We're good? Okay. Thanks, everybody. We're gonna get started with our next presentation from Mobileye, Dan Galves, Chief Communications Officer. I'm Adam Jonas, for those that don't know me. I run Morgan Stanley's Global Auto and Shared Mobility team, and we had the privilege of covering Mobileye in the first iteration as a public company.
Yep
Before it got bought by Intel. And then, and then, of course, since the IPO, the second iteration, a much more exciting time. This really is, in my opinion, it's, it's a strategically important company in a strategically important sector of computer vision, which I think has relevance, not just in auto, but in anything with a camera that makes decisions, whether that, whether that, box moves or doesn't move. No debt, positive free cash flow, dominant share in ADAS. Dan, thanks for spending some time with us.
We don't say dominant.
Okay, don't say.
Leading.
That's not a good word at a tech conference. Don't say dominant. Although 70% market share, well, you know, you can come to your own conclusion on front-facing camera. Dan, any messages just at the top here for the investment community?
Yeah, no, thanks, Adam. It's great to be here. It's great to be working with you again. I was also part of the first iteration of Mobileye and then took a break while we were inside Intel, and now really happy to be back for the last couple of years. So, you know, most of you know the basics around Mobileye. For many years, we've had a leading position, software and hardware, that supports driving assists safety systems. This business is very profitable, and this kinda core business is able to fund the approximately 80% of our R&D that's related to advanced products that are still low in volume, and leaves a lot of free cash flow beyond that.
Some of these advanced products, like SuperVision, are already in production and represented about 6% of our revenue last year on 0.2% of the volume. The power of these solutions is that they carry selling prices that are 20 to 60 times higher than our current high-volume products, and this has been showing up in the annual bookings that we report. For example, in each of the last two years, we've been awarded design wins from automakers that project to around 60 million units of future volume.
We actually shipped 37 million in 2023, and these design wins have, the last two years, projected to around $7 billion of future revenue in each year. Our actual revenue in 2023 was $2 billion. So you can see the growth potential of the business in these bookings numbers. The potential for success of these advanced products has been the main focus of investors since we've been public again. The confidence level of the market can move up and down based on the pace of design wins, based on other macro factors, and clearly, we're at a bit of a low point right now.
But nothing has changed in our view that the market for hands-free products will develop into a very large new TAM, that OEMs are bullish on this market, and we have the right technology, the right cost, and the right OEM relationships to take a leading position. In addition to that dynamic, the excess inventory issue that we became aware of very late in 2023 is causing a significant temporary reduction in our volumes. That's also very much in focus for investors right now.
This issue, which caused us to guide well below consensus for 2024, has hurt our credibility, and it's a very high priority across the company to regain this through execution and taking steps to ensure it won't happen again. I'd like to take the opportunity to explain a bit more of what happened and then give an update on where we stand. We ship product into 300 to 400 vehicle models globally. Those are produced by 30+ OEMs, but all of this product is sold through 5 or 6 Tier 1 auto suppliers.
Over time, we came to over-rely on the Tier 1s to balance their orders versus end-user demand, and they've done a very good job for 10+ years because we never had any type of, kind of, excess inventory buildup in the past. During the supply chain crisis, the dynamic changed, and some of these customers began to stockpile chips to make sure that they had some buffer in case our supply became tight. The good news is that we do have the ability to analyze shipments versus vehicle production, as each chip that we ship is coded to a specific vehicle platform, and we know how many units of each vehicle platform have been produced in a certain period.
We actually do have very good visibility, as it's not like we're shipping to distributors for different end markets who are in charge of finding demand or finding volume. We know what OEM and cars each chip is going to, and projections of how many of those cars will be produced in any given period is generally pretty accurate. We've changed our internal process to do this shipment-to-demand matching analysis on a regular cadence, both on a look-back and a look-forward basis, as well as other internal process changes to prevent any reoccurrence.
I'm not going to comment broadly on Q1 or the full-year outlook, as we'll provide a broad update when we report Q1 earnings in late April. But given the understandable focus on the pace of EyeQ volumes and excess inventory digestion, I can provide a brief update on that. On our Q4 call, we said that we expected Q1 EyeQ volumes to be approximately 3.4 million units, and that we would see an increase of at least 100% in Q2 versus Q1, and at that point, the vast majority of the excess inventory would be cleared out.
Based on our current view, which includes a significant number of customer commitments we've received for Q2, we believe we are on track to meet that outlook and those statements, and that we continue to expect volumes to get back to normalized levels in the second half of the year. Thanks.
Thank you. I hope you can put the phone away now.
Yeah.
That was a lot. Thanks, thanks for that, and it answered a couple of my questions. I was gonna ask, you know, what steps have you taken to make sure this doesn't happen again? Because it does seem like there were some internal changes in internal processes. I don't know if you want to elaborate on exactly what changes were made, but how much of it also reflects just that the customers that you're selling into are getting information from OEMs that are, frankly, in a chaotic environment, where they might have blinked or bet on their own ability to bring forward software-defined and electric vehicles as like sockets, if you will, for your-
Yeah
systems, and then that just got totally torn apart. So I don't know how you... You know, maybe tell-
I mean,
us what's within your control that you've changed
Yeah
... versus the reality of the industry that you sell into.
So I think we can separate the question into kinda core ADAS products and supervision, as well as kinda near term and long term. So I think for the core ADAS products, there's no kind of overexposure to EVs really, or anything else. In fact, we're probably underexposed because we don't have business with Tesla. And obviously, I think in the near term, you have better visibility on what's gonna be produced. You know, you have very good visibility on what has been produced, which makes, you know, the fact that this happened so frustrating because we could have seen it, right?
And I think I kind of explained that, and internal processes, besides like, you know, doing a much more regular kind of view of shipments versus production, is something that is fairly easy to do. We just weren't doing enough of it.
Mm-hmm.
And I think also, you know, changes within kind of the finance department to kind of spread the finance through the organization to ensure that they're kinda hearing the same things that sales or R&D are hearing as well. You know, it, it gives us a better ability to forecast, and now that we're a public company. I think, you know, in the long term, obviously things, you know... You, you have less visibility on kind of, you know, what specific vehicles are gonna be produced next year and the year after. And so I think we can, you know, if you, if you wanna talk more about-
Yeah
-supervision exposure-
Sure
-we can.
Well, yeah, let's move to SuperVision. I appreciate the update on the EyeQ system on a chip update that you're reiterating, the guide from before.
Yeah.
Without, with being true to you, you don't wanna break new information, but I, I didn't know if you were also on track for... Could you reiterate on trackness, if you will, of the supervision side, after the-
So I think SuperVision today is more exposed to EVs because that's who the customers are, right? The customers that we are in production with now are Zeekr, which is a start-up EV-
Mm-hmm
-maker, you know, majority owned by Geely. In China, they're a pure EV company. Polestar is our other kind of notable customer, that's in production this year. They're a pure EV company.
Also Geely.
Yeah, also owned primarily by Geely. So, you know, I think, you know, so far so good on the year. I think, you know, there are forecast of, you know, our guidance of 175 to 195 thousand units is significant growth over the full year last year, but it's not very significant growth over the fourth quarter run rate. We have additional, you know, Polestar was kind of very low, you know, not even material in Q4. That product is now being delivered in China. You know, we believe they're on track to start delivering to customers, you know, July, August, in Europe and the U.S.
Mm-hmm.
So that should be incremental volume. You know, there's a Smart vehicle. So Zeekr, I think they're, you know, January, February, were up, like, 18% year-over-year, on a combined basis against low comps, but tracking really well, and they just came out with a refreshed version of the Zeekr 001 that, you know, significantly higher range, significantly, you know, better charging times. And, you know, what we're hearing is that that's generated a lot of really good kind of pre-order activity, that refreshed vehicle, and probably held down February sales because of, you know, people kinda knew it was coming.
So SuperVision's been mostly a China story up to now, or entirely in terms of vehicles and production. What's taking the Western guys so long to get over the hump? I suspect if we had this conversation about the wins in your pipeline a year ago, you would've anticipated there would've been a bit higher cadence of announcements and/or, you know, moving forward. Is it... What's the issue?
I think the, you know, last year was actually a very kind of positive year in terms of business development for these advanced products. You know, we, we got a big win with FAW that looks like their, you know, every new car they produce under the Hongqi brand, which is like a 350,000 unit a year brand, is gonna have supervision starting, you know, Q4 this year. So that should generate a lot of significant growth in 2025, but China only, although they have ambitions for export.
Mm-hmm.
The Western OE that we haven't specifically named yet, that we announced, you know, in early January, was a huge win for us, something that we've, you know, worked on for 15 months. Seventeen additional models to have SuperVision and Chauffeur over time, starting in early 2026. The majority of the models are actually internal combustion models, so it's kind of very well-balanced between EV and ICE. And it really shows an automaker that's looking to kind of align their kind of future technology in this field with our product portfolio.
I think that there was, you know, the other thing that was positive in 2023 is, you know, we started the year with, I would say, 3 kind of good prospects, you know, including Geely. We talked about Volkswagen Group, we talked about Ford. I think that, you know, over the course of the year, that list of what we would call advanced discussions grew from three to 11.
Mm-hmm.
We haven't been able to kind of close a deal with Ford yet, and we're still, you know, uncertain about that. But the remainder of those, you know, prospects really came about in 2023 and are moving fairly fast. So 15 months for, you know, the Western OEM to close, I think we expect, you know, a little bit shorter time period for some of these others, but some of the discussions only began around, you know, midyear last year.
Dan, is China EV domination positive or negative for Mobileye?
Yeah, it's a good question. So I think that, you know, beyond kind of advantages that the Chinese OEMs have in EV costs and kind of access to supply and things like that, they're also companies that are extremely bullish on the types of products that we are looking to develop, right, or are developing. So last year in China, there were over 700,000 units produced with, you know, let's call it FSD-like-
Yeah
features, right? You know, get in the car, put in the address you're going, and the car should be able to take care of, you know, 90% plus of that driving on their own. Automatic lane changes, kind of understanding where the exit is, positioning you for that. Those 700,000 units were by four automakers: Zeekr, Li Auto, XPENG, and NIO. And that kind of consumer demand and kind of the amount of press and the amount of attention this product set is getting accelerated the development and the desire by other Chinese OEMs.
Mm-hmm.
It also, those Chinese OEMs are clearly looking to export and enter Europe in the near term. So this created... The fact that this market started to develop in China in 2023, created a sense of urgency on the part of Western OEMs. And, you know, by Western, I'm talking about non-Chinese.
Mm-hmm.
Right? So I think that that sense of urgency is something important for us because, you know, you know, the OEMs are, you know, by adopting a supervision type of technology, you're adding, you know, $2,000 of cost to the car.
Mm-hmm.
So you want to, you know, have a value prop for that. The value prop can be competing, like keeping up. It can also be safety-related, it can be convenience-related. And, you know, so we feel like that sense of urgency was never really, we always thought Tesla would be the one to kind of push that sense of urgency.
Mm-hmm.
I think in some ways, the OEMs sort of discounted a bit, kind of the power of FSD.
Yeah.
But the Chinese, it seems like that's really kind of driven a sense of urgency.
So from my discussions with OEMs and also investors, the perception, if not the reality, is that within the China EV market, there's BYD, and then there's everybody else. What is BYD's solution that you're observing and your intelligence is bringing you? How is BYD approaching Level One, Level Two, Level Two+, ADAS, and what opportunities could that bring for Mobileye?
Yeah. So I think that, you know, it's not necessarily specific to BYD, but, you know, Chinese OEMs do not want single source for anything.
Mm-hmm.
It's really part of the DNA. And, you know, even in the ADAS business, like, most of our kind of main non-Chinese customers were, you know, 90% plus of their ADAS business, if not a 100%. In China, you know, it's a mix. You know, with BYD, we're maybe 30% to 40% of their ADAS business. With SAIC, maybe it's 60%. With Chery, it's like 80 or 90. With Changan, it's zero, right? And so this kind of dual sourcing, kind of principle is also present in hands-free systems, where the Chinese customers that we're close to, you know, are all developing software internally-
Mm-hmm
... as well as looking for supplied solutions, right? And I think that the advantage of the supplied solution from somebody like Mobileye is the cost is lower, the probability of success is probably higher. We're already in production. The time to market is very clear, and the kind of ability to certify outside of China-
Mm-hmm
... is also much higher, right? So I, I think that this is as well as a path to eyes off, because the internally developed systems, you know, they're based on data from China only, right? So I, I feel like we have, you know, very good prospects to you know, penetrate into more Chinese OEMs over the course of this year. But we're also going to, you know, have internal competition-
Mm-hmm
... between us and the internally developed system, like is happening with Zeekr, and we welcome that. Really, it keeps us, you know, pushing forward fast.
Yeah. Let's talk about that a little bit. From my discussions with you in the past, you've acknowledged, you know, China ADAS competition is good, it's improving all the time, but not quite up yet for Western export market standards.
Yeah.
And that could be debated. I mean, can that be debated? And I'm curious, what is it about it? Is it because if it's not up to standard, let's say, in the eyes of maybe a regulator or a Western OEM, is it because of the tech or national security concern? And again, what's Mobileye's value proposition? Is it more on the tech side for certification, or is it-
Mm-hmm
... this trusted, let's say, geopolitical buffer, if you will?
Yeah. So I think, Horizon Robotics is our main competitor-
Mm-hmm
... on ADAS systems, kind of basic ADAS systems-
They should
... in China. So their pitch is basically 80% of the price for 80% of the performance, right?
Okay.
When we benchmark the actual performance, which we've done, which I've done-
Should be 20% of the price for 80% of the performance.
It's probably 80% of the price for 40% to 50% of the performance.
Okay.
But they have a good position in kind of low end. You know, there's been a lot of lower priced Chinese cars over the last few years that have gone from zero ADAS to some ADAS, and it's a bit of a check-the-box thing. And so, you know, this is where they're having some success. Again, it's not like they're taking business from us, it's business that really nobody had because it was... They didn't have ADAS on these cars.
Mm-hmm.
I think that the Chinese OEMs recognize that this kind of, you know, 40% to 50% of the performance would never meet standards outside of China. And I think in a more general point, it's like, you know, good enough in this area is not really good enough.
Mm-hmm.
You know, we just, there was news out last week about a recall of a front-facing camera system, where one of our kind of core customers had given a small program to, you know, not a Chinese, but a Western competitor, and now there's a false braking recall.
I believe it was Qualcomm, but we don't need to...
You know, I think that this-
Mm-hmm
... this is pretty clear to the automakers that, you know, you, you need high-level performance when it comes-
No, it's not Qualcomm, by the way. Sorry, it was Veoneer. Excuse me, just for the record, just for the record, Veoneer. Sorry.
Correct.
Continue.
Yeah, and I mean, it's-
Mm-hmm
... you know, I think it's just a better example that, you know, it's, you know, maybe you save a little bit or maybe you feel better because you have a second source, but if it's gonna cause a recall, then, you know, kind of all bets are off.
Okay. Dan, we've had this debate, and I wanna open it up. I'm gonna kind of prompt the audience now. Please, think of a couple of questions, take some pressure off me. It's always. You're gonna ask better questions than me anyway, and we got 20 minutes left here. But Dan, I'm gonna go into the EV debate. We've had this debate for a couple of years now. We're gonna have it right now. A mini version of it right now.
Let's go.
EVs are slowing. Is this good or bad for Mobileye?
I don't think it's good or bad. I think that there's nothing... Yeah, I think it's an important point. There's nothing about supervision, there's nothing about our products that work better in an EV versus an internal combustion. Now, to go beyond that, I think that what we. So I think OTA is kind of the critical point here. Over-the-air updates, right?
Mm-hmm.
We could certainly put a SuperVision product in a car that doesn't have OTA, but it wouldn't make a lot of sense, 'cause this is the kind of thing where, you know, the car has a specific driving style, driving experience. The car, the system is going to want to expand the road types, the type of situations it can handle over time, and you don't wanna put that system into a car and then not have the ability to execute those improvements while the car is on the road. Now, I think that OTA is something that doesn't require a complete tear down of the electrical architecture of a vehicle.
You know, it requires a modem, it requires cybersecurity, it requires a process from the OEM, it requires good, you know, wiring and kind of communication protocols, and the development of vehicle architectures, whether it's EV or ICE, OTA is in the plans.
Okay.
There's already a lot of cars that have OTA-
Right
... that are ICE cars.
Let me, let me kind of interject here. I get what you're saying in theory. In theory, there's no reason why a mechanical thermal car-
Mm-hmm
... a thermal classic, as I like to say, which might be on its last one or two generations, couldn't be rearchitected for OTA and a software, and a software-defined vehicle architecture that could, again, be a socket for supervision. But correct me if I'm wrong, as of, you know, March 2024, there are currently no full self-driving and/or supervision equivalent vehicles on offer that are internal combustion. They are 100% EV. Am I wrong? Today.
I think that that's, that's right.
Okay.
It's certainly right from our perspective.
So my
From our product
... where I'm going with this is, having covered legacy car companies for three decades and seeing the enormous pain that they've been facing with software, on even basic software from a clean sheet approach, going to a thermal classic vehicle and then re-architecting the software for, for the mechanical steering with the precision, the precision, Dan, required to do turn-by-turn navigation on public roads, man, that just is this, tell me, how did that happen?
I don't think that's the perspective of our customers that a tear down is required, right? I mean, I think you have new architectures that are coming, like the Western OEM. Like I said, you know, more than half of these 17 models are ICE cars-
Mm-hmm.
Because they have a new architecture that's targeted for ICE cars. Well, you know, will the software beyond... And we kinda control the software of our system.
Mm-hmm.
You know, I think that there's not a need for major changes to kind of the steering systems and things like this. But I will say that, you know, these types of new technologies or OEMs want to put these new types of technologies on cars that they need to work, right, that they need to sell.
Mm-hmm.
I think, you know, over the last few years, kind of the highest profile products in the portfolio of automakers have been EVs.
Mm-hmm.
Right? 'Cause they need them to work. They want them to be to stand out in the marketplace.
And this also reflects the type of buyer that they would have, someone that might... I mean, you can do your own surveys, but my wife doesn't care at all about Full Self-Driving. She doesn't, she really doesn't-
Mm-hmm.
You know, maybe on a highway, keeping a safe distance in front, that's fine, but the whole version 12 Tesla, she couldn't care less. Others, maybe the type of genre of customer, maybe Morgan Stanley clients that would be in the market for a really advanced EV, could be geeky enough to really want this.
Yeah
and appreciate the value it adds.
I mean, we don't-
There could be some psychographic overlap as well on a marginal buyer. Does that-
I think that that's something that's changing, too.
Uh-huh
... where we don't want this to be, like, a tech gizmo product.
Parler.
Like, we don't want this, that to be kind of the value prop-
Mm-hmm
... of supervision. No, the value prop of supervision is: it's cool, and, you know, if you have an hour commute on the highway, you don't have to do anything. And, you know, I think that there's a lot of evidence that you would get to work more refreshed, less fatigued.
No question about it.
But I think putting the high-resolution cameras on the side of the car, and you can see this with Tesla, they've benefited hugely from this. If you're, you know, going through an intersection and somebody's about to run a red light and T-bone you, if you don't have high-resolution cameras on the side, nothing is gonna give the car the confidence to stop short and-
Right
you know, not enter the intersection.
Right.
When you add those high-resolution cameras, that is very viable.
Okay.
There's also other, many other evasive maneuvers, and we and our customers need to do a better job of kinda highlighting the safety benefits.
Okay. I wanna go to the audience for just one, one more quick one from me. Are robotaxis dead? 'Cause it sure looks like they are. I was talking to... So Apple canceled the program, the stock went up.
Yeah.
GM, GM, you can read about what's going on there. But the stock seems to go up when GM pulls back on Cruise. But I was talking to a client yesterday saying, "Waymo seems to be expanding." I'm like: You know, the expansion comes before the pause. What's your perspective on this?
So I think investors don't like business models that have no certainty around kind of the path to first revenue, right? And I think when you focus all your efforts on a car with no steering wheel and no pedals, you have nothing until you can, with confidence-
Nothing until everything
... completely remove the driver.
Nothing until everything.
Right. Until you have regulatory approval, until you have consumer trust, like, you have nothing.
Mm-hmm.
And so this is why Mobileye never put our eggs in that basket, right? Where we always kinda tailored our technology to be able to generate revenue before you can go to that-
Mm-hmm
extreme part of the spectrum.
Thank you. Bob, I know you're a huge robotaxi pupil. Any questions in the audience for Dan? All right, one over here. Thanks.
Can you talk a bit about the margin structure?
Thank you. So could you repeat that-
The margin
Sorry. Oh, the margin structure?
Margin structure?
Yeah, both... Thanks, now it's working. Both gross and operating, right? So gross has continued to trend upward nicely, but of course, you've taken a hit on operating margin.
Yeah.
Does it mean anything for sort of the longer-term margin trajectory, and where would you think operating margins to pan out in the long term?
Yeah. So, during our time inside Intel, you know, this was a time to really kinda ramp up the development of these new solutions, and it led to a significant increase in R&D expense. Really none of that incremental R&D expense was associated with the kind of the core ADAS business, but it was in anticipation of, you know, the hands-free market developing and us having a full spectrum of solutions there. So I think that, you know, we've said that our operating expenses will grow 25% this year from a level last year that was much lower than we expected it to be for a variety of reasons.
And then we believe the operating expenses, you know, beyond that, the growth pattern will be much lower than that, and we'll start to see a lot of operating leverage. So, you know, over time, you know, what we've been public about is we expect to have 40% plus operating margins. We haven't put a timeframe on that, but I think that that's, you know, we feel highly confident in that type of number.
Another one from the audience? I wanna ask a technology question. Tesla is, their approach, and I'm oversimplifying it, is huge amounts of emphasis on compute in the data center, the learning computer, whether it's the billions they're spending on NVIDIA GPU clusters, and then their own internal, custom ASIC, efforts.
Mm-hmm
... to get more compute, no HD Maps, and their claim that Version Twelve doesn't no longer does auto labeling.
Mm-hmm.
Mobileye uses HD Maps, says that labeling is critical. Who's right?
Mm-hmm.
Or is it more complicated?
I think I think our team has a very pragmatic approach. You know, for example, with the HD maps, if you can pre-load information about how did the last 10 humans traverse this intersection? What was the exact drivable path? What was the kind of the common, the average speed? Where did they stop at a stop sign? So clearly, you know, if maybe there's a solid line there, but you can't see around the corner, so the common person kind of moves up so they can see, see around the corner. If you could pre-load all that information about the world into a car, it would take a lot of pressure off the real-time sensing.
Mm-hmm.
We have the capability to do that because, you know, we have tens of millions of cars on the road with our cameras inside. Now, the reality is we have 3 million cars collecting data today, but that's, you know, around 50 million kilometers per day of data, so we have a map that covers 90% of the U.S. and Europe, and it's growing very quickly in China.
I think with that map, you would collect as much data in a week or maybe a couple of weeks as Apple has, maybe in the history of their-
For sure
- program.
Right. Now, this is not full video data, right? These are... You know, 'cause it needs to be very light bandwidth to keep communication costs down and to be able to send over cell. But it's- we see it as a huge advantage for us, and it's something that's important to incorporate into the overall architecture of... Because, you know, in reality, what you're trying to do is you're trying to build a very accurate model of the environment around the car, and then you make decisions based on that. So I think the mapping is critical for us.
You know, another area we differ from Tesla is, you know, we, we feel like maybe there's one day where kinda cameras only could get to the accuracy levels to support eyes off, full disengagement, but we don't have a, a great sense of when that time would be. So if you can have two independent perception systems, both capable of driving the vehicle, that gives you real redundancy within the system, and that's part of the path from supervision to Chauffeur. In terms of kind of the end-to-end AI question, I think our view has always been, we need to incorporate the latest techniques in AI into our architecture, but there's no need to, like, over-rotate and completely tear up what you've done because we have a lot of solved problems. Like, vehicle detection, it can always get better, but essentially, it's a solved problem.
You know, we have, you know, 250 petabytes of video data that's been used over the years to train these systems. We have redundancy within the computer vision. So, you know, the deep neural network that we use may not have seen some weird agricultural vehicle along the side of the road. You don't want the system to see nothing.
Mm-hmm.
So we have a geometric modeling module within vehicle detection that essentially anything with mass that's elevated off the road is an object to be avoided. So, you know, I think that there's, you know... We're very careful to never say never.
Okay.
But we feel like kind of end-to-end AI techniques are a great way to get to kind of a comfortable, you know, well-performing system that works 95% of the time.
Mm-hmm.
But 95% of the time is not gonna be good enough if you ever wanna get to kind of a system where the driver can disengage.
Well, and I've asked. I've on a slightly different topic here on tech, I've asked Amnon this before. I'll ask you again. My understanding is that you don't really rely on NVIDIA, or you're not buying as many GPU clusters as you can, or otherwise renting them to train a neural network. Why is that? Could that change? Is there a reason... Explain to the audience why that's not necessary for you-
Yeah
... when so many other companies collecting monumental amounts of vision and other ambient data-
Yeah
do rely on such technologies?
So I think like a third-50% of our CapEx is buying NVIDIA clusters, you know, for our kind of on-prem data-
A third to 50-
1/3 to 50%. Now, that's it's a very small number compared to what others are talking about.
Sure. Dollar terms, yes.
Then we have, you know, we spend that type of dollar on off-prem, you know, on cloud-
Mm-hmm
... data services. This is kinda what we use to, for kind of training, processing power.... I think that, you know, we're not talking about, you know, completely training the system through pure neural network training. I think, like I said before, we've got a lot of solved problems-
Right.
So we can, you know, if we're having trouble with some type of object or some type of shape, we can query our database, which is, you know, I think last maybe seven or eight times higher than what Tesla has in terms of full video. We can search that database, find examples of what we're having trouble with-
Mm-hmm.
and kind of train the system using specific
Mm-hmm
video, not just, like, all of the world's video.
I just kinda wrapping up here. You're a former sell-side analyst. So, you know, I'm not asking you necessarily to play that role right now, but what are... In your, in your opinion, what are investors missing or mismodeling about Mobileye, and, and how do you value- how would you value this company?
Yeah, I think at this point, nothing is being mismodeled, right? I think everybody is gonna have, you know—I think for the next few years, we feel quite comfortable with where consensus is, and consensus has come down a lot, right? And, you know, now we feel very achievable. I think that, you know, the way I would value it is by sizing what this market could become, and I think by the end of the decade, you know, the numbers I kinda hear, you know, maybe 10% of cars produced are eyes on and 10% eyes off, and, you know, put an ASP or a, you know, content per vehicle on that number, and then make a judgment on what our share will be.
And then you have to kind of probability weight that 20 to 30 number. I think, you know, the numbers can get extremely large, but I think that, you know, that probability factor is based on, is the market gonna develop, and does Mobileye have the right technology? I think that those are, you know, questions that I'm very willing to talk through with people one-on-one.
Yeah.
But there's not, you know, 100% clarity right now.
I appreciate that. Well, you know, Dan, I was supposed to be... I was originally planning on being in Jerusalem this week, and I was gonna be visiting with Amnon as well, but, you know, your being here and being generous with your time, I, I postponed that one. So we will get to Jerusalem. We will come back. It's gonna happen this year.
That's awesome.
I do appreciate you spending time with us, and, and I know you're heading there next week.
Next week.
Safe travels.
Yeah. Thanks a lot, Adam.
Thank you so much.
Yeah. Thanks, everybody.