology, but I'm quite curious to get your perspective on looking 5 years out, where do you think you could see the most promising avenues for the company?

Well, like I said, if we accept the argument that autonomous driving is a robot, then that is the most popular one and probably going to give us a lot of benefit. But I don't think that's what you're asking. I really think, on the industrial side, I think that smart manufacturing, including moving things around, people using robot to move equipment or products, we think manufacturing site automotively, and all of that, the smart manufacturing environment, which is, I think, is a huge opportunity. Today, it's people putting pieces in there. I think, eventually, people need to figure out how to using a smart design and going through a flow design that no person interference in that smart manufacturing process. I think that's one big important thing. I also believe, at home, there is a huge opportunity, too. I look at the value -- the vacuum cleaner market, which is already a big business, but I think that's just at a very early stage of the robots at home. I think we can imagine that for the -- for people to help the seniors or help providing better service in moving things, there are things that we can do with the computer vision at home in a much smarter way. But it will take time. So I really think that -- I look at -- I want to spend more time on -- more of our business development time to understand the need on the smart manufacturing and smart home. And hopefully, that we can figure out where that's -- that are the most important application that will drive these 2 directions.

Right. So I mean, effectively, it seems to be driving us to 2 key application markets. I mean, one, we could call smart home and care as well. It's not quite health care, but personal care, effectively. We talked about, for instance, you being able to see what your grandparents are doing, right, remotely, if you're in the U.S. and in Taiwan. The -- and the other one is more industrial applications, right, for warehouses, smart warehouses, sensing, accounting, et cetera. And we could have to this access control, smart locks. I mean, it's already an existing market. And obviously, in Asia, more developed perhaps than in the western world. But there's probably a lot to go. So I guess a key question when I think about all that is, how -- from your end, do you manage to do a push and pull marketing-wise to reach all those markets, right? Because the channels for those are very fragmented, to the exception perhaps of a few large appliances companies like Edge, Samsung, et cetera. So how do you find the leverage basically to reach those channels?

So that's definitely an important task for us. For automotive, it's easy. We're only building -- we are building a strong team on the business on-site to approach customers, which is -- because the market is well understood and big enough so we can justify it. So on the other markets that -- which is now as well defined, we have to really focus on the areas that we -- where we think that is going to be happening soon. For example, you mentioned that smart access control, which is something we believe there's a huge potential. And we put our own business development resource and working with some component. And we work with some really good partners, which are providing a structure light. And they help us to do -- introduce us to their customer, and we help them to introduce our customer to them. So like you said, when there's a market, it's really there's a momentum. It's not only that. As soon as we approach a customer, the customer say, great, this is exactly what we want. Then they are kicking off product immediately. I think that's the market we're going to focus more. And there are also market that we keep trying to go in there. Like you said, because there are only a couple of a big customer working there, it will take time to get in there. So we are -- our approach is more trying to identify and bet on the market that we think there's a short-term return and continue to work on those much longer-term market with business development.

Great. And I guess maybe last question related to about is, you talked about this design win pipeline of $600 million, I think, revenues, $200 million secured as design wins versus design in a quarter ago. So where are you now for the secured design wins? And how do we think about the time line for this pipeline to develop and materialize over the next few years?

Right. So I just want to be clear that for that $600 million pipeline, $400 million are secured. When we say secured, we received the letter from the customer telling us that they kick off the project with us, and they tell -- they even give us their forecasts in terms of the unit number they require from the beginning of project to the end of the project by year. And also, we have engineering resources working on it. And then we know we have secured that design win, and then we can -- but, however, we didn't use their forecasts in putting to our pipeline. We have a judgment on that based on the different multiplier and maybe push out the schedule a little bit, maybe reduce the forecast a little bit. So that's how we come to the design win. From -- for the second part of the $200 million is the engagement that we are going further along enough. And we have engagement, we stand for maybe even 12 months and that we understand the structure. We are -- we have some confidence that we might get design wins, and so that -- we will use that as the second half. But we definitely have a multiplier, which is more aggressively to reduce the potential returns in those. For example, the potential -- the possible design wins, the size of the design win and the time lines design win, then we use a multiplier to reduce the expectation. So that's how we generate the pipeline. I think this is very important exercise that show you not only just what we have won, but also we think that we show you more things that we think that's in the pipeline in the -- that we are cooking. This combined $600 million really is giving a number that our investor can track because, as you know, automotive is such a long-term development. And you cannot wait until revenue -- you start receiving revenue, then you can start talking to investor about your progress. It's too long. So we have to figure our way that the -- our other competitors are using or other automotive semiconductor company are using this method to communicate to the investor about their progress, the progress they're making. And I think that's a fair way to do it, and we try to do it conservatively, but we still think it's an important exercise to show our investor that our confidence on our progress.

Great. And I think with that, Fermi, we're reaching then our time. So Tim and I would like to thank you very much as well as Casey and Louis for being here today. And hopefully, we'll speak to you very soon. Thanks, everybody, for dialing in.

Thank you.

Thank you.

Thank you. Bye-bye.

Thank you. Appreciate it.