All right. Hitting the home stretch of the forum, but you know, really, really excited to have our next guest. Look, longtime friend of the team, buy side, sell side. We've known each other-
Everywhere.
for a long time. The Chief Communications Officer at Mobileye, Dan Galves. Dan's responsible for overseeing the company's investor relations, comms, corporate strategy efforts, but given his sort of history in you know auto, sitting in my seat as well, he's always been very insightful, and he does a great job of putting things into the context that you as investors want to hear. So with that introduction, Dan, thanks so much for being here today.
Thanks, Chris.
We always do this sort of overview in your own words, and the way I would talk about it is, you know, I think as we have discussed, and you'd be probably, you know, one of the first to admit, this has been a little bit of a year to forget. Right? For-
If I could only forget.
And, you know, there's been issues on the inventory build first, on SuperVision. There was, you know, some of the production cuts, and everything in China.
Yeah
... has come so fast for Tier 1s, and that obviously that leads into Tier 2s, and you know sat on a prominent platform in Zeekr. So maybe the first question is, how should we think about, I call it a reset-
Mm-hmm
... over the next couple of years? And then we'll talk about some of the technology that's obviously exciting that we're gonna get updates on-
Yeah
in the next couple of months.
No, that sounds good. I think if we, you know, if we take a step back and think of kinda what our, you know, what our business is today and kinda where we're trying to get it to, you know, the last few years and kinda through next year are really kind of, you know, phase one of a growth story into advanced products. So I think, I think the growth story of Mobileye going forward is going to be more sophisticated technology, which requires more compute, more software, and generates more revenue per car for Mobileye as the car starts taking over more and more of the driving.
Mm-hmm.
And I think phase one, you know, is kind of the strategic objectives are, you know, kinda number one, you know, maintain our kinda core ADAS business, maintain profitability in that business. Number two is acquire, you know, deepen the relationships with our core customers in terms of getting design wins for these advanced products. And then, you know, another strategic objective is to execute the EyeQ6-based programs that we have, which, you know, the focus right now is on VW Group, to launch a SuperVision product on a number of vehicles in 2026, and Chauffeur in, you know, late 2026, early 2027. And these are the objectives of the company.
And then phase two, which is really kinda 2027 to 2030, is if we do our job during phase one, then we'll be able to scale these products in that time period, and it creates, you know, a lot of opportunities for investors.
Across the board, the broader portfolio of VW, for example.
Across the broader portfolio of VW, and, you know, the likelihood is many more customers as well. So I think what we'd like to see is, you know, our North Star is autonomous vehicle technology, but the way that we build the architecture is such that we can use modules or building blocks of kinda the complete stack to create products along the way. You know, everything from Surround ADAS, which is kind of a multi-sensor, highway, hands-free system that also helps automakers meet, you know, the much more difficult regulatory requirements at the end of the decade. Supervision is an eyes-on, hands-free, you know, that operates in all environments. Chauffeur is a, you know, an eyes-off that most likely starts mostly on highways-
Mm-hmm
... you know, at the start. And then you have Drive, which is the complete self-driving system-
Mm
... that would turn into kind of a robot, robotaxi service. So this is the portfolio. It's all kinda built across the same backbone, and I think the optimal kind of outcome for Mobileye in the 2027 to 2030 time period is that you have multiple automakers that, you know, maybe 10% of their volume is SuperVision, you know, 30% is Surround ADAS-
Yep
... the rest is, you know, the core ADAS that we have today, and then there's pieces that are Chauffeur that, you know, would tend to scale up towards the end of the decade, but that, that's really kinda what we're working towards. If you look at kinda what's happened, so that, that's kind of the goal. That's really the North Star of the company. I think it's kinda helpful to take a step back also and look at what's happened over the last two years in terms of the positives and negatives since the IPO, 'cause there's been a lot of changes, which you mentioned a lot of. I think, you know, obviously, negative changes tend to get more, you know, attention, but let's start with the positives.
So in terms of, like, the core ADAS business, outside of China, over the last two years, we've won almost 100% of the RFQs that we've participated in, which basically means that across our top ten customers, which account for around 75% of our revenue, we've locked in that ADAS business through 2032, 2033... and there's been no kind of notable new competitors that have kind of pushed us, either on pricing or performance in that area. The regulatory environment, number two, has gotten kind of much more positive for us because kind of the newer, you know, NCAP is kind of the group that creates testing requirements for ADAS systems that results in, depending on how you score, you get a three-star crash rating, four-star, five-star, and this is kind of-
Yeah
... very important for consumer sentiment, you know, to different brands, different vehicles. The kind of new test regime that's gonna be in place for 2028 in Europe, and that tends to kind of, you know, cut across the rest of the developed markets soon after, is such a step change that it's going to require data from more than just a front-facing camera. So that means more sensors, which means more processing power, more sophisticated software, and more kind of revenue per unit for us. So that's been another positive. And then in terms of the SuperVision Chauffeur engagements and Surround ADAS, let's say we've gone from essentially, you know, one kind of high-probability customer in VW Group-
Mm-hmm
... a second one in Ford that didn't work out, to now we have kind of deep engagements with nine out of the ten top customers. So I think that that's actually been very successful in the process of acquiring customers. You know, so I think that that's how I would characterize-
Yeah
... you know, outside of China. Inside China, there's been a ton of headwinds. I think that, you know, on the ADAS side, there really are no clear performance standards in China. There are new local competitors that have come in at lower price and, you know, to kind of create more of a race-to-the-bottom dynamic. We declined to participate, and we've lost some share within the domestic China OEMs. On the kind of advanced product side, there's just a huge kind of push to make this, you know, China-for-China technology. There's a huge push for OEMs to kind of show their software capability, their technology capability. And despite the fact that the systems that are on the road today that compete with Supervision are, you know, twice as expensive and, you know, for the same-
Well, performance
... Relatively the same performance, there's a lot of kind of traction of this. So, you know, in a lot of ways, we're seeing kind of the auto market, you know, the auto industry really more quickly than we expected, turn into two almost completely separate industries, one in China-
Yeah
... and one outside of China.
A bit of a de-globalization. That theme has sort of come up across the day. And to paraphrase, this is mostly like Horizon, for example, on the low end. There are some others that are out there, but Horizon has made some inroads with BYD. And on the upper end, these are typically more expensive, NVIDIA Compute, multiple LiDAR, but an internal driving policy. Is that fair to say?
Exactly. So, you know, the competition on the ADAS side is really all about price, and it's from third parties, so it's not the OEMs, like, creating their own. So yeah, two or three different competitors that have sort of pushed us on price. It kind of kept on falling. It doesn't seem like these companies really care about making money any-
Mm-hmm
...anytime soon. So that's the situation there. You know, we still, the business that remains for domestic China OEMs is kind of primarily focused on export, where you-
Because they have the NCAP, they have to hit it.
Exactly. So you have to hit those standards, so I think that creates kind of some level of leverage or opportunity for us. And then on the upper end, yeah, it's basically combination of camera, radar, LiDAR data on kind of very heavy compute, you know, from a third party like NVIDIA, and the software is developed internally by the OEMs, you know, thousands of people, lots of cost.
Yeah. So, Dan, maybe if we could bridge this near term to that sort of medium term, 2027 to 2030 story. So sort of some quick math that I've put together. So when I think about that short-term path forward, 'cause what I think a lot of people wanna do is use that base of core ADAS, which is maybe, you know, $0.40-$0.45 next year by our numbers, and then it would need something like 200,000+ units in supervision, a couple of million units on ADAS to start to get to the $0.70-$0.75 numbers that are more like 2026.
Mm-hmm.
Can you put some numbers around that? Do we need external wins to hit 2026 numbers, or do you have some visibility, given Audi and FAW, that is sort of in a realm of expectations?
Yeah, good question. Like, I don't wanna talk too specifically about, you know, specific timing.
Sure.
'Cause I think in a lot of ways, you know, kind of the next couple of years, like with SuperVision, like, the intent was that China would create a kind of a growth avenue-
Mm-hmm
... for supervision. I think that's been diminished quite a bit. But I can kind of give a framework, and I think I'd start by saying, like, number one, like, our priority over the next six to 12 months has to be to kind of meet expectations, you know, to-
Yep
... to kind of consistently meet expectations that the market has, and to put up new design wins for the advanced products, like, full stop. Now, it's very encouraging what we're seeing. So, you know, in terms of supervision engagements, you know, there are six or seven that are in, you know, the RFQ process. You know, the timing is not certain by any means, but, you know, these are programs where, you know, things have happened that are very kind of strong signs that there will be a decision, and we're not facing much competition. These RFQs, you know, when you get to scale with the program, are anywhere from 250-300 thousand units per year, per customer.
Per customer.
Right? And that's, you know, if you kind of use some, you know, rough assumptions, that's maybe $0.18-$0.19 per program in terms of-
Mm-hmm
... kind of incremental EPS. Again, like, once you get to scale with the program. And on the Surround ADAS side, we have four RFQs that, you know, are... we're responding with, you know, EyeQ6 High at around $175-
Mm-hmm
... $200 per unit. Now, this is gonna be, you know, replacing a, an ADAS unit.
Yes.
So, you know, if you kind of use incremental-
And net.
You know, yeah, net-net. Well, so each of these are about a million units a year per customer, or averaging out. And that kind of creates maybe $0.08-$0.09 per customer. And I think what's even more encouraging is some of these customers are in both, and so what we're seeing is kind of this segmentation building up, where certain OEMs are looking at, you know, 10% of their business being these kind of aspirational systems, you know, maybe 30%-40% of their business being Surround ADAS, and then the remainder would still be kind of regular-
Yeah
... ADAS or emerging market ADAS.
You, if we were to use that VW analogy, you're gonna put supervision on maybe Audi, but Surround would make sense for VW. Again, that sort of segmentation of the market, 'cause you can't price maybe supervision for-
Yeah
... for the lower-end VW.
Exactly.
Yeah.
If you look at, like, a pure premium automaker, which we're also involved in, it's, you know, the numbers are different. You know, maybe it's 50%-
Yeah
... of their volume is Surround ADAS, and 25% is Supervision.
Yeah.
You know, it's encouraging, but we need to go out and kind of, you know, complete these processes.
Then, Dan, I always hate being like the Wall Street guy on pushing on timeline, but there was always... You know, Amnon used to make the point of, we are what, 99%-
Yeah
... along, and RFQs always slip. Literally, one of my first slides was, you know, as an auto analyst, you're just used to one- to two-year push-outs. Has there been, over the last twelve to eighteen months, things like DXP that have extended timelines, where maybe some of these customers are looking at using supervision across more platforms?
Mm-hmm.
But anything you could say about the timeline, specifically on those SuperVision ones which you've been talking about for years?
Yeah. So it's, I think that there have been, not recently, right, so no change-
Yeah
... to kind of what we've recently said, but, you know, kind of the EV/ICE replanning process was a slowdown in the ability to make decisions on new technologies, right? So from maybe October of last year until, like-
Yeah
... February, March, not the same kind of level of progress with these programs. I think at the same time, you had China, like, you know, what are we gonna do about China?
Yeah.
Also taking up a lot of kind of management airspace. And that did lead to some kind of a you know lack of progress during that period. But recently there's been you know significant progress, and this is not like we're having like a catch-up meeting every two weeks.
Mm-hmm.
These are, you know, specific kind of, like, prototype vehicle expeditions and, you know, demos for this group of executives in this region, demo for this group of executives in that region. You know, start of commercial discussions. So, like, a lot of work goes into these programs, and, you know, I don't wanna kind of commit to timing.
Mm-hmm.
I think that there's enough of them that the you know, the opportunity to have some, you know, positive news by the end of the year is good. But I think the most important is, like, we've seen these programs reach periods where it becomes embarrassing to-
Yeah
... to not move forward.
And even slow-
Like, when-
Even slow legacy OEMs also have their timeline that they want to sort of complete something by a rough date, yeah.
Yeah. I think, you know, 2027 is important. Like, you know, automakers believe that, you know, Tesla will continue to double down on their technology and improve it. And, you know, you have Volkswagen Group, that is gonna be launching these systems in 2026, which creates kind of pressure on the other-
Yep
... OEMs. So there is some level of timing pressure, and, you know, and the stuff that you talked about has had an impact, too. Like, the Volkswagen deal, like, actually took a lot longer than we thought, but it resulted in much higher volumes than they were originally talking about.
And then, Dan, a realistic reset on China, right? Amnon talked about, I think it was late 2025 into 2026, so the chip will be ready in 2026, a lower-end, more simplified chip to compete. But like you said, it's a race to the bottom.
Mm-hmm.
So how do you think about what is a realistic Mobileye business in China over the next, you know, two to five, two to five years?
Yeah. So I think the exposure is already, like, a lot smaller. Like, you know, from our filings in the first half, you'll see that China was about 30% of revenue.
Mm-hmm.
... what was that, right? So I think if you go back to 2023, you know, you had, well, let me just look ahead. So, like, 30% of revenue in the first half, you know, rough numbers within our guidance for the second half, it becomes 15%-
Yeah.
of our revenue.
With Zeekr being a material step-down, right?
Right.
Before. Yep.
If you think about domestic China OEMs, specifically, it's about 4% of our revenue. The exposure is much lower. Now, you know, it's not just this, so in some ways, the 4% is probably more solid than the re-
Yeah.
... the balance of the fifteen, because that re-
It gets blocked.
The balance of the 15 is foreign automakers producing cars in China, using our ADAS, and those companies-
Sales are going lower.
are likely to see.
Yeah
... volume declines. You know, the 4% includes some level of Zeekr volumes, and then about 1 million units per year of ADAS chips to domestic China OEMs, of which 500,000 are for export. So I think if you, you know, our strategy in China is not to try to claw back market share by lowering pricing. Our strategy in China is basically to kind of present to the OEMs, which we have been for a while now anyway, that if you have global ambitions, if you have ambitions to export into developed markets, we can be a really good partner for you, 'cause we can help you get to five-star-
Yeah
... safety ratings, like, in a lot of different markets.
Which is a BYD next three to five years, maybe not the next one to two years, once, like, Hungary and other export facilities are actually in place in Europe.
Exactly. I think that the kind of the export into developed markets from Chinese automakers has gone a lot slower than we expected. And the kind of the market share gains within China has gone a lot faster than we expected, so it's, it hasn't been a great environment for us. But we do see opportunities, and kind of the remaining business in China for ADAS is a lot of export and, you know, with a couple of companies that, you know, have kind of indicated that they want a long-term relationship. So we'll see. I think we're, you know, we're trying to kind of you know, investors should be not expecting, you know, any type of kind of recovery in China.
Yep.
And that's kind of the part of the process that, you know, we feel like we're pretty far through now.
Well, well, well, Dan, one of the things that I've said throughout the day is I play referee, and I always do the good and the bad. I feel like I already hit you with the tough questions. One of the things I wanted to discuss is, you know, the real excitement that will come, whether it be Q7, whatever, I can guess on what the Audi vehicle is in 2026, 2027, where Chauffeur is sort of Mobileye's way of, you know, saying Level 3, Level 4, essentially hands-off and eyes-off.
But what I wanted to dive in, because I don't think everyone here sort of knows the Mobileye framework, I put it up in one of the slides, is I've asked this question to many people over the course of the day: What is the bar-
Mm-hmm
... that an autonomous system needs to be gauged against? And I've gotten answers as, "We don't know. It's vague." Mobileye seems to sort of plant their flag that you want to be five, 10X better than a human driver over, I think, 400 million miles. Can you go into a little bit that framework? Because you are sort of, you know, someone that is, and Amnon has always been very specific that you're not going to go before the technology is ready-
Yeah
... because regulators in society aren't gonna stand for that, you know, when lives are at risk.
Yeah, it's a difficult question. I, I think that we're listening to our customers, which are kind of very clearly saying that for kind of an unsupervised system where the driver is out of the loop, they'll be taking on the liability-
Mm-hmm
... when that is there, and that they don't believe there will be much tolerance for machines causing accidents, and for that reason, that they're setting a very kind of high performance requirement. There is a European regulation that has kind of the words, like, "better than human performance.
Yeah
... in it. Nobody's defined that yet, so it's kind of a very difficult thing to get your hands around because, like, or, you know, what type of a failure are you talking about? A failure that would cause a fender bender, a failure that would cause, like, an injury, a death? You know, the statistics vary based on kind of those metrics. What we've always kind of targeted is a million hours between what I would call, like, a significant failure-
Yeah
... which is, you know, somewhere between 10 and 50 times kind of human performance, the way that our customers are looking at it. And that's really high, right? So the plan, basically, is to have two separate perception systems, each of which is capable of driving the car on its own, that has different failure modes that, you know, like a corner case for one system wouldn't be a corner case for the other.
Yeah.
If you had sun glare, it would affect the one system, it wouldn't affect the other, so fail in different ways is, like, a big philosophy at Mobileye, and so the plan is to get to 1,000 hours between failure for each of those systems, and kind of theoretically, if you have both of them independent, then you could multiply those failure rates together, and you get to about 1 million hours, which is, you know, maybe 30 million miles, something like that. Now, if you know, regulators decide that the bar is lower, or you have kind of competitors that, you know, set the bar at a lower rate.
'Cause they're on the road, yeah.
That would, because they're on the road, that wouldn't be the worst thing for us. But I think ultimately it's gonna be up to our customers, and I think regulators will have a say as well. The other thing to kind of keep in mind, because it's like a lot is gonna depend on how many cars are on the road as well. So it's like if you have a system that's, you know, one failure over, you know, every eighty thousand miles or-
You'll never see it.
... something like that, then, you know, and but you only have one car-
Yeah
... you're probably okay. But if you have a hundred thousand cars, you're gonna have-
Yeah
... you know, three accidents a day.
Yeah, that's one of my slides. Yeah.
Oh, really? Okay, yeah. That's good. So it's, yeah, it really, it depends on a lot of factors. So I think for Waymo, you know, the opportunity to have, like, a small fleet and we're super kind of, you know, impressed with what Waymo's done. 'Cause people like, you know, ask me like: "Well, Waymo's saying this, and, you know, they're saying that." I think for Waymo, it's like, you know, when you have 700 cars on the road-
Exactly
... it's not like you're gonna end up with, you know, a hundred different court cases, you know? But, you know, with a big automaker that's looking to put these systems on the road and tens of thousands, hundreds of thousands-
Yeah
... of units, they have to think a little bit differently.
Can you talk? You know, Tesla talks about Shadow Mode to the idea of this validation, particularly of Supervision into a Chauffeur. Again, not knowing what the exact vehicle, but if we use Audi as an example, do I have the right understanding that a vehicle will be launched with all of the hardware?
Mm-hmm.
Capable of Level 3, Level 4 Chauffeur, but maybe for the first X number of months, that Chauffeur would be running in Shadow Mode?
Yeah.
So that you, you'd only be able to have a supervised system, and then at some point, Audi would say, "For the price of X, you can upgrade, and now you can, you know, theoretically fall asleep in the car.
That's the plan. Like, I've seen kind of the timelines for the Audi Chauffeur program, and yes, like, the launch date is X, launching with Supervision capability, eyes on.
Mm-hmm.
So at launch, you'll be able to drive hands-free, but you'll have to keep your eyes on the road. The system is gonna monitor the performance of the car, you know, in the background, you know, monitoring interventions of the driver, but also kind of like, you can understand failures as they happen, you know, in Shadow Mode, and basically, you wanna build up enough miles. If you have, you know, 30, 40, 50 thousand cars on the road, like, those miles are gonna pile up really fast, so you can kind of get to this, you know, proof point of, okay, like, these systems drove, you know, 8 million hours and had six failures, or something like that. Because, you know, I think that they're... Especially in Europe, I think that the regulators are gonna want proof.
Yeah.
Mm-hmm.
I wanna give ample time for Q&A before I keep on rattling off. Anything from the audience? Don't be shy, don't be shy. It always takes a... Anyone? Neil.
Hi. How's it going?
Hey, good, good. How are you?
Good, good. I wanted to ask about just the LiDAR news, and I think there were several, you know, reasons cited in the press release around why Mobileye is shutting that down, you know, doubling down on Imaging Radar, advancements computer vision.
Mm-hmm.
Can you give a little bit more color on that? Is the view that because of these advancements in other areas, the relative, you know, importance versus a year ago or two years ago-
Yeah
... is LiDAR losing that positioning? Or is it really about, you know, the rest of the ecosystem, other players in the market being at a sufficient level, sufficient cost, where Mobileye doesn't need to, you know, invest in that area anymore? Just any color on that would be helpful.
Yeah. No, it's a good question. I think we're... Yeah, it doesn't mean at all that we're not planning to use LiDAR in these advanced systems like Chauffeur and Drive. I think in a lot of ways, like, our view of LiDAR is kind of always like, what do we need from it, given what we're gonna get from the computer vision and the imaging radar system? And so over the last couple of years, we feel like, you know, Transformer-based architectures have kind of helped robustify our computer vision quite a bit. So kinda what we're seeing in kinda the EyeQ6-based testing for computer vision, with all of the new kind of AI technology within it, is promising, is better than we thought, you know, a couple of years ago, in terms of robustness.
The imaging radar is meeting the performance specs that we were targeting, but now it's meeting them with, like, B samples, so this is much more kind of almost production ready for us. So, you know, that kind of confidence level that it would hit the specs, like, has gone up a lot. So that sort of led us to believe, like, okay, we probably don't need, like, a, an advanced LiDAR, even in the second generation. It's, you know, what time of flight gives us is probably gonna be plenty. And then I think the price gap between, you know, what we had line of sight for our FMCW versus kind of where time of flight is now or looking like it's gonna be in a couple of years, you know, there's always a price-performance trade-off, right?
And the extra performance you get from FMCW, in our view, might not be worth kind of the price gap, like, for, you know, with what we're doing. And we were gonna need to make a decision to kind of move forward on production, which means you've gotta get contract manufacturers and start, you know, putting, like, you know, significantly more dollars behind it. So it turned into an easy decision to make, but I don't think it really says anything about LiDAR specifically.
Dan, is the right context there that so Mobileye has two different OS perception stacks, so camera and radar, LiDAR. So on the radar, LiDAR perception stack, you are getting the velocity vector from the imaging radar, so you don't need it from FMCW. So time of flight, you know, whether it be Luminar or Innoviz, whoever it is-
Yeah
... gives you that information.
Yeah.
Yeah.
Yeah, exactly, and I think like the you know the good LiDAR suppliers have really been performing, too, in terms of kind of performance and, you know, production capability. So, you know, we feel good about our access to the technology from third parties.
Any other questions? One from me. Yeah. Dan, you mentioned. Look, and I'm guilty of this over, you know. I always say I'm not a computer science, not a, you know, not an engineer. Years back, I used to put your approach in sort of this bucket of rules-based.
Mm-hmm.
Right? RSS is sort of a rules-based overlay, a do-not-violate sort of safe space, but I think we've underestimated the amount that you are using, you know, transformer models, Gen AI, but in a hybrid, I think Amnon's called it a compound AI approach.
Yeah.
Do you wanna talk a little bit about how you use Gen AI in driving policy?
Yeah.
And then maybe how it's different than the likes of Wayve or Tesla or Waabi.
I think that we've always. You know, our our CTO is, you know, one of the premier, you know, AI machine learning academics in the world, and Amnon is no slouch either. You know, they've published all kinds of white papers and, you know, patents, and so, you know. And they both maintain kind of an academic position as well, which I think helps keep them up to date with like, you know, everything that's happening on the cutting edge. And so we've always integrated, like, deep neural network, deep learning technology. You know, that's been in the system. I think the first deep neural network technology in a car was in the first Tesla Autopilot-
Mm
... to do a specific task. But our system has always been mixed in with other techniques and also some rules-based systems. So, you know, for example... So I think that this concept of fail in different ways, you know, we've been, you know, detecting vehicles with a deep neural network, you know, machine learning technology for, you know, six, seven years. But we've also, you know, maintained a legacy kind of appearance-based system, where all you... You don't really care if it's a vehicle, a pedestrian, a cyclist. What you care about is it has mass, and it's above the road, and you don't wanna hit it.
You need to stop it.
Right? So I think that this kind of fail in different ways, because, like, you know, the AI more of a, you know, more learning-based technology is, you know, amazing, but if your data set doesn't include what you're seeing, there's a chance that it would be ignored. Now, we used kinda convolutional neural networks to do kind of a lot of the perception stack, you know, from 2016 and through EyeQ5. Now, CNN is great, but you can tell it's taking, like, a lot of brainpower here for me to describe this. But, you know, if you get an image from the camera or a frame from the camera, it's made up of lots of different images, and each one of those images has to be analyzed separately. So, you know, imagine, you know, a view of, like, 50 different vehicles-
Yeah
... on the highway.
It's a lot, yeah.
You have to pixel by pixel. You have to kinda analyze and detect each vehicle separately. You have to stitch the frame together to, like, an environmental model. You have to kind of bring it up from 2D to 3D. There's a lot of steps in the process. Transformers allow you to analyze the entire frame as one, which is a kind of huge positive in terms of efficiency and, you know, reduced complexity. But you still have the same issue of, you know, a data set that needs to be, you know, complete enough to include, you know, every possible scenario you can see out there.
So it still, in our view, requires internal redundancies to have different kind of approaches and different modules that would detect things in different ways, so that you can, you know, boost the performance to super high levels. What we don't wanna be really in the business of is, like, going out and trying to find failures, and then developing more and more data-
More training, yeah.
... to kind of fix those failures, retrain the network, hope that it doesn't, you know, create additional kind of failure modes that you didn't have before. So I think that that's what, you know, we wanna kinda set ourselves up so we have kind of a clear path to the performance requirements that we were going over before, without, like, a lot of uncertainty about how long is it gonna take us to find all these corner cases, retrain the networks? How much data are we gonna need? How much compute are we gonna need? It's... In our view, it's not the right way to do it.
... Last one from me, just the opportunity for parting thoughts. I always, you know, like to push the envelope. Anything that we may should start to think about, topics for maybe the investor day in December? Anything to end on?
Yeah, no, I think that, yeah, it's been a tough year with a lot of different headwinds. I think a lot of it's been focused on China, you know, maybe some of it on things that we're, you know, not really in control of, like our, you know, our parent company. I think what, you know, we as a management team wanna do is kind of get back to basics and sort of, you know, reset, you know, the way we talk about Mobileye in terms of kind of what are our competitive advantages, what are our moats? Kind of what does our business look like? How do you value these deals if they come, right? You know, the framework-
Give us a framework for how to look at an announcement and put it into-
Exactly
... a generic bucket.
I think the analyst day in December is gonna be really, really good because you'll get, like, a kind of a full, like, kind of exposure of our technology stack and the way we're approaching AI, the way we're approaching, you know, mean time between failure. You'll get a lot of information on kind of the progress that we're making with Porsche and Audi. I mean, this is like, it's incredibly... It's the most intense kind of execution phase that Mobileye's ever gone through, you know, us as a Tier 1.
You'll also be the... I think it's public that you'll be the Tier 1, right? I mean-
Yeah, we're gonna be the Tier 1, which creates, you know. I mean, there's 40,000 requirements that we have to meet, you know, with this program. But I think on the other side, like, the Porsche and Audi engineers that have kind of launched complex programs like this many times in the past, like, we're working together to create kind of a platform that can then be scaled to other automakers as well.
Okay.
I think it's, you know, we're kind of building these muscles, executing. Things are really on track, and I think we'll be able to kind of demonstrate that at the day. Some of these frameworks for, you know, how to think about, you know, the business.
Yeah
... looking ahead.
You're always in a rock and a hard place, so you can't really say too much, so it's more like a general announcement. Could be, you know, X, X number of units-
Yeah
... and then come back.
I think automakers are more kind of open to announcements with us now because, you know, we're a tech provider.
Mm-hmm.
You know, and so I think that they kind of understand that investors are not looking for, you know, science projects.
Yep
... anymore. So things seem really good. And hopefully we can provide some, you know, proof points pretty soon.
Excellent. Exciting next couple of months. Well, everyone, round of applause for Dan Galves and Mobileye.
Thanks, everybody. Thanks, man.
All right. Hitting the team, buy side, sell side. We've known each other-
Everywhere.
... for a long time. The Chief Communications Officer at Mobileye, Dan Galves. Dan's responsible for overseeing the company's investor relations, comms, corporate strategy efforts. But given his sort of history in you know, auto, sitting in my seat as well, he's always been very insightful, and I... And he does a great job of putting things into the context that you as investors want to hear. So with that introduction, Dan, thanks so much for being here today.
Thanks, Chris.
So, we always do this sort of overview in your own words. And the way I would talk about it is, you know, I think as we have discussed, and you'd be probably, you know, one of the first to admit, this has been a little bit of a year to forget. Right? For...
If I could only forget.
And, you know, there's been issues on the inventory build first, on SuperVision. There was, you know, some of the production cuts and everything in China-
Yeah
... has come so fast for Tier 1s, and that obviously, that leads into Tier 2s and you sat on a prominent platform in Zeekr. So maybe the first question is, how should we think about, I call it a reset-
Mm-hmm
... over the next couple of years? And then we'll talk about some of the technology that's obviously exciting that we're gonna get updates on-
Yeah
... in the next couple of months.
No, that sounds good. I think if we, you know, if we take a step back and think of kind of what our business is today and kind of where we're trying to get it to, you know, the last few years and kind of through next year are really kind of, you know, phase one of a growth story into advanced products. So I think the growth story of Mobileye going forward is going to be more sophisticated technology, which requires more compute, more software, and generates more revenue per car for Mobileye as the car starts taking over more and more of the driving. And I think phase one, you know, is kind of the strategic objectives are, you know, kind of number one, you know, maintain our kind of core ADAS business, maintain profitability in that business.
Number two is acquire, you know, deepen the relationships with our core customers in terms of getting design wins for these advanced products. And then, you know, another strategic objective is to execute the EyeQ6-based programs that we have, which, you know, the focus right now is on VW Group to launch a SuperVision product on a number of vehicles in 2026, and Chauffeur in, you know, late 2026, early 2027. And kind of these are the objectives of the company. And then phase two, which is really kind of 2027 to 2030, is if we do our job during phase one, then we'll be able to scale these products in that time period, and it creates, you know, a lot of opportunities for investors.
Across the broader portfolio of VW, for example.
Across the broader portfolio of VW, and you know, the likelihood is many more customers as well, so I think what we'd like to see is, you know, our North Star is autonomous vehicle technology, but the way that we build the architecture is such that we can use modules or building blocks of kind of the complete stack to create products along the way. You know, everything from Surround ADAS, which is kind of a multi-sensor, highway, hands-free system that also helps automakers meet, you know, the much more difficult regulatory requirements at the end of the decade. SuperVision is an eyes-on, hands-free, you know, that operates in all environments. Chauffeur is a, you know, an eyes-off that most likely starts mostly on highways.
Mm-hmm
... you know, at the start. Then you have Drive, which is the complete self-driving system that would turn into kind of a robotaxi service. So this is the portfolio. It's all kind of built across the same backbone, and I think the optimal kind of outcome for Mobileye in the 2027 to 2030 time period is that you have multiple automakers that, you know, maybe 10% of their volume is Supervision, you know, 30% is Surround ADAS-
Right
... the rest is, you know, the core ADAS that we have today. And then there's pieces that are Chauffeur that, you know, would tend to scale up towards the end of the decade. But that, that's really kind of what we're working towards. If you look at kind of what's happened... So that, that's kind of the goal. That's really the North Star of the company. I think it's kind of helpful to take a step back also and look at what's happened over the last two years in terms of the positives and negatives since the IPO, since there's been a lot of changes, which you mentioned a lot of. I think, you know, obviously, negative changes tend to get more, you know, attention, but let's start with the positives.
So in terms of, like, the core ADAS business, outside of China, over the last two years, we've won almost 100% of the RFQs that we've participated in, which basically means that across our top 10 customers, which account for around 75% of our revenue, we've locked in that ADAS business through 2032, 2033. And there's been no kind of notable new competitors that have kind of pushed us either on pricing or performance in that area. The regulatory environment, number two, has gotten kind of much more positive for us because kind of the newer, you know, NCAP is kind of the group that creates testing requirements for ADAS systems that results in, depending on how you score, you get a three-star crash rating, four-star, five-star, and this is kind of-
Yeah
... very important for consumer sentiment, you know, to, to different brands, different vehicles. The kind of the new test regime that's gonna be in place for 2028 in Europe, and that tends to kind of, you know, cut across the rest of the developed markets soon after, is such a step change that it's going to require data from more than just a front-facing camera. So that means more sensors, which means more processing power, more sophisticated software, and more kind of revenue per unit for us. So that's been another positive. And then in terms of the Supervision, Chauffeur engagements and Surround ADAS, let's say, we've gone from essentially, you know, one kind of high-probability customer in VW Group-
Mm-hmm
... a second one in Ford that didn't work out, to now we have kind of deep engagements with nine out of the 10 top customers. So I think that that's actually been very successful in the process of acquiring customers. You know, so I think that that's how I would characterize-
Yes
... you know, outside of China. Inside China, there's been a ton of headwinds. I think that you know, on the ADAS side, there really are no clear performance standards in China. There are new local competitors that have come in at lower price and, you know, to kind of create more of a race-to-the-bottom dynamic, we declined to participate, and we've lost some share within the domestic China OEMs. On the kind of advanced product side, there's just a huge kind of push to make this a, you know, China-for-China technology. There's a huge push for OEMs to kind of show their software capability, their technology capability. And despite the fact that the systems that are on the road today that compete with supervision are, you know, twice as expensive and, you know, for the same-
Relatively.
relatively the same performance. There's a lot of kind of traction of this. So, you know, in a lot of ways, we're seeing kind of the auto market, you know, the auto industry really more quickly than we expected, turn into two almost completely separate industries, one in China-
Yeah
... and one outside of China.
A bit of a de-globalization. That theme has sort of come up across the day. To paraphrase, this is mostly like Horizon, for example, on the low end. There are some others that are out there, but Horizon has made some inroads with BYD. On the upper end, these are typically more expensive, NVIDIA compute, multiple LiDAR, but an internal driving policy. Is that fair to say?
Exactly. So, you know, the competition on the ADAS side is really all about price, and it's from third parties, so it's not the OEMs, like, creating their own. So yeah, two or three different competitors that have sort of pushed us on price. It kind of kept on falling. It doesn't seem like these companies really care about making money-
Mm-hmm
... any, anytime soon. So that's the situation there. You know, we still, the business that remains for domestic China OEMs is kind of primarily focused on export, where you have-
Because they have the NCAP. They have to hit it.
Exactly. So you have to hit those standards, so I think that creates kind of some level of leverage or opportunity for us. And then on the upper end, yeah, it's basically a combination of camera, radar, LiDAR data on kind of very heavy compute, you know, from a third party like NVIDIA, and the software is developed internally by the OEMs, you know, thousands of people, lots of cost.
Yeah. So Dan, maybe if we could bridge this near term to that sort of medium term, 2027 to 2030 story. So sort of some quick math that I've put together. So when I think about that short-term path forward, 'cause what I think a lot of people wanna do is use that base of core ADAS, which is maybe, you know, $0.40-$0.45 next year by our numbers, and then it would need something like 200,000+ units in Supervision, 2 million units on ADAS to start to get to the $0.70-$0.75 numbers that are more like 2026.
Mm-hmm.
Can you put some numbers around that? Do we need external wins to hit 2026 numbers, or do you have some visibility, given Audi and FAW, that is sort of in a realm of expectations?
Yeah, good question. Like, I don't wanna talk too specifically about, you know, specific timing.
Sure.
'Cause I think in a lot of ways, you know, kind of the next couple of years, like with SuperVision, like, the intent was that China would create a kind of a growth avenue-
Mm-hmm
... for SuperVision. I think that's been diminished quite a bit. But I can kind of give a framework, and I think I'd start by saying, like number one, like our priority over the next six to 12 months has to be to kind of meet expectations, you know-
Yep
... to kind of consistently meet expectations that the market has and to put up new design wins for the advanced products, like, full stop. Now, it's very encouraging what we're seeing. So, you know, in terms of SuperVision engagements, you know, there are six or seven that are in, you know, the RFQ process. You know, the timing is not certain by any means, but, you know, these are programs where, you know, things have happened that are very kind of strong signs that there will be a decision, and we're not facing much competition. These RFQs, you know, in, you know, when you get to scale with the program, are anywhere from two hundred and fifty to three hundred thousand units per year, per customer-
Per customer
... right? And that's, you know, if you kind of use some, you know, rough assumptions, that's maybe $0.18-$0.19 per program in terms of-
Mm-hmm
... kind of incremental EPS. Again, like, once you get to scale with the program. And on the Surround ADAS side, we have four RFQs that, you know, are, we're responding with, you know, EyeQ6 High at around $175-
Mm-hmm
... to $200 per unit. Now, this is gonna be, you know, replacing a, an ADAS unit.
Yes, yes.
So, you know, if you kind of use incremental-
Net net.
You know, yeah, net net. Well, so each of these are about a million units a year per customer, or averaging out. And that kind of creates maybe $0.08-$0.09 per customer. And I think what's even more encouraging is some of these customers are in both, and so what we're seeing is kind of this segmentation building up, where certain OEMs are looking at, you know, 10% of their business being these kind of aspirational systems, you know, maybe 30%-40% of their business being Surround ADAS, and then the remainder would still be kind of regular-
Yeah
... ADAS or emerging market ADAS.
You, if we were to use that VW analogy, you're gonna put supervision on maybe Audi, but surround would make sense for VW. Again, that sort of segmentation of the market, 'cause you can't price maybe supervision for-
Yeah
... for the lower-end VW.
Exactly. And if you look at, like, a pure premium automaker, which we're also involved in, it's, you know, the numbers are different. You know, maybe it's 50%-
Yeah
... of their volume is Surround ADAS, and 25% is SuperVision.
Yeah.
You know, it's encouraging, but we need to go out and kind of, you know, complete these processes.
Dan, I always hate being, like, the Wall Street guy on pushing on timeline, but there was always... You know, Amnon used to make the point of, we are, what, 99%-
Yeah
... along, and RFQs always slip. Literally, one of my first slides was, you know, as an auto analyst, you're just used to one- to two-year push-outs. Has there been, over the last 12 to 18 months, things like DXP that have extended timelines, where maybe some of these customers are looking at using supervision across more platforms?
Mm-hmm.
But anything you could say about the timeline specifically on those Supervision ones which you've been talking about for years?
Yeah. So it's, I think that there have been, not recently, right, so no change-
Yeah
... to kind of what we've recently said, but, you know, kind of the EV/ICE replanning process was a slowdown in the ability to make decisions on new technologies, right? So from maybe October of last year until, like-
Yeah
... February, March, not the same kind of level of progress with these programs. I think at the same time, you had China, like, you know, what are we gonna do about China?
Yeah.
Also taking up a lot of kind of management airspace, and that did lead to some kind of a you know lack of progress during that period, but recently there's been, you know, significant progress, and this is not like we're having, like, a catch-up meeting-
Mm-hmm
... every two weeks. These are, you know, specific kind of like prototype vehicle expeditions and, you know, demos for this group of ex-executives in this region, demo for this group of executives in that region. You know, start of commercial discussions. So, like, a lot of work goes into these programs, and, you know, I don't wanna kind of commit to timing.
Mm-hmm.
I think that there's enough of them that the opportunity to have some, you know, positive news by the end of the year is good. But I think the most important is, like, we've seen these programs reach periods where it becomes embarrassing to-
Yeah
... to not move forward.
And even slow-
Like, when you-
Even slow legacy OEMs also have their timeline that they want to sort of complete something by a rough date, yeah.
Yeah. I think, you know, 2027 is important. Like, you know, automakers believe that, you know, Tesla will continue to double down on their technology and improve it. And, you know, you have Volkswagen Group that is gonna be launching these systems in 2026, which creates kind of pressure on-
Yep
... the other OEMs. So there is some level of timing pressure, and, you know, and the stuff that you talked about has had an impact, too. Like, the Volkswagen deal, like, actually took a lot longer than we thought, but it resulted in much higher volumes than they were originally talking about.
And then, Dan, a realistic reset on China, right? Amnon talked about, I think it was late 2025 into 2026, so the chip will be ready in 2026, a lower-end, more simplified chip to compete. But like you said, it's a race to the bottom.
Mm-hmm.
So how do you think about what is a realistic Mobileye business in China over the next, you know, two to five, two to five years?
Yeah. So I think the exposure is already, like, a lot smaller. Like, you know, if... From our filings in the first half, you'll see that China was about 30% of revenue.
Mm-hmm.
What was that, right? So I think if you go back to 2023, you know, you had... Well, let me just look ahead. So, like, 30% of revenue in the first half, you know, rough numbers within our guidance for the second half, it becomes 15%-
Yeah
... of our revenue.
With Zeekr being a material step down, right?
Right.
Before. Yep.
And if you think about domestic China OEMs specifically, it's about 4% of our revenue, so the exposure is much lower. Now, you know, it's not just this, so in some ways, the 4% is probably more solid than the-
Yeah
... the balance of the fifteen, because that
It's close
... the balance of the 15 is foreign automakers producing cars in China, using our ADAS, and those companies-
Sales are going lower.
... are likely to see-
Yeah
... volume declines. You know, the 4%, you know, includes some level of Zeekr volumes, and then about a million units per year of ADAS chips to domestic China OEMs, which half of that is for export. So I think if you... You know, our strategy in China is not to try to claw back market share by lowering pricing. Our strategy in China is basically to kind of present to the OEMs, which we have been for a while now anyway, that if you have global ambitions, if you have ambitions to export into developed markets, we can be a really good partner for you, 'cause we can help you get to five-star-
Yeah
... safety ratings, like, in a lot of different markets.
Which is a BYD next three to five years, maybe not the next one to two years, once, like, Hungary and other export facilities are actually in place in Europe.
Exactly. I think that the kind of export into developed markets from Chinese automakers has gone a lot slower than we expected. And the kind of market share gains within China has gone a lot faster than we expected. So it's, it hasn't been a great environment for us, but we do see opportunities, and kind of the remaining business in China for ADAS is a lot of export and, you know, with a couple of companies that, you know, have kind of indicated that they want a long-term relationship. So we'll see. I think we're, you know, we're trying to kind of... You know, investors should be not expecting, you know, any type of kind of recovery in China.
Yep.
That's kind of the part of the process that, you know, we feel like we're pretty far through now.
Well, well, well, Dan, one of the things I, that I've said throughout the day is I play referee, and I always do the good and the bad. I feel like I already hit you with the tough questions. One of the things I wanted to discuss is, you know, the real excitement that will come, whether it be, I don't know, Q7, whatever, I can guess on what the Audi vehicle is in 2026, 2027, where Chauffeur is sort of Mobileye's way of, you know, saying Level 3, Level 4, essentially hands off and eyes off.
But what I wanted to dive in, because I don't think everyone here sort of knows the Mobileye framework, I put it up in one of the slides, is I've asked this question to many people over the course of the day: What is the bar-
Mm-hmm
... that an autonomous system needs to be gauged against? And I've gotten answers as, "We don't know. It's vague." Mobileye seems to sort of plant their flag that you want to be five, 10X better than a human driver over, I think, four hundred million miles. Can you go into a little bit that framework? Because you are sort of, you know, someone that is... and Amnon has always been very specific that you're not going to go before the technology is ready-
Yeah
... because regulators and society aren't gonna stand for that, when, you know, when lives are at risk.
Yeah, it's a difficult question. I think that we're listening to our customers, which are kind of very clearly saying that for kind of an unsupervised system where the driver is out of the loop, they'll be taking on the liability-
Mm-hmm
... when that is there, and that they don't believe there will be much tolerance for machines causing accidents, and for that reason, that they're setting a very kind of high-performance requirement. There is a European regulation that has kind of the words, like, "better than human performance"-
Yep
... in it. Nobody's defined that yet, so it's kind of a very difficult thing to get your hands around because, like, you know, what type of a failure are you talking about? A failure that would cause a fender bender, a failure that would cause, like, an injury, a death? You know, the statistics vary based on kind of those metrics. What we've always kind of targeted is a million hours between what I would call, like, a significant failure-
Yep
... which is, you know, somewhere between 10 and 50, 50 times kind of human performance, the way that our customers are looking at it. And that's really high, right? And so the plan basically is to have two separate perception systems, each of which is capable of driving the car on its own, that has different failure modes that, you know, like a corner case for one system wouldn't be a corner case for the other.
Yeah.
If you had sun glare, it would affect the one system, it wouldn't affect the other. So, fail in different ways is, like, a big philosophy at Mobileye. And so the plan is to get to 1,000 hours between failure for each of those systems, and kind of theoretically, if you have both of them and they're independent, then you could multiply those failure rates together, and you get to about a million hours, which is, you know, maybe 30 million miles, something like that. Now, if, you know, regulators decide that the bar is lower, or you have kind of competitors that, you know, set the bar at a lower rate-
Because they're on the road. Yeah.
... that wouldn't, because they're on the road, that wouldn't be the worst thing for us. But I think ultimately it's gonna be up to our customers, and I think regulators will have a say as well. The other thing to kind of keep in mind, because it's like... A lot is gonna depend on how many cars are on the road as well. So it's like if you have a system that's, you know, one failure over, you know, every 80,000 miles or something like that-
You'll never see it.
... then, you know, but you only have one car-
Yeah
... you're probably okay. But if you have 100,000 cars, you're gonna have-
Yeah
... you know, three accidents a day.
Yeah, that's one of my slides. Yeah.
Oh, really? Okay, yeah. That's good. So it's, yeah, it really depends on a lot of factors. So I think for Waymo, you know, the opportunity to have, like, a small fleet, and we're super kind of impressed with what Waymo's done. 'Cause people like, you know, ask me, like, "Well, Waymo's saying this, and, you know, they're saying that." I think for Waymo, it's like, you know, when you have 700 cars on the road-
Exactly
... it's not like you're gonna end up with, you know, 100 different court cases, you know? But, you know, with a big automaker that's looking to put these systems on the road and tens of thousands, hundreds of thousands-
Yep
... of units, they have to think a little bit differently.
Can you talk? You know, Tesla talks about shadow mode to the idea of this validation, particularly of supervision into a chauffeur. Again, not knowing what the exact vehicle, but if we use Audi as an example, do I have the right understanding that a vehicle will be launched with all of the hardware?
Mm-hmm.
Capable of Level 3, Level 4 Chauffeur, but maybe for the first X number of months, the Chauffeur would be running in shadow mode?
Yeah.
So that you would only be able to have a supervised system, and then at some point, Audi would say, "For the price of X, you can upgrade, and now you can, you know, theoretically fall asleep in the car.
That's the plan. Like, I've seen kind of the timelines for the Audi Chauffeur program, and yes, like, the launch date is X, launching with Supervision capability, eyes on.
Mm-hmm.
So at launch, you'll be able to drive hands-free, but you'll have to keep your eyes on the road. The system is gonna monitor the performance of the car, you know, in the background, you know, monitoring interventions of the driver, but also kind of like, you can understand failures as they happen, you know, in Shadow Mode. And basically, you wanna build up enough miles. If you have, you know, 30, 40, 50 thousand cars on the road, like, those miles are gonna pile up really fast. So you can kind of get to this, you know, proof point of, okay, like, these systems drove, you know, 8 million hours and had six failures, or something like that. Because, you know, I think that they're, especially in Europe, I think that the regulators are gonna want proof.
Yeah.
Mm-hmm.
I wanna give ample time for Q&A before I keep on rattling off. Anything from the audience? Don't be shy, don't be shy. It always takes a... Anyone?
Oh.
Neil.
Hi.
How's it going?
Hey, good, good. How are you?
Wow, good, good. I wanted to ask about just the LiDAR news, and I think there were several, you know, reasons cited in the press release around why Mobileye's shutting that down, you know, doubling down on imaging radar, advancements in computer vision.
Mm-hmm.
Can you give a little bit more color on that? Is the view that because of these advancements in other areas, the relative, you know, importance versus a year ago or two years ago?
Yeah
... is LiDAR losing that positioning? Or is it really about, you know, the rest of the ecosystem, other players in the market being at a sufficient level, sufficient cost, where Mobileye doesn't need to, you know, invest in that area anymore? Just any color on that would be helpful.
Yeah. No, it's a good question. I think we're. Yeah, we're. It doesn't mean at all that we're not planning to use LiDAR in these advanced systems like Chauffeur and Drive. I think in a lot of ways, like, our view of LiDAR is kind of always, like, what do we need from it, given what we're gonna get from the computer vision and the imaging radar system? And so over the last couple of years, we feel like, you know, transformer-based architectures have kind of helped robustify our computer vision quite a bit. So kinda what we're seeing in kinda the EyeQ6-based testing for computer vision, with all of the new kind of AI technology within it, is promising, is better than we thought, you know, a couple of years ago in terms of robustness.
The imaging radar is meeting the performance specs that we were targeting, but now it's meeting them with, like, B samples, so this is much more kind of almost production ready for us. So, you know, that kind of confidence level, that it would hit the specs, like, has gone up a lot. So that sort of led us to believe, like, okay, we probably don't need, like, an advanced LiDAR, even in the second generation. It's, you know, what time of flight gives us is probably gonna be plenty. And then I think the price gap between, you know, what we had line of sight for our FMCW versus kind of where time of flight is now or looking like it's gonna be in a couple of years, you know, there's always a price-performance trade-off, right?
The extra performance you get from FMCW, in our view, might not be worth kind of the price gap, like, for, you know, with what we're doing. We were about to need to really kind of make a decision to kind of move forward on production, which means you've gotta get contract manufacturers and start, you know, putting, like, you know, significantly more dollars behind it. It turned into an easy decision to make, but I don't think it really says anything about LiDAR specifically.
Dan, is the right context there that Mobileye has two different OS perception stacks, so camera and radar, LiDAR. So on the radar LiDAR perception stack, you are getting the velocity vector from the imaging radar, so you don't need it from FMCW. So time of flight, you know, whether it be Luminar or Innoviz, whoever it is-
Yeah
... gives you that information.
Yeah.
Yeah.
Yeah, exactly. And I think, like, the you know, the good LiDAR suppliers have really been performing, too, in terms of kind of performance and, you know, production capability. So, you know, we feel good about our access to the technology from third parties.
Any other questions? One from me. Yeah. Dan, you mentioned... Look, and I'm guilty of this over, you know. I always say I'm not a computer science, not a, you know, not an engineer. Years back, I used to put your approach in sort of this bucket of rules-based.
Mm-hmm.
Right? RSS is sort of a rules-based overlay, a do-not-violate sort of safe space. But I think we've underestimated the amount that you are using, you know, transformer models, gen AI, but in a hybrid, I think Amnon's called it a compound AI approach.
Yeah.
Do you wanna talk a little bit about how you use gen AI in driving policy?
Yeah.
Then maybe how it's different than the likes of Wayve or Tesla or Waabi.
I think that we've always, you know, our CTO is, you know, one of the premier, you know, AI machine learning academics in the world, and Amnon is no slouch either. You know, they've published all kinds of white papers and, you know, patents and so, you know, and they both maintain kind of an academic position as well, which I think helps keep them up to date with, like, you know, everything that's happening on the cutting edge. And so we've always integrated, like, deep neural network, deep learning technology, you know, that's been in the system. I think the first deep neural network technology in a car was in the first Tesla Autopilot-
Mm-hmm.
to do a specific task. But our system has always been mixed in with other techniques and also some rules-based systems. So, you know, for example... So I think that this concept of fail in different ways, you know, we've been, you know, detecting vehicles with a deep neural network, you know, machine learning technology for, you know, six, seven years. But we've also, you know, maintained a legacy kind of appearance-based system where all you... You don't really care if it's a vehicle, a pedestrian, a cyclist. What you care about is it has mass, and it's above the road, and you don't wanna hit it.
You can stop it.
Right? So I think that this kind of fail in different ways because, like, you know, the AI more of a, you know, more learning-based technology is, you know, amazing, but if your data set doesn't include what you're seeing, there's a chance that it would be ignored. Now, we used kinda convolutional neural networks to do kind of the, a lot of the perception stack, you know, from 2016 and through EyeQ5. Now, CNN is great, but you can tell it's taking, like, a lot of brain power here for me to describe this. But, you know, if you get an image from the camera, a frame from the camera, it's made up of lots of different images, and each one of those images has to be analyzed separately. So, you know, imagine, you know, a view of, like, 50 different vehicles-
Yep
... on the highway.
Pixel by pixel, yeah.
You have to. It's kind of pixel by pixel. You have to kinda analyze and detect each vehicle separately. You have to stitch the frame together to, like, an environmental model. You have to kind of bring it up from 2D to 3D. There's a lot of steps in the process. Transformers allow you to analyze the entire frame as one, which is a kind of huge positive in terms of efficiency and, you know, reduced complexity. But you still have the same issue of, you know, a data set that needs to be, you know, complete enough to include, you know, every possible scenario you can see out there.
So it still, in our view, requires internal redundancies to have different kind of approaches and different modules that would detect things in different ways, so that you can, you know, boost the performance to super high levels. What we don't wanna be really in the business of is, like, going out and trying to find failures, and then developing more and more data-
More training, yeah.
... to kind of fix those failures, retrain the network, hope that it doesn't, you know, create additional kind of failure modes that you didn't have before. So I think that that's what, you know, our... We wanna kind of set ourselves up so we have kind of a clear path to the performance requirements that we were going over before, without, like, a lot of uncertainty about how long is it gonna take us to find all these corner cases, retrain the networks? How much data are we gonna need? How much compute are we gonna need? It's... In our view, it's not the right way to do it.
Yeah. Last one from me, just the opportunity for parting thoughts. I always, you know, like to push the envelope. Anything that we may should start to think about, topics for maybe the investor day in December? Anything to end on.
Yeah, no, I think that. Yeah, it's been a tough year with a lot of different headwinds. I think a lot of it's been focused on China, you know, maybe some of it on things that we're, you know, not really in control of, like our, you know, our parent company. I think what, you know, we as a management team wanna do is kind of get back to basics and sort of, you know, reset, you know, the way we talk about Mobileye in terms of kind of what are our competitive advantages, what are our moats? Kind of what does our business look like? How do you value these deals if they come, right?
To give us a frame-
The framework.
Give us a framework for how to look at an announcement and put it into-
Exactly
... a generic bucket.
I think that the-
Yeah.
I think the analyst day in December is gonna be really, really good because you'll get, like, a kind of a full, like, kind of exposure of our technology stack and the way we're approaching AI, the way we're approaching, you know, mean time between failure. You'll get a lot of information on kind of the progress that we're making with Porsche and Audi. I mean, this is like, it's incredibly... It's the most intense kind of execution phase that Mobileye's ever gone through, you know, us as a Tier 1.
And you'll also be the... I think it's public that you'll be the Tier 1, right? I mean-
Yeah, we're gonna be the Tier 1, which creates, you know... I mean, there's 40,000 requirements that we have to meet, you know, with this program. But I think on the other side, like, the Porsche and Audi engineers that have kind of launched complex programs like this many times in the past, like, we're working together to create kind of a platform that can then be scaled to other automakers as well.
Okay.
So I think it's, you know, we're kind of building these muscles, executing, things are really on track, and I think we'll be able to kind of demonstrate that at the day. And some of these frameworks for, you know, how to think about, you know, the business-
Yeah
... looking ahead.
You're always in a rock and a hard place, so you can't really say too much. So it's more like a general announcement. Could be, you know, X, X number of units-
Yeah
... and then come back.
I think automakers are more kind of open to announcements with us now because, you know, we're a tech provider-
Mm-hmm.
you know, and so I think that they kind of understand that investors are not looking for, you know, science projects-
Yep
... anymore. So things seem really good. And hopefully we can provide some, you know, proof points pretty soon.
Excellent. Exciting next couple of months. Everyone, round of applause for Dan Galves and Mobileye.
Thanks, everybody. Thanks, man.