Mobileye Aktie

Greetings, and welcome to Mobileye's First Quarter 2026 Earnings Conference Call. [Operator Instructions] As a reminder, this conference is being recorded.

It is now my pleasure to introduce your host, Dan Galves. Mr. Galves, you may begin.

Thank you, Maria. Hello, everyone, and welcome to Mobileye's First Quarter 2026 Earnings Conference Call for the period ending March 28, 2026.

Please note that today's discussion contains forward-looking statements based on the business environment as we currently see it, including regarding our future financial outlook. Such statements involve risks and uncertainties. Please refer to the accompanying press release, which includes additional information on the specific factors that could cause actual results to differ materially.

Additionally, on this call, we will refer to both GAAP and non-GAAP figures. A reconciliation of GAAP to non-GAAP financial measures is provided in our posted earnings release.

Joining us on the call today are Professor Amnon Shashua, Mobileye's CEO and President; Moran Shemesh, Mobileye's CFO; and Nimrod Nehushtan, Mobileye's Executive Vice President of Business Development and Strategy.

Thanks. And now I'll turn the call over to Amnon.

Thank you, Dan. Hello, everyone, and thanks for joining our earnings call. We delivered very good results in the first quarter. Revenue was up 27% year-over-year. Adjusted operating income was up 61% and our operating cash flow was again strong at $75 million despite working capital timing that was a modest drag. We have seen upward pressure on demand for our EyeQ product for the last several quarters. That continued in Q1 and is what we expect for Q2 as well.

As a result of higher volume and revenue in Q1, we have raised our 2026 outlook towards the high end of our original guidance, leaving the outlook for the remaining 3 quarters essentially unchanged. The geopolitical and economic environment remains volatile, but based on our visibility for Q2, we believe there is sufficient conservatism baked into the second half.

Diving deeper into the drivers of our business, our ADAS business is very strong with very high margins and cash generation. Design wins over the last several years have secured our position with our main customers over the long term. India looks like a meaningful growth opportunity and our focus over the last couple of years on supporting Chinese OEMs on their export ambitions is paying dividends. Finally, the Surround ADAS segment gives us the opportunity to replace many of these base ADAS programs with much higher average selling prices over time.

On our advanced product portfolio, the current priority remains execution, and that is going very well. We have a number of production programs running in parallel, 2 of which start production in the relative near term. These are SuperVision with Porsche and the Drive robotaxi with MOIA, the Volkswagen Group's autonomy division. For both programs, Mobileye is responsible for the development of comprehensive advanced ADAS and autonomy platforms, integrating hardware, software, data and maps into a complete system that must be probably safe, predictable and verifiable.

These solutions need to meet tens of thousands of requirements set by the automaker and need to be homologated to automotive grade standards. Each program gives us the ability to prove that Mobileye is the leader in developing and executing complex AI-based systems in the physical world at global scale, systems that can be validated under strict standards, something that many companies talk about, but few besides us are actually executing on this vision.

Specific updates as it relates to SuperVision are as follows: progress is strong with performance tracking well to our objectives. As a concrete example, 6 weeks ago, we had the first OEM directed drives in the U.S. for this system, having only tested in Germany and Israel previously. Our first task was a 2,000-plus kilometer drive in a vehicle equipped with production EyeQ6 High SoC and ECU hardware with the latest software engines integrated into the production architecture. We had no prior knowledge of the route, which was across a diverse set of urban, suburban and highway road types and severe weather, including heavy snow.

The SuperVision system performance was outstanding with very few interventions encountered. This was an important proof point for our out-of-the-box performance and ability to generalize to a brand-new geography. We have a couple of more software releases to make and then expect to have the capability to demonstrate to other potential customers in the various key geographies.

On robotaxi, we continue to make rapid progress. In Q1, Volkswagen announced the start of pre-series production of the ID.Buzz autonomous vehicle in its Hanover facility with vehicles coming off the regular assembly line with Mobileye's fully integrated self-driving system. Volkswagen ability to produce fully integrated robotaxis at scale from an active automotive production line is very unique.

MOIA, the Volkswagen division that will deploy these vehicles announced that testing has begun in L.A. for the Uber collaboration. They also announced today that Orlando is the first launch city in collaboration with BEA. For both of these efforts, the path to commercialization is as follows: we continue the current process of testing, data collection and validation. Once we achieve sufficient proof points, we'll begin accepting commercial riders with a safety driver until the required level of performance has been proven that allows us to remove the safety driver.

That is the point where the scaling advantages of our approach, including crowdsourced mapping, our deep and diverse global data set and Volkswagen ability to ramp up production rapidly will be self-evident in terms of ability to expand geographic areas of operation more rapidly than competitors. And it's another opportunity for Mobileye to prove its end-to-end capability in terms of executing complex physical AI systems at scale. All of this experience over the next 2, 3 quarters would feed back to further improvements and fine-tuning to be ready for scaling in Europe once the ID.Buzz is fully homologated and certified, which is targeted for the first half of 2027.

Turning to the Mentee side, components of the version 3.2 of the robot have arrived and will demonstrate incremental capability soon. The hardware road map for version 4 is nearly complete and is expected to be ready for demonstration by early 2027. This will be the version that we expect to commercialize for initial use cases and market entry and will be cost and weight optimized and offer enhanced dexterity and manipulation capabilities.

Finally, on the buyback we announced this morning, we are a cash-generative company, which is unique in this space. That gives us the ability to pursue growth opportunities like we did with Mentee, but also be opportunistic with our equity. While we are making strong progress on our advanced products and conversion of our large future revenue pipeline, the reality of automotive development time lines and OEM confidentiality agreements limit what we can disclose publicly.

In an environment where technology competitors are generating significant news flow, we believe that this lack of visibility has weighed on our stock price. While we continue to execute, we see an opportunity to deploy cash towards share repurchase, which will benefit all shareholders by partially offsetting dilution from stock-based compensation and addressing dilution from the Mentee transaction at significantly more attractive prices than those embedded at closing.

I'll now turn the call over to Moran.

Thank you, Amnon, and thanks for joining the call, everyone. Before I begin, please be aware that all my comments on profitability will refer to non-GAAP measurements. The exclusion in Mobileye non-GAAP numbers are typically amortization of intangible assets, which is mainly related to Intel's acquisition of Mobileye in 2017 and stock-based compensation. This quarter, we also excluded the goodwill impairment loss referenced in the press release and transaction costs associated with the Mentee acquisition, which closed in early February.

First quarter revenue of $558 million was up 27% year-over-year. This compared to the indication we gave on the January call of about 19% growth. We had assumed shipments of approximately 10 million EyeQ units in the quarter, including some recovery of safety stocks at customers, which has ended 2025 at a very low level. The uptake in the quarter was a combination of higher share and higher ADAS shipment rates at core Western customers and more meaningfully from robust Chinese OEMs volume from the export market, a segment where we have higher share than we do on Chinese OEMs vehicles sold domestically.

Adjusted operating income was $95 million, up 61% year-over-year. Adjusted operating margin was 17% up about 4 percentage points versus Q1 2025. Profitability was largely as expected. Strong mix to our top 10 customers was a bit of a tailwind, offsetting the higher China OEM volumes, which typically carry lower pricing and profitability. Operating expenses were as expected, representing about 25% of our full year expectation of around $1.1 billion and were up versus Q4, mainly due to engineering reimbursement timing that relate to production program milestones and also the consolidation of Mentee expenses as of early February.

As I noted on the January call, we've been seeing consistent positive revisions from our customers throughout 2025, and that continued in the first quarter of the year. we continue to expect that underlying demand trend is in the low 9 million units per quarter. Q1 was a bit above that, even excluding the adjustment of safety stock at our customers, but we prefer to continue to forecast a reversion to that low 9 million unit trend over the balance of 2026, particularly given the geopolitical and economic volatility that Amnon mentioned earlier.

Turning to full year guidance. We are increasing the revenue outlook to $1.975 billion at the midpoint, which implies 4% year-over-year growth. This is underpinned by about 38 million EyeQ units, which is up a little less than $1 million from the prior outlook, accounting for the upside in Q1. A bit more granularity on the volume is that the forecast assumes the current SMP production forecast of our top 10 customers, which is currently minus 3.5% year-on-year.

It also assumes that the run rate of China OEM volume in the second half of 2026 comes down meaningfully from the first half levels. We aren't sure what will happen, but given low visibility on that part of the business, we prefer to stay conservative. We are increasing our outlook for adjusted operating income to $210 million at the midpoint, up from $195 million in the prior outlook.

The 2 items impacting revenue to income conversion are: number one, a good portion of the incremental revenue is related to China OEM volume, which converts at lower revenue per unit and profitability than the rest of our volume. Number two, on the SuperVision side, volume is consistent with our prior outlook, but we do have some incremental costs for the ECU, particularly related to memory. Our assumption of operating expenses are unchanged at approximately 10% year-over-year growth to around $1.1 billion.

Finally, on the full year, we have now provided an outlook for GAAP operating income. At the time of the January call, the impact of the amortization and stock-based comp from the Mentee acquisition was not able to be estimated precisely. Now it is. The only thing to note is a reminder that only a portion of the shares issued as part of the acquisition will show up in the share count this year. That is because the majority are tied to vesting requirements for the Mentee family. Therefore, the relevant accruals are included in the projected share-based compensation expenses referred to in the guidance.

Another important point to note is that while there will be share-based compensation expense and some impact to the share count associated with these shares in 2026, it will be gradual as full vesting only occurs for 50% of the shares in 2028 and the remainder in 2030.

Turning up to second quarter. We are assuming about 9.3 million EyeQ units and for revenue to decrease approximately 6% on a year-over-year basis. We would expect gross margin to be slightly below Q1 levels based on the mix of orders we are seeing currently and for operating expenses to be consistent with Q1 levels, maybe slightly down. To conclude, we are almost 4 months into 2026 and continue to see positive demand signals from our customers on the core business.

As Amnon discussed, we are also seeing very good execution progress ahead of a large number of advanced product launches over the coming 1 to 2 years, which we expect to create significant growth for the company. Finally, I am pleased that we are able to begin a share buyback program as we believe it takes advantage of the strong cash flow of our business and benefits our shareholders by offsetting a portion of RSU issuance, which is a critical part of Mobileye's employee compensation structure.

Thank you, and we will now take your questions.

[Operator Instructions] Our first question comes from Edison Yu with Deutsche Bank.

This is Winnie on for Edison. First question is on the ADAS side. It seems like the year's guide is raised reflective of the 1Q beat. So just curious what conditions you're seeing now on the channel and customer, you've given us some guidance for the rest of the quarter. So can you just refresh some of that to reflect what you're seeing in the first half of the second quarter?

I think that basically, in terms of the guidance, we are reaffirming our guidance from January call and adjusting it for the upside that we're seeing in Q1 as we don't anticipate this upside to impact the rest of the year. So that's the main reason.

And as for the upside, I can now briefly mention what we're seeing. So first on the China OEM export volumes, we're seeing -- I think that half -- probably half of the upside is coming from there. We're seeing very strong demand for both Q1 and also incorporated into Q2. For the second half, we're still conservative and this market is volatile, but we're seeing very good demand on that and some of our customers have very significant year-on-year growth on exports.

Secondly, ADAS fitment rates is increasing in 2026 for our top OEM customers. We're seeing constant demand here throughout the year. So this is, I think, unchanged from our January call. And third thing is the safety stock inventory adjustment for our customers. We talked about it that 2025 ended at very low level, even below 3 weeks or so. And they have now increased their safety stock to approximately 4, 5 weeks, which is kind of normal. And we don't anticipate this volume to reverse this year as it's kind of normal stock they need for their ongoing shipments.

Just to add on the macro side that there are a few tailwind effects that we are benefiting from. The first one is the increase in export volumes by the Chinese OEMs that our strong customers. Interestingly, these volumes are in emerging markets like Asia and South America and are not necessarily competing with the European volumes that we have. So we can benefit from an overall increase in volumes.

The second is that we have increased our market share in our key customers. Although we have been in the majority of the volumes of our top 10 customers, it wasn't necessarily 90-plus percent in all of them and gradually over the past 2 years, we have been increasing our market share, replacing older solutions by competitors. So that is also a tailwind effect. And overall, we don't anticipate these 2 trends to weaken. We expect them to continue as they were, and it's Kind of what stands behind the revision to the guidance.

That's very helpful. A follow-up on Mentee. I was wondering if you can give us an update on the progress made thus far? And would it be reasonable to assume some kind of proof of concept later in the year with external customers?

We we're making progress in 2 fronts. One is the hardware. What we have shown a month or 2 ago was version 3.1 -- version 3.2 is being assembled now, better dexterity, improved hands as well. Software-wise, we are integrating VLMs into the system, designing tasks that are more targeted to home use tasks or to B2C domains. Version, we have another version 3.5 of the hardware in 2 months from now and version 4, which is the hardware to go into mass production should be ready by end of this year, early next year.

Regarding the proof of concept, we are still analyzing the domains. Part of our analysis is the viability of the B2C model rather than B2B or starting B2C and then B2B. So we're still analyzing the opportunities of the use cases that we are building for the robots.

Our next question comes from Chris McNally with Evercore ISI.

Amnon, just wanted to focus on the upcoming KPIs for the Driver Out, as mentioned in Los Angeles and Florida and maybe what's to come after Driver Out with respect to commercial scale. So if I divide it into 2 parts. On the Driver Out, what's left in your time line to validate the service for that fall in Q4 launch, as you mentioned?

So our milestones for the Driver Out is first to start validation on the final Level 4 vehicle. So there is still a few more months until we get the final vehicle ready for series production, then we'll start the validation. Also, there are some things that we need to close with the remote operators to make sure that everything there is running as we plan. And then we start with commercial drives with a safety driver and towards the end of the year to remove the driver. And we are on track with all our plans in that area. What comes after that? Well, we will see.

Our first priority is driver out on an SDS system that is fully homologated, both software and hardware, automotive grade. This is a huge moat. This is very, very important. Once we get that, the second is scale. We want to see 2027, at least 6 cities, hundreds of vehicles at minimum. That's the next real big milestone. And then we'll look at the market and see whether we need to simply remain an SDS provider, which at the moment is our plan A or to extend our vertical integration. We'll see what happens by end of 2027.

Perfectly clear. And I think you basically hit on the first part of my follow-up. But if we take the second, I think we all understand Driver Out is not really the end of AV development or service. So could we talk about the original -- the ODD expansion, will the first commercial service go on the highways? And how do we think about those AV improvements, which are non-safety critical, smoothness of the ride and essentially your service getting better as it ramps in 2027 in the U.S.

No, we're talking about robotaxi. So robotaxi is full deep urban point-to-point in cities. We have the capability also to support highways, but we will start in deep urban inside cities and then gradually expand into highways as well. In terms of comfort, this is part of our KPIs today. I don't see us coming out with a commercial service that doesn't have the necessary comfort level of driving.

But of course, the most important is the safety level. But we are in our KPIs, measuring also what would be called roadmanship, making sure that at the comfort level, we are also meeting our KPIs. So all of that should be in 2026. 2027 is more focused on scaling, both scaling number of vehicles, also reducing the ratio between teleoperators to vehicles. That will be the goal for 2027.

Our next question comes from Joe Spak with UBS.

One quick follow-up on the guidance and then a bigger question, Amnon. On the guidance, I know you mentioned that one of the reasons for the EBIT flow-through versus the revenue flow-through was some of the China mix. But if I understood correctly, I thought the better China volume was in the first quarter and that you're still assuming sort of that low 9 sort of pace globally for the rest of the year. So maybe just could help me understand some of that conversion. And if you could, I think you mentioned 2Q is still trending pretty well there. So maybe some even more near-term expectations on the quarter.

Okay. I'll give it to Moran.

Yes. So I think for the China OEMs that you mentioned, it's not just in the first quarter, it's also in the second quarter. So in terms of the years -- in the year, the portion of China OEM has increased in a few hundreds of thousands of chips, which impacts, of course, the conversion of revenue to profitability and also the export volume in China is for a lower ASP than what we sell in the West. So that's for the guidance clarification.

Just to add to this to clarify, these China export volumes do have lower ASP. However, they are for new markets that today -- until today, we did not have any sales in. So it's not that there is a competition between higher ASP, higher-margin European business, for example, or American business for us that now comes from a lower ASP from China. These China volumes go to, let's say, blue oceans when it comes to ADAS penetration. So it's a net gain for us.

Okay. And then just, I guess, to follow up on Chris' prior question with the Drive product. Like I appreciate the commentary on the KPIs, but like what's really like the process like here between the different parties? Like where does sign off on some -- like moving to the next phase lie? Is it with you? Is it with MOIA? Is it with the TNC?

And then you did briefly sneak in there at the end that you'd look after these launches, whether it makes sense to remain an SDS player or extend that vertical integration. I mean the latter would clearly give you more freedom. Is there anything that prevents you from doing that from a partnership or exclusivity perspective?

No, nothing prevents us to pursue the right business direction. It also depends on how the future plays out? Are there going to be 1 or 2 SDS suppliers out there, which is our current assumption? Or there are going to be multiple. If they are going to be multiple, maybe the right business decision is to go more vertically integrated. But it's too early to tell.

Right now, our focus is on the SDS, on the driver route, the SDS, the hardware, the software, the driver route, the roadmanship, the teleoperation, there's lots going on there, the maps, making sure that the maps scale so that we can scale quickly from city to city during 2027. That is the focus of the company. We have no limitations on how to pursue our business model.

And as for the first part of your question, the driver out eventually depends on the customer, which is MOIA and Volkswagen. We are supplying the technology. We do not determine when the driver would be out, but our KPIs and milestones of both parties are targeting end of 2026.

Our next question comes from Josh Buchalter with TD Cowen.

I'll start with one on the model. I'm a little confused on the ASP trends implied in the guidance. So you mentioned China tends to be lower ASP. But if I sort of run this low 9 million EyeQ shipments per quarter through the rest of the year, it implies ASPs continuing to trend down through the rest of the year despite China becoming a lower part of the mix and potentially some advanced ADAS solutions later in the year. Can you help walk me through the ASP trends through the year? And if we should indeed be modeling low 9 million EyeQ shipments per quarter through 2026?

Yes. So I believe in January earnings call, we discussed ASP with regards to the second chip that we have this year. We have approximately -- we have a program and one specific program with a dual chip when the second chip is discounted. We have approximately 800,000 units this year. So this is an ASP headwind of like $0.80. And with the China OEMs, we did increase, as I mentioned before, the China portion in terms of volume for 2026. So this is an additional maybe $0.30 or $0.40 decrease in ASP since our last estimation, although volume has increased significantly. That's the explanation.

Yes. I think it's pretty -- it's difficult to be precise about it because there's other parts of the business as well. And just to be clear, we're not assuming additional advanced product launches for this year.

Okay. And then maybe a bigger picture one. Amnon, given your position in the industry, I was hoping you could maybe reflect on how the regulatory environment for autonomous mobility broadly and robotaxis has changed over the last year. And when we should expect that to be a more meaningful part of Mobileye's model?

In the U.S., it's a self certification, which is very convenient to start ramping up. In Europe, the bar is much higher in terms of homologation, and this is the advantage of our partnership with MOIA and Volkswagen that they take the homologation part to homologate the vehicles in Europe. And I believe that as robotaxis will start proliferating from the thousands of units to tens of thousands to hundreds of thousands, we would see more regulation coming in everywhere, not only in Europe, but also in the U.S.

So having a very clear and precise and crisp definition of safety. In our case, it's RSS and PGF stuff that we talked about back in the past is very important to prepare the company towards an environment in which the regulatory profile is going to be much more stringent.

If I may add to this, I think that if you see the communications from other companies on robotaxi launches, it's primarily either in China or in the U.S. You see much less noise or news, sorry, coming for European market. We think that some of the reasons for that is because of the regulatory requirements in Europe that we are -- that we've been actively working on with VW over the past 1.5 years almost.

Through this engagement, we have exposure to what -- how regulators view this business. And they do require specific APIs and very detailed explanations on validation concepts and testing methodologies and how can you overcome different unexpected events and safety assurances, et cetera, that is much more nuanced than just the high-level technological debate that is being made on public stages.

So I think we have a significant advantage in being fairly advanced in this process. And this will prove, we believe, as a competitive advantage in the next few years as being one of the only, if not the only robotaxi enabler in the European market, which in and of itself has potential of tens of millions of commuters.

Our next question comes from George Gianarikas with Canaccord Genuity.

I was wondering if you could comment on some of the recent traction that NVIDIA has seen with their reference design and what your pitch is to OEMs in terms of total cost of ownership and why they should pick your solution?

At the end of the day, it's a combination of performance and cost. If you refer to Alpamayo, we downloaded Alpamayo. It doesn't seem like a production-worthy system. It's something nice to play with, but it's not anywhere close to production worthy. Whether an OEM can take it and upgrade it or refine it for production worthy system is yet to be seen. I would add that 2016, NVIDIA had something similar with pixel labeling that they announced open source for the automotive industry, OEMs, nothing really -- there was no real traction for it.

Bringing something into production is tough. Taking a demoware or taking a nice demo into production, there is a death valley in between. And this is something that Mobileye is very good at. This 2,000-kilometer expedition that I mentioned in my script, it's very meaningful. It's an OEM taking a number of competing systems. One of them is the Mobileye system with Porsche, which is not yet ready. It's maturing over time. It's maturing this year to be ready for start of production, but it's not yet fully matured. And doing a 2,000-kilometer expedition without us knowing the route in advance in very significant weather conditions, bad weather conditions, urban, suburban, highways, day and night and snow and our system really excelled.

So this shows that going from demo to production is an art. It's a science and art and something that Mobileye excels in. So it's not just a matter of here's an open source network that does something cool. Can we then refine it and bring it to production. Just to mention that the production program we have with an OEM has about 60,000 requirements. This is what it takes to go from a demo to a production vehicle. And this is one of the strengths of Mobileye. It's not only that we are experts in AI, that we built an AI systems, we're experts in machine learning. We have the cost-optimized solutions. We know how to bring stuff into series production. And this is difficult.

And maybe as a follow-up, there's a lot written about Volkswagen and their future strategy. I was just wondering if you could please comment on your relationship there and their commitment to deploy your solutions over time.

Yes, I can take this. So I think ultimately, the reality today is that all of the upcoming SOPs, product launches across all brands of Volkswagen Group mostly, spanning from base ADAS in lower-priced vehicles to robotaxi and everything in between, all of the upcoming SOPs are with Mobileye products. And this is the plan of record. It has been a plan of record in the past couple of years, and it did not change. If anything, we managed to expand our business with Volkswagen in these 2 years, also winning strategic projects for the base segment, introducing strong ADAS for the first time with Volkswagen Group on very high-volume vehicles.

And we are seeing kind of pulling additional vehicle platforms to the already nominated products we have with them. So I think we need to kind of distinguish between some like news that comes out for -- that serve certain interest to the realities of their planning schedules. And our experience in this industry shows that the first thing to change if there is indeed a decision to take a different product is these planning schedules when they did not change. If anything, they change for the better for Mobileye.

So we're not seeing any evidence of change, of course. We're not seeing risk to our existing projects as a consequence. Of course, we need to finish the execution and to get to the SOP date, but the business opportunity remains very, very significant for us when we will finish this execution.

Our next question comes from Shreyas Patil with Wolfe Research.

Maybe, Amnon, just to follow up on some of your earlier comments. I'm just curious what you're seeing in the pipeline amongst OEMs. From the outside perspective, it does seem a bit jumbo. We've seen Mercedes and BMW appear to be pulling back from L3 in Europe, focusing on effectively SuperVision like products. Ford and GM are talking about deploying their own solutions within the next 3 years. Others are partnering with AV players such as Nissan and WAVE. So how many opportunities are actually available to pursue in areas like SuperVision and Chauffeur in your view? Or have OEMs sort of laid out their plans for autonomy over the next few years?

I think, by and large, OEMs have not yet made up concrete plans. We see opportunities for SuperVision. We see even more opportunities for Surround ADAS. With Level 3, I believe we'll see the bigger opportunities with Level 3 as we get closer to the production with Audi on Level 3 or as we get the driver out of our robotaxi and also showing a significant cost reduction of the robotaxi stack, which we have -- which we can show by the end of the year. So SuperVision and Surround ADAS, we see significant opportunities. But with OEMs, it takes time, and we cannot predict the timing at this point. So our focus is really the execution. Execution will bring more opportunities.

Okay. Great. And maybe just a quick modeling follow-up. I think you talked about higher DRAM costs for this year. Maybe if you could help quantify that. And is that something you can pass along via price adjustments?

So the DRAM is the responsibility of our Tier 1s, Mobileye just sells the chip.

Yes, the DRAM, we spoke about, yes, it's in the SuperVision area where we buy the memory directly for the issue. But this is -- it's a relatively small business. We're talking just about a few millions. And yes, we are passing that through to customers, but this is -- it's a very -- the dynamic there is changing. So it's really -- it's not something that is expected to impact significantly on our cost, but it is a few millions currently.

Our next question comes from Mark Delaney with Goldman Sachs.

The company spoke to the performance of its preproduction vehicle in the U.S. with EyeQ6 High. You spoke to that doing well across urban, suburban and highway settings and achieving your mean time between failure objectives. Can you remind investors what Mobileye is targeting for MTBF for this product, how it compares to competitors? And maybe most importantly, given what you were able to see on the unplanned route, is it catalyzing any incremental OEM business interest?

We are not sharing our MTBF goals. SuperVision is an eyes-on system. So MTBF is important, but not crucial as it is important for robotaxi. There are other KPIs like comfort, not only disengagement, but also comfort ODDs, what kind of ODDs can you satisfy? There's a long list of requirements. It's not just one number that determines the driving experience of the product. So it's not something that we can share with the public. It also changes from OEM to OEM and the OEM has a lot to say about the driving experience because they set the requirements.

So it's not only the base technology that determines the driving experience. There's a lot that goes into it. But what I can say is that this 2,000-kilometer expedition has shown the excellence of our product, even though the product is not yet finished, especially compared to other competing demo systems that were part of this expedition -- means it shows that the gap, the discrepancy between all the talk that you hear and the actual performance is huge.

My other question was related to Mobileye's efforts in AI. And now that you have Mentee Robotics transaction completed, can you speak more to the synergies between the existing Mobileye efforts in AI, what Mentee brings? And are you able to work better jointly to accelerate your efforts in real-world AI?

We are planning an AI Day around the July time frame, where we are going to lay down our complete vision of AI. Just to give the perspective, the system, the software running today on our EyeQ6 High internally, we call it Gen 1.5. By in about 2 months, it will be Gen 2.0. And by the end of the year, it will be Gen 3.0. So we're working very fast on software rewrite in order to accommodate the best AI has to offer, whether it's Gen AI, whether it's simulators, everything. And we'll be very transparent about it in our AI Day. So we expect around July kind of a consolidated view of how we take modern AI and bring it into physical AI, both in terms of robotaxi and in terms of robotics.

Our next question comes from Luke Junk with Baird.

First, I wanted to ask, just bigger picture, as already referenced, there's been just a lot of chatter about OEMs pulling back from L3 applications and refocusing on L2+. And just wondering if you're seeing any broader applications of this, specifically in terms of Surround ADAS and the amount of interest you're seeing at the front end of the funnel. It seems like it's really an area the market is consolidating around right now.

I believe that we do have -- we are engaging with OEMs on Level 3. But I would say that Level 2+ our SuperVision is gaining more traction with OEMs and Surround ADAS is gaining even more traction with OEMs. I believe that Driver Out with robotaxi, especially when you have a cost down path, a credible cost down path will reignite Level 3 and Level 4, consumer Level 3 and consumer Level 4 programs with OEMs. Maybe you want to add something there?

Yes. I think that the debate around Level 3 is not new. It's been going on and off, and there has been cycles of, let's say, excitement versus skepticism for 10 years now. Ultimately, it's a very challenging product because it requires the robotaxi performance levels, but a privately owned vehicle, so the cost is supposed to be significantly lower in a much more efficient system. And also the -- in order to have a useful product, it needs to be available in a broad enough ODD or at least in a broader enough area to satisfy the needs of the consumers.

So we believe that our product, the Chauffeur product with Audi, which is progressing well, is going to satisfy these key requirements. As we get closer to -- or as we progress with execution, we'll be able to show this and expose this to the OEMs that it's not an if question, it's a one question and the one is imminent because I don't think that any OEM has question marks on the value proposition to consumers. I think it's a consensus that the ultimate value proposition to consumers is eyes off and mind off and giving back time to driver.

This remains a compelling case, although OEMs maybe are more cautious in going all in and developing this when it's not clear there is an available solution. And we believe that we will be providing this available solution very quickly relative to others. And this can reignite the momentum with OEMs.

Maybe a related question. Some OEMs of robotaxi offerings have been recently really trumping the advantages of their data collection efforts. And just curious to get an update on where Mobileye is making strides in this regard in terms of extracting more data and run and maybe some of the specific benefits of your test fleet, both the advanced products and robotaxi.

We have no shortage of data. As we mentioned back a year or 2 years ago, we have hundreds of petabytes of data that we can leverage for our development. Not only that, we added simulators that can run billion of hours of driving experience overnight. I talked about it at CES. So it's not that we lack data. We're just doing with the robotaxi. We just need to do the validation with the final hardware in terms of the vehicle platform, and then should be done in a few months from now.

Our next question comes from Gary Mobley with Loop Capital Markets.

I wanted to ask you about Surround ADAS. Perhaps you can give us an update there. But more specifically, looking at your top 10 OEM customers, what percentage of those have committed to conversion to Surround ADAS? And maybe you can give us an update as to the timing or contribution for revenue from Surround ADAS and the ASP impact.

Yes. So I think our first Surround ADAS design win announcement was roughly a year ago, and it was with Volkswagen Group, which basically committed to upgrade their entry fleets for -- to Surround ADAS starting 2028. That is -- just some kind of reference numbers, the average ASP is around $100 to $150. That's the range we referred to in the past. So with a similar gross margins to our base ADAS volume, which is roughly 70%. So that numbers remain.

Over the past 2 quarters, we managed to add 2 additional OEMs. So we have today 3. One is the major U.S. OEM that we announced back at CES, which in a similar fashion to VW decided to upgrade the entire electric fleet to Surround ADAS from base ADAS today, actually with a higher ASP than what VW has with more content. And recently, we also announced Mahindra, the first Indian OEMs to adopt Surround ADAS. So now we have 3, 2 of them are today our top 10 customers. And we believe that Mahindra represents a significant growth opportunity given that the Indian market is just now starting to adopt ADAS.

In India, less than 10% of vehicles have ADAS at all. And regulation coming up in 2027 is expected to accelerate this to the higher 90s in just a couple of years, which is a huge organic opportunity for us. And through this product with Mahindra, we can benefit and be a market leader in India. So zooming out and Surround ADAS, I think in just a year to have 3 design wins, 2 out of the top 10 OEMs with significant volumes. This in and of itself without new design wins can represent when these will be launched, more than 10% increase in revenue on a yearly basis.

So of course, as this gains momentum, as we make progress in execution, which we are, we show this to more and more OEMs. We expect this to generate more interest and these growth numbers can be even improved in the future.

I appreciate that color. And for Moran, I had more of a housekeeping question. Can you give us some context around the goodwill impairment charge in the quarter?

Yes. So in Q1 versus our previous valuation from December, market cap has went down like 35%. So we had to do -- this was a trigger for impairment assessment in the quarter. I have to say this goodwill is kind of unique in its nature since its goodwill pushed down to Mobileye from Intel on the acquisition of Mobileye in 2017. So even initially, it was a very significant portion of our net assets, which is not something reasonable for a company to have goodwill on its own assets.

I can say on the valuation itself, we recognized a goodwill impairment of $3.8 billion. On the business aspect, we kept the same projections, but reflected a higher risk premium because of macroeconomic environment, geopolitical environment. So that has impacted the valuation, and we recognize this impairment in Q1.

Our next question comes from Aaron Rakers with Wells Fargo.

I wanted to ask first on kind of SuperVision. I apologize if I missed it. Can you help us appreciate the volumes that were shipped this last quarter in SuperVision? And how do we -- any updated views on the volumes as we start to think about the Porsche ramp going forward as we move through '26 and into '27?

Yes. So we delivered in Q1 20,000 units. We are seeing stability in demand. 2025 was high in SuperVision. For Q2, we estimate 15,000 units, so roughly the same number. We are still pretty conservative for the second half. And for the full year, we're still estimating like 50,000 units or a bit more kind of consistent or a bit lower number than in 2025 in case that demand is changing or there is any further impact, but it's not something that we're seeing. So orders keep coming and this is business with stability for the last few quarters. And as for Porsche, we're not anticipating any Porsche volume in 2026.

It will start -- the ramp-up will start in 2027 towards the second half of the year.

To recap, we did not change our SuperVision volume assumptions for the year.

Perfect. And as a quick follow-up, I want to go back to the memory question. I know that you guys talked about your partners, obviously, handling the pricing dynamics. But at a higher level in the current situation that we're under, are you seeing any risk from just actual supply of memory impacting any of your OEM customers or your partners from a procurement perspective? Is that a headwind that we should think about? Or have you not seen any of that?

So we did not see direct reporting or direct planning from our customers to accommodate for this. So our revised guidance reflects the recent discussions we had with our customers. And of course, they would -- they baked in all of these risks into their current predictions and estimates. Of course, we need to keep close tap on the situation and monitor it, but we are not seeing any direct imminent change.

Maria, this next question will be our last question today.

Okay. Our last question will be from Steven Fox with Fox Advisors.

I'll try to make it a good one. I was wondering if you can go back and maybe expand on the initial comments you made in the prepared remarks about India. It sounded like you were saying you're more bullish about it. How much -- and if you could talk about why and whether there's any influence potentially down the road from your position with Chinese exports?

Yes. So the Indian market has been lagging in terms of ADAS adoption rates compared to Europe, U.S., China, Japan and Korea. Recent numbers suggest roughly 8% ADAS take rates in the Indian market, which refers to vehicles sold in India by both Indian OEMs and foreign OEMs. Just for reference, the Indian automotive market is roughly 5 million units per year. So in a pure size, it's a very significant opportunity.

There is a regulation coming up in 2027, which is expected to incentivize and mandate OEMs to adopt ADAS solutions starting 2027, and we expect this to increase the ADAS penetration rate from the 8% it is today to 70%, 80%, 90% in a few years -- in 2 or 3 years.

Now we are today very strong with Indian OEMs. It's the 2 major Indian OEMs. This recent announcement on Surround ADAS and SuperVision with Mahindra reflects the kind of the strength and leadership position we have and also that the Indian market is not necessarily just for entry solutions, but also for more advanced higher ASP products as well as there is more and more demand by Indian consumers for advanced functionalities.

We think that for -- in Mahindra's case, for example, ADAS has been ranked as one of the key reasons for Mahindra's increased sales year-on-year. They have been growing very fast, and their customers vote for ADAS is one of the reasons for it. So we believe that there is a strong demand by consumers. There is going to be a regulatory push and just the sheer size of the population all suggests that it can be an organic growth opportunity, and we're very well positioned by not just Mahindra, also others that are selling into the Indian market.

We have reached the end of our question-and-answer session. I would now like to turn the floor back over to Mr. Galves for closing comments.

Thanks a lot, Maria, and to the Mobileye management team, and thanks, everyone, for joining the call. We will talk to you next quarter. Thank you.

This concludes today's teleconference. You may disconnect your lines at this time. Thank you for your participation.

Greetings, and welcome to the Mobileye Fourth Quarter and Full Year 2025 Earnings Call. [Operator Instructions] As a reminder, this conference is being recorded. It is now my pleasure to introduce your host, Dan Galves. Mr. Galves, please go ahead.

Thank you, Donna. Hello, everyone. Welcome to Mobileye's Fourth Quarter and Full Year 2025 Earnings Conference Call for the period ending December 27, 2025.

Please note that today's discussion contains forward-looking statements based on the business environment as we currently see it. Such statements involve risks and uncertainties. Please refer to the accompanying press release, which includes additional information on the specific factors that could cause actual results to differ materially.

Additionally, on this call, we will refer to both GAAP and non-GAAP figures. A reconciliation of GAAP to non-GAAP financial measures is provided in our posted earnings release.

Joining us on the call today, as usual, are Professor Amnon Shashua, Mobileye's CEO and President; Moran [ Shamash ], Mobileye's CFO; and Nimrod Nehushtan, Mobileye's EVP of Business Development and Strategy. Thanks. And now I'll turn the call over to Amnon.

Hello, everyone, and thank you for joining our earnings call. As I look back on 2025, there are a number of meaningful positives to highlight both for our company and the industry.

In a very uncertain geopolitical environment, demand for our products came in higher than expected throughout 2025, demonstrating the resilience of the auto industry and our product offerings.

Results were quite strong with revenue up 15%, adjusted operating profit up 45% and operating cash flow up more than 50%. The industry began to clarify the structure and features of the next generation of ADAS for mass market vehicles, several forces are coming together here. Demand for incremental safety, demand for convenience in the form of highway hands of driving and the need to consolidate the technology on a single [ ECU ] to keep the systems cost low. Mobileye's EyeQ6 high chip is very well positioned, and we won the first 2 major programs with two of the biggest OEMs in the world.

Waymo's commercialization provided a number of supporting proof points on consumer acceptance and demand for autonomous mobility services. This led to a major uptick in demand signals from transportation network companies and public transport groups, which led to an expansion of expected volume through our Volkswagen ecosystem to 100,000 units by 2033. We are now 1 year closer to the launch of our advanced products with the Volkswagen Group. We expect the first major public milestone to be removal of the safety drivers in MOIA in MOIA's robotaxi fleet in 2026.

We are implementing a unique first think slow-set structure to our advanced products that we believe accelerates both precision and scalability. This includes novel technologies like vision lag with semantic action models and artificial community intelligence.

And finally, Mobileye took a decisive step to expand its footprint into the humanoid robotics field with the acquisition of [ Menti ] robotics. [ Menti ] has achieved a fully vertically integrated low-cost, highly capable robot that has a clear path to commercialization into the structured environment of industrial and logistics services field and with its distinctive technology to cater to unstructured environments like home use cases.

Aside from our 2025 results, we detailed all of these areas in my CES talk on January 6. I encourage anyone with an interest in Mobileye or just physical AI in general to make sure to view that presentation.

Turning to guidance. Moran will spend some time on it, but I'll address it briefly. We're encouraged by the volume growth we are expecting despite global auto production that's expected to be flattish again. And while we don't expect the volume levels of Q1 to be sustained throughout the year, it's a strong signal for the year and order flows for Q1 has been rising for the last month or 2.

Turning to technology. At CES, I talked a bit about the debate around approach, specifically this concept of data in command out, which is a false debate because no legitimate actors in our field are actually doing that. There's always a need for structure and architecture and everyone's architectures have evolved given advancements in AI over the last few years, including ours.

We introduced two new innovations that are accelerating our path to position scalable autonomous vehicles. One is Artificial Community Intelligence referred to as ACI. This is a simulation concept using a sales play reinforcement learning technique that we are using to train our planning engine, also known as driving policy. This is the first ever productization of a technique propulsion academic research. A strong motivation for ACI is that the sample complexity for planning is much higher than for perception because the multi-agent nature of driving where actions that you take will impact the actions of other world users. Therefore, the amount of data one needs to collect could be [ unwieldily ] even for large data collection fleets. At this solution, we have created simulators that can achieve 1 billion hours of training overnight.

Mobileye has unique advantages here since our [ REM ] maps, which cover much of the globe can be used as a realistic and diversifying structure for the 20. The other advantage is we have developed sophisticated [ same to wheel ] techniques that have the required understanding of the noise model of our perception engine when transferring the driving policy to the real world. That [ same to wheel ] technology is also very relevant to humanoid robotics and will be a key area of technology sharing between Mobileye and [ Mentee ]. We also introduced a fast thing slow thing concept that utilizes specialized vision language models to provide contextual information and to address robustness to vehicle decision-making.

This is not necessarily about safety. It's more about understanding the semantics of complex in. For example, a field where a policeman signaled that the role they would like to take is blocked. The safety layer ensures that we won't hit the policeman, but we also need to understand the them, figure out that we shouldn't try to overtake the police member rather, we should either wait or take a different route.

This is what slow thinking gives. Since this is not safety critical, the contextual information can be imputed into the system at a lower frequency than perception, which is typically analyzed at 10 trains per second.

Structuring our architecture with fasting and sloping components, states compute and even brings use of cloud-based compute into the picture. As a result, we can put a very sophisticated [ VLM ] on the in-car compute, but call on much bigger [ VLM ] in the cloud when the situation evolves. This has been very positive -- this has very positive effects on the meantime between intervention metrics, but can also eventually lead the scalability benefits in terms of cars per tail operator as the [ VLM ] can replace a human teleoperator in many cases.

Turning briefly to our announced acquisition of [ Mentee ] Robotics. Most of the AI that humans are using every day is in the digital world. The two main applications of AI in the physical world, our autonomous vehicles and robotics. It makes sense for these two expressions of physical AI to be together because there's a great deal of technology overlap. Both extensively use computer vision and control, cash flow thinking concepts, make heavy use of [ VLMs ] and extensive simple techniques.

Menti itself compared to other companies we evaluated had a superior combination of strength, including a high level of vertical integration, a pure AI approach with ability to demonstrate high-level capabilities with no teleoperation, a design strategy that results in an optimized cost versus usefulness ratio and above all, a distinctive AI technique continues to do continuous on-the-job learning from patent demonstration.

A truly practical approach to capitalize on the most near-term industrial and logistics markets and then expand to more challenging markets over time. We believe access to mobilized tools, simulation and data training infrastructure will accelerate Mentee's development. And the number of technologies developed for robots, such as self-play simulation and [ same to wheel ] techniques and also bolster mobilized [ AV ] development.

Finally, there is potential for catalysts as we continue to demonstrate the strong capabilities of the [ Mentee ] Robot and execute on customer proof-of-concept work in the near term. I will now turn the call over to Moran.

Thank you, Amnon, and thanks for joining the call, everyone. Before I begin, please be aware that all my comments on profitability will refer to non-GAAP measurement. The primarily excluding mobilized non-GAAP numbers is amortization of intangible assets, which is mainly related to [ Inter ] acquisition of [ Mobileye ] in 2017.

We also exclude topic commutation. Our full year 2025 revenue of $1.9 billion slightly exceeded the high end of our car guidance. Fully revenue was up 15% year-over-year. Compared to our original guidance of 6% growth at the midpoint. It was a very good year where a combination of minor upsell in global production trends ice program ounces and higher-than-expected ADAS and supervision volumes from China [ OEM ] on led to significant growth.

Full year adjusted operating income was $280 million, up 45% year-over-year. and margin was 15%, up about 300 basis points versus 2024. The fourth quarter included a nonrecurring expense of $7 million related to workforce efficiency initiatives we undertook in Q4. That expense was not part of our guidance as of the October earnings call. So if you exclude that, adjusted operating income would have also been slightly above the high end of the guidance.

Like I said earlier, we saw consistent positive revisions from our customers throughout 2025. and we've continued to see that over the last month or 2 during our 2026 planning case. 2025 EyeQ volume was $35.6 million across the full year. which was well above our original expectations of $32 million to $34 million. We've seen a fairly consistent demand trend of 9 million units recording with some minor fluctuations across quarters. For example, in 2025, Q2 and Q3 were higher than trend. Q1 and Q4 were a bit lower. One more point on Q4 before turning to the future, modest upside to the high end of our prior guidance was related to higher-than-expected supervision in IQ volume was consistent in the high end of our guidance of about 8.2 million. This level when the start of the quarter looked a bit below the demand trend and our customers' desire to end the year with NIM inventory, but the demand trend in Q4 ended up higher than we expected. As a result, we believe that inventory as our Tier 1 customers and in 2025, extremely low. We believe there is some level of adjusting safety store that will occur in Q1 to get back to normal level.

We expect about 10 million EyeQ units shipped in Q1, which supports an outlook of approximately 19% year-over-year growth in the first quarter. After that, customer forecast indicates a reversion to the trend of slightly above 9 million units per quarter.

Turning to the full year guidance. we are expecting revenue in the range of $1.9 billion to $1.98 billion, representing flattish to 5% growth. The midpoint of our guidance incorporates [ RQ ] volume of slightly above 37 million, which again consists of 10 million units in Q1 and an assumption of a bit over 9 million per quarter in the balance of the year.

If we look specifically at our top 10 customers, we are assuming that the overall production is down 2% our volume with those customers is up 6% at the midpoint. This includes about 700,000 units for a new OEM program that requires two EyeQ4 chips per car, that program is clearly a positive, and we will generate higher gross profit in dollars per vehicle. This is the second chip is lower price relative to the first. It has an impact on overall ASP and gross margin.

For Chinese OEMs we are expecting a decline of about 0.5 million units compared to 2025, which was a bit above 3 million. We are encouraged by the significant growth in China OEM volumes in 2025. It is aligned with the export volume growth, the area where our business is the strongest with those customers. We've seen all reasons why that wouldn't continue into 2026, but prefer to remain conservative given we only have a short-term visibility into order flow with China OEMs.

Gross margin will be down some out on a year-over-year basis, driven by continuation of EyeQ5 related cost heavy. We discussed this on the October earnings call as an impact to the second half of 2025 that would continue through 2026 and then will gradually decline beginning in 2027.

We also have modest vehicle mix headwinds and the impact of the [ Dolce ] program mentioned about.

Turning to operating expenses, 2025 ended up at $1.03 billion. This was slightly above our original budget of $995 million accounted for by the nonrecurring termination related bookings in Q4 mentioned above. In 2026, we are expecting around $1.1 billion or 10% growth. The underlying growth in OpEx is approximately 5%, consisting of normal salary and benefit inflation as well as additional infrastructure to support execution of the advanced products in 2026 and 2027. On top of that, we are including in the R&D expenses.

Finally, we are experiencing an FX headwind related to appreciation of the Israeli currency, versus the U.S. dollar, that meaningfully raises our head count cost in dollar terms. This is being mostly offset by the workforce efficiency initiatives noted above, but not completely. To conclude, we are almost 1 month into 2026 and continue to see positive demand signals from our customers on the core basis.

As Amnon discussed, we are also seeing very good execution progress ahead of a large number of advanced product launches over the coming 1 to 2 years as well as accelerating momentum in customer demand from next gen, higher ASP ADAS and the transformative robot taxis. Thank you, and we will now take your questions.

[Operator Instructions] Our first question today is coming from George Gianarikas of Canaccord Genuity.

I'd like to ask first maybe on the -- on your view on the competitive environment, particularly in light of some announcements at CES from NVIDIA and others, just your view on what's happening in the advanced economy solution space.

So I think that we have been -- we've obviously seen a lot more announcements and excitement around the van solutions and autonomous driving in general, and also robotics, it was one of the key things at CES this year for everyone who attended. We still believe that we are closer to launching our advanced products than other competitors. And this is one of our strongest advantages, combined with the maturity of our technologies and the advancements of our technologies. And we are, as we said, 1 year closer to launching a spec from a product that stands from [ Solana ] supervision to offer a robotaxi starting from '26 and through 2027. We believe this will be a major transformation for kind of positioning mobile in the market. It's having proven product on the field.

Right now, there is a lot of demos, a lot of referrals to technologies and emerging technologies, and there is some, we think, noise and maybe simplistic description of some of these technologies and how useful they could be for a reliable system. There is a recent announcement by NVIDIA about their open source model, [ Altimeo ] that they announced and supposedly given others' ability to -- [indiscernible] maybe want to say a word about this. But we don't see that as something that changes and about positioning in the market leading.

And maybe I can ask a follow-up on Mentee specifically. I mean you mentioned yourself that there were a significant amount of start-ups and competitors at CES in the humanoid space. I'm just maybe a synopsis, a brief bullet point or 2 as to what the differentiation from Mentee will be as you try to attack the marketplace and commercialize the product.

I think the other backups mostly in China, a lot of sum-ups in the area of humanoids, many of the demonstrations that you see out there are teleoperated. Now the -- to win this game, you need to have a fully autonomous control of the robot from perception to action, to understanding the team, having an eye back that can control the global autonomously. And this is a Mentee has been demonstrating quite consistently over the past year, 1.5 years. And the [indiscernible], as I showed the number of trips, Mentee is also fully vertically integrated with the design of the actuaries the year. The AI itself, all the software component, the electronics, which is which is the crucial if you want to have a true end-to-end system.

Another, I would say, distinctive element is the ability to do continuous learning. So Mentee has developed an AI technology that allows the robot to passively use a human performing the task and innovate that task in a very, very short period of time without having any special equipment, now VR goggles or special suits, just passively observing a human performing the task. This is, I think, very important as we move from structure to unstructured environment like home use.

So taking everything together, we have here a company, that is both the thinking practical, what is going to be the first domain launch, which has structured environment like fulfillment centers, assembly, assembly plants, retail and also developing the technology for the next deployment for unstructured environments like home lease fully vertically integrated, very strong component, whether it's reinforcement [indiscernible], very strong, very interesting overlap technology overlap with Mobileye that can go both ways to the synergies.

So overall, this is a very good step for Mobileye to take a decisive steps towards owning physical AI in its holdco.

If I may add to this, I think to kind of differentiate between the different actors, Mentee as we know, the -- probably the only Western Humanoid robot company that is actively engaging with customers on proof of concepts and pilots that involve pure AI operations with no remote operation. We show something about the advancement of the use of robots in a setting that a customer is willing to evaluate and deploy in kind of a non-sterile setting. Unlike maybe some height videos that show something on a outlet. This is an actual testing environment is a different stage of maturity and in engagement with potential customers. .

And we believe that through integrating mobilized Technologies, we have obviously strong strength in computer vision in AI using cameras and using sensor fusion and designing a systems for safety and reliability and how to integrate system in a very cost-efficient manner and efficient compute, all of these will help them even accelerate the progress they made so far.

The next question is coming from Mark Delaney of Goldman Sachs.

Yes. For Surround ADAS, the company has already reported on some strong momentum. You spoke about the two big OEMs that have already committed and given serious production awards. As you look at the opportunity set in 2026, could you give a bit more details on the number of OEMs you're engaged with for Surround ADAS? And how many might be able to convert into awards this year?

So I think the important point about Surround ADAS is that this is a product that addresses a very clear pain point for customers and for OEMs. I mean, in the sense that it simplifies the system, it reduces costs, it provides advanced functionality needs each regulation. So it takes most of the boxes that OEMs want to take for the high-volume vehicle segment in the upcoming years. .

Therefore, we -- the first OEM that we announced with Volkswagen was kind of starting a trend that created a flywheel effect of more and more OEMs being interested. Now having announced the second design win with -- we now have two out of the top 6 OEMs in the planet in major markets, adopting this and launching this in a few years. This has definitely created a stronger realization amongst other OEMs at this has to happen for them also, at least to some degree.

We have seen an increase in the amount of engagement we have I don't want to predict timing and quantities, but we're definitely encouraged by the increase in different engagements we have with multiple OEMs across our customer base.

And we also believe that we have inherent advantages for this product category because it requires a very reliable system performance, very high safety standards, advanced functionalities like hands-free driving in primary roles and so on, but also be extremely cost efficient. And just to give you some sense, these two programs we won are going to be integrated in the kind of a standard trade across the highest volume in vehicle categories for these two OEMs.

So every dollar counts, and also the implications for the OEMs to adopt this product means how much conviction they have that they need such a product for -- it's not a balloon project in a small amount of vehicles that it failed then nothing happened. If this project is delayed, for example, this is obviously affect the entire vehicle portfolio. So it shows about the confidence that the Mobileye much conviction they have in this product, and it's definitely an encouraging sign.

My other question was on Menti. Given the announcement and engagements that you've had with potential customers and industry participants. Can you help us better understand to what extent it's catalyze additional interest in partnering with Menti Robotics including opportunities to have your humanoid robots in factory and commercial environments to gather data. And as you think about that 2028 commercialization target you shared at CES, how important is that data collection and gathering for hitting that time frame?

I think it's a very interesting question. We've had since 2 weeks since the announcement at CES, and we have -- I've received reach-out from a significant number of customers asking about our interest and readiness to support on-site pilots and proof of concepts and kind of starting from our industrialization partners that want to contribute in manufacturing and components because they understand we do the full robot starting from that and really trying to attract us to work with them for manufacturing and for all of our industrial partners, whether it's Tier 1s, OEMs and others that want to see how they can work with modeling integrating robots into their logistics centers, warehouses, manufacturing lines, the need is definitely there. And for them, it made perfect sense. I think one of the encouraging signs that we've seen is that already at CES, we've had meetings with OEMs. And in most of these meetings, it came up, let's take a follow-up and schedule when we can actually talk about a plan to deploy this in our environment.

And I think that the fact that the no mobilize from the automotive business and they trust the standards of the company and that they have a need for longer-term finding solutions for human labor that is becoming a bigger and bigger problem for them, especially in developed countries. This gives them an easier path to evaluate a new technology with a partner they trust as opposed to working with the start-up in humanoid, who knows what you can get from them and whether or not they can deliver. And definitely, we're leveraging these relationships. So we definitely think of this as an area to continue to develop in the next few months.

Our next question is coming from Chris McNally of Evercore ISI.

This is John [ Sagar ] on for Chris McNally. Amnon, you made the sandwich analogy for ADAS and Navy demand, basically with high demand first round at the low end, or drive at the high end. If we could focus on just drive for the time being, you guys announced VW MOIA, one of the two big partners in [ Marubeni ] and an unnamed ON, but the forecast is for a fleet of 100,000 AVs by 2033. And obviously, a bit of a ways away.

So my question, can we give a sense for what the near-term demand for your drive system might be for just like the next 2 to 3 years, '27, '28 or on Phase 1 for a growing -- on the growing list of cities?

Well, we announced together with Maria cities to expand the 60 cities in 2027, and that includes as Los Angeles together with Uber. We have another high-volume program with Holland that will come 6 months later, have also its expansion.

As the CEO of [ A&P ] on stage, mentioned that they foresee about 100,000 vehicles in the next 8 years. The exact numbers of the rollout will depend on the success of 2027 deployment of the first six cities, but we are talking about thousands of vehicles at this point.

Just to add to this, I think it may be somewhat challenging to understand what it means 100,000 because it sounds like a big number. I think what we're taking away from this, what it means is that Volkswagen has in place the manufacturing capacity to produce as many vehicles as needed. The 100,000, if we're successful in 2026 and '27 and then in '28, which we have high confidence in our chances, means that 100,000 can also be a small number in hindsight. The manufacturing capacity they have and the funding they've moved into this in the past few years to build everything needed to produce robotaxis in scale eventually is Volkswagen. So they can produce 10,000 per year, 50,000 per year, 75,000 per year when the demand will be there. And the demand from mobility operators, P&C municipalities is far greater than tens of thousands per year globally.

While the technology gets to this maturity level and allows quick economic and geographic expansion, which we believe we have clear advantages in, then the numbers will -- the demand will not be a problem. And we have a partner that can scale and give the supply at the best extent possible.

Understood. And just how should we think about like the volume for a Phase 1 launch? Like should that be like 1,000 to like 1,500 like Waymo in San Francisco? .

You can think of it as a few hundreds of vehicles per city as a good testing is a good debt measuring stick. Just also seeing Hawaii rollout that's roughly the numbers they've had in some cities is 200, some cities close to 500. That's a sufficient number to kind of facilitate for the mobility demand in that city and also to build a meaningful business. And that's also roughly what we're planning.

Just one last follow-up. Do the AV customers pay for anything before the purchase of the $45,000 drive content like R&D in advance? Or do you get any protection if their volumes are less than planned?

Without going into the details of our contracts, we are receiving -- we're delivering samples and engineering samples throughout the year, and there is an engineering budget that covers the direct engineering costs and development costs. So there is definitely a good amount of investment well before the commercialization.

So I think when you have high confidence in our chances of getting to driverless, I think we're not that concerned about the downside potential.

The next question is coming from Joseph Spak of UBS.

Thanks. Hello, everyone. Just to maybe talk about a couple of more near-term things. Obviously, memory has become a larger issue and concern in the automotive industry. And I know -- or I believe you don't really buy a lot of that memory directly, but clearly, it is used in the modules at your Tier 1 customers to assemble to sort of ship on to the OEMs. So I'm just curious, what you're hearing from your customers and the supply chain and whether this is really a pricing issue? Is it an availability issue? Is there any sort of volume risk embedded in your outlook? And because -- even if it's a pricing issue, I guess do you see any risk of decontenting?

So as you said, Joe, we're not -- let's say, the exposure that we have is not direct because we're not purchasing a lot of these units. It's mostly indirect through the fact that our Tier 1 customers are purchasing memory components. .

What we've been doing in the past few months, and we have been actively working on this, I think well before it was a public knowledge that this dynamic is developing, is to create of maximizing our supply of these components and working with multiple vendors to ensure that we have enough flexibility to kind of mitigate the direct cost impact from specific vendors and that we will be able to ensure that vehicle manufacturing will not be impacted by these fluctuations.

And we haven't -- I think that like we did last year, our forecast for this year is maybe opting for the conservative side. You can see the difference between Q1 and other quarters. It does bake in some level of understanding that there is some volatility in the industry. So we want it to be on the more conservative side. But we haven't seen any, let's say, direct evidence or indication that there is an imminent change to volume as a consequence. But we will keep close monitoring lines as it develops, and we're doing everything in our power with our Tier 1 partners to create the availability of these components.

Okay. The second question, just on -- you mentioned some of the appreciation of the shekel. And I know you get very helpful exposure in your 10-Qs on what a change in that currency can do to your cost base. But I believe also like at this time a year ago or earlier in '25, you made a comment on one of the calls about how a lot of the costs on the shekel were hedged. So did something change with the hedging strategy? Like maybe you could just sort of update us on sort of why it's a little bit maybe more of an issue now than you thought a year ago?

Yes. So I will start with 2025. So we have a hedging plan but basically cause that we can meet our OpEx expectation for 2025. So for example, for the second half of 2025, the rate that we had in our financials was like 5% or 6% favorable than the average macro trade. So -- and these are transactions that we made in the beginning of 2025. But as appreciation of the shekel continues into 2026, and we're talking about, I think, 10% or 12% in the last year. We still have hedging in place for 2026. So we are more than 50% hedged on our payroll expenses at a favorable rate. But the risk is obviously heavier as deterioration gets bigger, but we took it into account in our guidance.

So we took into account some further hedging, but it would be at a favorable rate, the fact that we are already more than 50% hedged. I think we're in a good place in terms of the rate, but it's still the year-on-year impact because it's a significant impact. It's worth mentioning.

The next question is coming from Aaron Rakers of Wells Fargo.

I wanted to kind of double click on the Porsche and VW and the Audi kind of programs. I'm curious as you kind of thought about your guidance for this year, I think the initial expectation was maybe early volumes on Porsche late this year. Just give me an update on where we stand on some of those programs and how we should think about volumes, appreciating that I think 2026 in the past has been more characterized as an execution year.

2026 is an execution year. The supervision on Porsche and Audi should start Q1 next year, Q1 2027. There was some pushback of deadlines unrelated to us that pushed the SOP to the first quarter of 2027.

And just to say that it didn't really change the plans of the project. It's just a 1-month change between December 26 to February 27. So it's not really material. And we think made clear in previous calls as well that we do not expect meaningful volumes from these programs in 2026.

Right. appreciate that. And then as a quick follow-up. In the prepared comments, you talked about inventory levels at your customers, your major OEMs being fairly lean. I know you've guided 10 million EyeQ units this quarter. I'm just curious, can you kind of go a little bit deeper on what you're seeing as far as the inventory levels your customers are holding? And do you expect any replenishment when you gave the unit expectations for this year? Or is it more lean inventories continue? I'm just curious how you kind of bake that into your guidance.

Yes. So I think for what we're seeing, and I mentioned it also in the remarks, we are seeing increased demand in terms of order flow. So 2025 ended higher than expectation. 2026 is constantly increasing in terms of production. But what happened specifically in Q4 was also that the orders were relatively low in the first place as December is a slow month in ordering, it's a short month with holidays, et cetera. So it was low and then production level came up even higher. So it basically means that we believe the inventory levels at our customers are not reaching their inventory target at the end of 2025. So they are tighter than usual. And it has some impact on Q1. But again, we're also seeing very good demand. 2026 production levels are going up. But yes, Q1 does have some impact of Q4 low volume combined with heavier or bigger demand.

The next question is coming from Edison Yu of Deutsche Bank.

I want to follow up on Menti. Can you give us a sense of what are the next steps with some of these customers you're talking with? I know you mentioned proof of concept. Are you going to basically ship maybe a few units? And then if that turns out well, you'll ship 30, 40 and then much more? How do we think about that -- those kind of next steps to commercialization?

I believe that 2026 is going to be maybe high tens of units in terms of POC. 2027 should be more and 2028 should be even further. 2027, we go also into production with the production partner. So 2026 is tens of units. What we would like also is to have in addition to the POCs also to produce more units for the sake of Mobileye to start experimenting with robots, not only main themselves. But again, it's going to be a high double-digit number of robots in 2023.

And maybe just to add, from the viewpoint of the customers in these pilots, the purpose is to start with a smaller amount of robots that perform specific tasks that they're kind of outlining. And you just go to the logistics center, for example, and there is a few shelves with boxes and human beings today are moving boxes according to their instructions and so forth. And basically want to see how robots can perform over a certain period of time, what's the precision, reliability, durability, maintenance and so on and gradually afterwards, expand this to more and more tasks and in larger volumes.

We are talking about companies that employ tens of thousands of employees today in these types of positions. So I think it's, again, going to be a question of supply. And that's why it's so important to have a manufacturing partner, as Amnon said, that already in '27 is able to produce robots in a serious production manner, which is important both for costs and also for scale, for volume.

Understood. Appreciate the color. A follow-up on robotaxi. Obviously, there's a lot of excitement coming out of CES. Has your view on owning more of the, should we say, ecosystem changed at all? And I ask this in the context of you obviously have a lot of parties involved. could that kind of hinder the speed of deployment or some of the logistical aspects? And obviously, it would require capital, but I think that's not that big of an issue anymore.

I think the current arrangement we have with Volkswagen and MOIA is really optimized for speed of -- for volume of deployment. So going right now more vertically integrated is not going to increase the volume of deployment. This is something perhaps to be considered towards the end of the decade or further than that. I think what we have in place is really optimal to where Mobileye is at. Mobileye will be producing the self-driving system as a Tier 1, taking responsibility not only for the electronics, but also for the sensors, of course, the software stack and all the validation and the revenue per vehicle plus the recurring revenue per mile is very, very attractive. The focus is execution.

The next question is coming from Tom Narayan of RBC Capital Markets.

The first one I have is on the '26, I guess, adjusted operating expenses. I think you guys said it's up $100 million. And I know the FX was mentioned, some other issues. But then the one that I'm wondering if that's the biggest piece of it is the Menti R&D or maybe consolidating Menti if it's operating at a loss? Just curious how we should think about that OpEx going higher. What's really the biggest driver of that? And I've got a follow-up.

Yes. Okay. So I think I've mentioned that we have incorporated Menti R&D into our guidance. So the guidance includes in terms of operating expenses, mainly 5% of regular inflation enhancement or comp and then and the additional portion is Menti R&D. So these are the significant two items. I also mentioned we have also a headwind from the FX rate, but that is mostly offset by the efficiencies initiatives we did in Q4. So I hope that answers the question.

Yes. Just to follow up. I mean I think that's pretty clear. But just to follow up, like we do expect to have normal OpEx inflation per year of around 5%. This relates to salary and benefit inflation as well as kind of additional infrastructure to support the AV activities and the advanced product activities. That's normal. On top of that this year, we are assuming consolidation of the Menti R&D expenses. We talked about that as somewhere in the lower single digits, but probably think towards 4% type of thing. And additionally, we do have this FX headwind, which is mostly offset by the workforce initiative we did in the fourth quarter, but not completely. So that should give you a decent walk from 2025 to 2026.

Yes. I think you described it quite accurately, right? But take into account that also growing in terms of being a Tier 1 in a Tier 1 position with our programs with Porsche and Audi and DRIVE. And sometimes you need to make adjustments in terms of increasing headcount. Again, this is nonmaterial compared to the overall OpEx of -- but it adds a few percentage. So take the walk-through that you mentioned, what was quite accurate and accurate few percentage of growth that we need to account for when we are taking a Tier 1 position and investing heavily into the future. So 2 years ago, we calculated our OpEx growth, but we cannot be precise single percent in an area which is experiencing the rapid growth.

Got it. And for a follow-up on Menti. I know this is very early to ask this question. But I mean, look, we're seeing the market's reaction to the potential news out of -- from Hyundai with Boston Dynamics and the credit that Hyundai is getting. Is this something you guys might think about in maybe the distant future, I don't know about trying to crystallize the value. Right now, there's so much appetite for the capital market certainly. Is this something you could consider monetizing Menti some way? Or do you believe that together as a combined entity that that's how you kind of view the business?

I think that the market is taking some time to internalize the news of the acquisition or the news of Mobileye entering into humanoid. I do believe that in some near future, this would create a dividend, the like of what's happened between Hyundai and Boston Dynamics. Menti has all the potential to make big steps forward, has demonstrated quite a mature technology as the clips that I have shown and the clips that they have on their website. And together with Mobileye, they can make rapid steps forward. Now whether we are going to see this dividend in a month or whether we're going to see this in a year, I don't know, but it has the potential to catalyze the same benefits that Hyundai is receiving from Boston Dynamics.

The next question is coming from Colin Rusch of Oppenheimer.

Can you talk a little bit about the near-term pricing dynamics on EyeQ? Just curious how much movement there really is as you see some of these larger volumes move through in the first part of the year and how we should think about that trending through the balance?

So maybe if you refer to the EyeQ prices, but to make sure I volume or prices. I didn't get the question.

I'm concerned about pricing as you ship a higher volume here and then how that -- the pricing trends through the balance of the year as you normalize out.

Yes. So the pricing is -- every year is affected by the mix of EyeQ. And as you know, the EyeQ have different generations with different software features and software packages. And this has somewhat of a different price. But overall, on average, there is no, let's say, meaningful change in the prices. There is a different mix this year compared to last year, as Moran said in her remarks, where we see higher volumes of EyeQ5-based ADAS products and EyeQ5 has somewhat of a higher cost, but still it's advanced products with meaningful volumes this year compared to last year. So this does have some impact, but it's all natural mix.

And also the second chip that I mentioned, the second chip that I mentioned combined -- the fact that the second chip is at a lower price than the first one. So it's a higher gross profit per vehicle, but lower ASP. I think that the combined natural mix that Nimrod mentioned and the second chip is approximately like $0.80 or so year-on-year.

Yes. But just to clarify, the second chip or a car has two EyeQ4 chips, this is a one-off thing. It's not that we see a trend having two EyeQ chips in the car with one of the EyeQ chips at a discounted price. This is what we call a bridge. This is a bridge towards EyeQ6 light. The carmaker wanted to meet certain regulatory environments that EyeQ4 alone could not meet. Therefore, a second EyeQ4 was added. But again, this is a one-off. We don't expect it to be a trend.

Okay. Perfect. And as you think about doing the Driver demonstration here later this year, can you talk about the regulatory process and any bottlenecks or hurdles that are still remaining here, things that are of concern that you guys are focused on getting ready to do that demonstration?

Well, in the U.S., it's self-certification. We have stringent KPIs in terms of lead time between [ sads ] that we are meeting towards going driverless. Outside of the U.S., there is [ homologation ]. And as we mentioned together with Volkswagen homologation will occur in 2027 outside of the U.S.

But nothing on a regional basis or a city basis that you guys are concerned about?

No, we don't see. And actually, the [ homologation ] in Europe will have a stronger tailwind given that the vehicles are produced by Volkswagen, Level 4 vehicles and our cooperation together with AT&T and MOIA and Volkswagen will allow us to go through the [ homologation ] in a much easier way than if we were doing it alone.

And this is a significant entry barrier to the European market as it involves a lot of activities and the direct engagements with the regulatory bodies that we are already doing with Volkswagen. So getting this approval in '27, as Amnon mentioned, will also separate us in the European market from others.

Yes, it's an important point. And just to clarify on the timing of [ homologation ], we're saying that it will be completed in 2027 and start in 2026. And the six cities commercialized in 2027 that Volkswagen talked about includes some European cities, which is going to require the [ homologation ] process to be completed.

Our next question is coming from Joshua Buchalter of TD Cowen.

I guess to start, you highlighted the potential for conversions on surround ADAS this year, but you haven't made the same comments about SuperVision and Chauffeur. Are -- maybe you can provide an update there. Are you guys deemphasizing that in your go-to-market and conversations with customers? And then for my follow-up, that's on a completely unrelated topic.

Amnon , you've touched on this in the CES presentation, but I was hoping you could provide some more details about specifically how your EyeQ road map is going to accelerate Menti's time to market? And perhaps as important, how much software development is needed to move further into robotics, given EyeQ is designed specifically for autos?

So we have multiple engagements also on SuperVision Chauffeur. So there's definitely an active engagement there with the market. Just to put things in perspective, our relationship with Volkswagen Group with the different brands on these products started maybe in 2021.

And it took us a couple of years to kind of cross all the items that is needed. And we are also focused now on opportunities that have a meaningful business potential as opposed to smaller scientific projects that some OEMs are trying to explore, maybe in some cases, it's in-house development that they're doing and they want to allocate one car in the future and see if it works.

We're trying to focus on opportunities that present significant volumes, multiple vehicle models with concrete time lines so that we can scale the products. We are not looking for the first opportunity. We want to kind of scale and we have several of those. I don't want to predict timing, but we are encouraged by the activity there.

Regarding the Menti, we feel that it's too early to talk about EyeQ chips on humanoid robots. We think this is a longer term -- this is a longer horizon issue. Currently, the robots are based on NVIDIA's chips, and we see that we are very proud of that relationship, and we see that going on for the near -- for the foreseeable future. When we go into really high-volume production where every cent counts, then I think EyeQ8, EyeQ9 could be quite relevant, but it's not in the foreseeable future.

We're showing time for one final questioner. Our last question is coming from Samik Chatterjee of JPMorgan.

This is [indiscernible] on for Samik Chatterjee. My first one would be like since you said that Porsche and Audi programs are now pushed out to 1Q '27, will DRIVE or Robotaxi be the biggest swing factor for 2026 revenues? And on that itself, like any updated thoughts on the monetization for DRIVE in terms of upfront revenues versus recurring consumption-based revenues?

And sorry, for my follow-up, I wanted to ask on the second surround ADAS customer. You said that you -- there could be a potential decision for the second architecture with this customer in EyeQ. If that happens, like will that potentially double your pipeline with that customer? That's it.

The answer to your first question is that there -- we did not expect any meaningful impact from the advanced products in 2026. We've been saying that for the last several quarters. So there's no change related to what you talked about.

And we did not account for DRIVE revenue in 2026 guidance. So there is no -- it's not in the guidance. Regarding the second question on the second design win for ADAS. So the discussions are obviously ongoing, and we are making good progress. And again, I don't want to go into predicting time, but we continue to work on this and it's progressing.

Thank you. At this time, I would like to turn the floor back over to Mr. Galves for closing comments.

Thanks, everyone, for tuning into our earnings call, and we'll talk to you next quarter. Thank you very much.

Ladies and gentlemen, thank you for your participation. This concludes today's event. You may disconnect your lines or log off the webcast at this time, and enjoy the rest of your day.

Thank you and welcome, everyone, to the annual keynote address of Mobileye's CEO, Prof. Amnon Shashua.

I'm Dan Galves, Chief Communications Officer. We have a really interesting agenda for you today. Of course, it's exciting to share the news, about 30 minutes ago that Mobileye has agreed to acquire Mentee Robotics. A portion of Amnon's speech will cover that. We're also really excited to provide updates on the automotive side as we always do at CES. This includes our collaborative work with VW's MOIA Group to bring autonomous vehicles to the roads of U.S. and Europe. On that note, we're very gratified to have Christian Zenger, CEO of Volkswagen Autonomous Mobility to join us for a conversation with Amnon. Thanks for being here, Christian.

Before Amnon begins, I'll read a forward-looking statement and then I'll play a short video. Please note that today's discussion contains forward-looking statements based on the business environment as we currently see it. Such statements involve risks and uncertainties. Please refer to the accompanying presentation and Mobileye's periodic reports and other filings with the U.S. Securities and Exchange Commission, in particular, the section therein entitled Risk Factors, which include additional information on the specific risk factors that could cause actual results to differ materially. Thanks, and we'll start with a short video.

[Presentation]

I'd like to welcome Amnon to the stage. Thank you very much.

Hello, everyone and good afternoon. I'll address the scope of robotics, where Mobileye is entering into the full aspects of physical AI with the acquisition of a humanoid robotics company. But let's start first with Mobileye, our main business. And then towards the end of my talk, I'll cover aspects of humanoid robotics and I'll show you some clips never shown before by any company in humanoid robotics.

So let's begin. Normally first, how we did back in 2025. So when we look at our RFQs with our top 10 customers, we won 95% of all RFQs during 2025. We had 2 new design wins with OEMs that we did not have a relationship with, Volvo and Subaru, in terms of our EyeQ6 Lite. EyeQ6 Lite is the chip for the basic ADAS, the high-volume part of our business. During 2025, our pipeline of EyeQ6 Lite was 3x -- 3.5x compared to 2024. So very, very strong momentum. In terms of our pipeline, in the next 8 years, $24.5 billion, just to remind everyone that we are a bit short of $2 billion revenue 2025. So it means if you take this amount and you divide it by 8, it's already 50% more than our revenue in 2025. So this is a very strong pipeline. This is a pipeline based on signed contracts with OEMs. $18 billion of it was awarded since we started our IPO, 40% increase of pipeline since 2023.

And then we are competing on RFQs with potential of 100 million more chips. So that will even increase our pipeline. And if we looked at our deployment since inception of the company, more than 230 million chips, 230 million cars on the road. I think it's 1/8 of the total cars on the planet. So this is a very, very big number. In terms of our REM, our technology for building high-definition maps, we have 8 million cars sending us -- we call this harvesting, sending us data every day. 32 billion miles were recorded, harvested during 2025 and 3 new OEMs have joined this harvesting fleet. So this allows us to build high-definition maps, which we call REM maps, high-definition maps with coverage that is expanding Europe, U.S., Asia and so forth.

We just announced yesterday a U.S. OEM with Surround ADAS, Surround ADAS is our system with EyeQ6 High, our latest high compute chip, 5 cameras and 5 radars, 5 or 6 cameras, so front-facing camera, 4 parking cameras plus 1 to 5 radars, We believe that this is the evolution of ADAS. ADAS eventually will move from front-facing to what we call the Surround ADAS and 9 million vehicles shows that this is really high volume. And SOP is middle of 2028. So when we look at Surround ADAS, we have already 2 design wins, Volkswagen, which we announced several months ago and the major U.S. OEM. Together, it's 19 million units. And it shows that this is on its way to become the next step of driving assist, the next evolution of ADAS with ASPs when we are a Tier 2, somewhere between 2 to 4x our normal ASP. And when we are a Tier 1, somewhere between 12 to 14x.

And just the status of our advanced product, Surround ADAS, I just mentioned. with Volkswagen, it's already in B-sample stage. This is together with Valeo. Valeo are the Tier 1, we are the Tier 2. And the full software would be installed end of this year for an SOP of 2028. SuperVision with Porsche and Audi, the Volkswagen Group, premium brands of the Volkswagen Group. We're already in C-sample hardware. We have dozens of vehicles in test phase. And our Gen 2 stack, which I'll talk about this later, would be ready in April this year and SOP is 2027. Chauffeur, our Level 3 with Audi. This is a B-sample stage. We have prototype vehicles in testing and we are working on specific functions for eyes-off. Again, this is an SOP 1 year, 1.5 years later.

In terms of our -- if we look at our robotaxi activity, we have multiple sites, both with our partner, MOIA and the Volkswagen autonomous vehicles with the ID. Buzz, multiple locations in Europe, countries like Oslo, which have lots and lots of snow. So we're not only testing in sunny conditions. We're testing in all terrain -- all possible conditions. This is with ID. Buzz, with the Holon vehicle. So see that we are covering multiple sites in multiple cities, multiple geographies.

So let's start with our robotaxi. This is a project -- the lead project is with Volkswagen and MOIA. And just to recap where we're standing in terms of the technology, the sensor configurations, 13 cameras. There are 3 long-range LiDARs by Innoviz, 6 flash LiDARs also by Innoviz and 5 imaging radars. These are manufactured by Mobileye, very special radars. We talked about that last year. And our -- there are about 100 vehicles today test driving all over the sites that I mentioned before. The coming milestone, the Level 4 ready vehicle would be ready in February this year, so next month. And we're going driverless Q3, Q4 in the U.S. And then start expanding in 2027 to multiple cities with also partners like Ruter, Holo, Uber. It's all under the roof of Volkswagen and MOIA. So here is just to show kind of the breadth of testing with the ID. Buzz, multiple cities, multiple geographies, also snow. One of them is called U.S. Launch City. We're still keeping the identity of the city in wraps but the idea is to start launching there in Q3.

So I'm happy to invite Christian Zenger. He's the Autonomous Mobility -- Volkswagen Autonomous Mobility CEO and also the Chairman of MOIA. And we have here a clip that will run in the background while we speak.

Please, Christian.

So Christian, thank you for joining us on -- joining me on stage. We have been working for the last 2 years on the ID. Buzz. I think it's a great partnership. Give me your perspective, why you think we fit together? What's going well in this kind of partnership?

Well, thank you, Amnon. I think the short answer is that we have an absolute clarity of roles and a shared vision. From day 1 on, we were aligned on a simple idea. Autonomous mobility will only scale cost effectively if each partner focuses on what each partner is best on. And so Volkswagen brings industrial scale vehicles and homologation. Mobileye brings the state-of-the-art Level 4 system, which we integrate very well by MOIA, a new entity of the Volkswagen Group into a turnkey solution that makes autonomous mobility as a service product. And we are all doing what we do best. And this is why our partnership works really well.

Very good. So let's highlight what are the next milestones. We have a very exciting year ahead. Our test fleet has grown a lot. We have now more than 100 ID. Buzz vehicles in multiple sites around the world. So I think we're in a very good position. Let's talk about the new -- the next milestone.

Exactly. Today, we have more than 100 ID. Buzz AD vehicles running across Europe and the U.S., more to come. We are in Munich, Hamburg. We are in Oslo but also in Austin and Los Angeles. More locations are coming soon. The scale of real world test across very different traffic, weather conditions and also legal systems is rare in the industry. And the ID. Buzz AD has a perfect platform, leveraging technology from across all the Volkswagen Group.

Yes. I think also there is a great synergy in what we do together. I think it's a 2-way synergy. On one hand, we benefited from the economies of scale working with SuperVision and Chauffeur with Porsche and Audi and other premium brands of Volkswagen. So there's common hardware. For example, the ID. Buzz has 2 ECU boards, which are duplication of the SV62, which we have with the Porsche SuperVision. So there's economy of scales on this hand. And at the same times, we benefit from all the data that we collect through the ID. Buzz. It's priceless data where we have 360 collection, 360, not only video but LiDARs and radars and we have technologies that create ground truth, automatic ground truth from this data. All of that allows us to bolster and improve all our sensing technologies, which is relevant also to our SuperVision and Chauffeur. So this 2-way synergy is very, very strong.

It's really a good point. I remember in prior years, everyone talked about creating a scalable self-driving system from Level 2 to Level 4. Now the Volkswagen Group, MOIA and Mobileye is actually doing it. And the development process we have seen in the year 2025 is even strengthening our confidence in this path we are on. And I really spend a lot of time in the vehicle, seeing the performance and seeing the focus we have on continuous improvement, what makes the product over time. And we have already revealed the production version of the ID. Buzz in full self-driving version last year, it's the first fully self -- purpose-built vehicle.

We see it on the charts behind us. But forward-looking by the end -- by the third quarter of this year, we expect to go live with the entire MOIA ecosystem, including the vehicle, the Mobileye self-driving system, the software platform for fleet control, remote guidance and all the secondary driver tasks. And even more, this will be followed by the launch of driverless services by end of 2026 in the U.S. And followed by EU homologation in 2027, we will attempt to rapid market expansion thereafter.

Yes, very, very good. We're ready to support all the scale. I wonder what's your view of where the market is developing, where the market is heading?

The key lesson is that autonomous driving has shifted from technology challenge to scaling and business model challenge. So scaling remains difficult. The breakthrough comes from specialization and a strong ecosystem model. And MOIA sits precisely on this intersection. We have 2 task to do. First, we integrate vehicle and self-driving system, combine this into our own software and service tools and making this into a real true turnkey solution for our customers. So in short, we integrate what the supply side delivers with what the demand side needs into a turnkey solution ready for autonomous mobility. This is our ecosystem logic and our starting lineup.

First, the ID. Buzz AD, a purpose-built safe vehicle with self-driving system, Drive 64 installed, already in the production line, enabled by Volkswagen industrial scale and backed by logistics and aftersales capabilities. Second, the digital driver by Mobileye, a Level 4 self-driving system built on more of 20 years of ADAS experience, a global data foundation, what you explained very well and an industrial experience. Third, the ecosystem platform by MOIA and this is new. A software backbone that unites the ecosystem, including passenger management, fleet control, remote guidance, safety oversight and real-time monitoring and it enables to be connected to several booking platforms. And fourth, what truly differentiates us in our ability to combine the best of both worlds, Volkswagen's industrial scale and high-volume manufacturing and MOIA's strong technology expertise with partnerships and a deep experience in urban mobility.

The platform supports multiple use cases from robotaxi to ride pooling to shuttles and line services for demand generators and operators around the world. And because it's built as an ecosystem, it scales faster, significantly reduces cost and reaches breakeven much earlier. And this leads us to some very clear targets, 6 cities by end of '27 and more than 100,000 active self-driving vehicles on the road by end of 2033.

Amazing. This is truly, truly amazing. So thank you again for joining us and for the strong partnership, where we've been working in a great partnership for the last 2 years and I'm looking forward for a very bright future. And really, as you said, the name of the game is scale, right? Technology, it's quite clear that the technology can work. Now it's a matter of how to scale. And with MOIA, Volkswagen, I believe we can do that. Thank you very much, Christian.

Thanks a lot, Amnon.

Okay. Let's continue. And I would like now to take the opportunity to go a bit under the hood. I like every time to pick your brains about technology. So let's go under the hood. I'll go under the hood in our robotaxi technology stack but there's lots of shared components with all our stacks, SuperVision, Chauffeur and even going down to Surround ADAS and front-facing ADAS. So there are basically 3 things that one needs to consider when building such a stack. One is, how do we harness -- the best way to harness the modern AI, generative AI, whether it's vision language models, vision language action models, fast and slow. I'll talk about this later. There are all sorts of caricature approaches, which people mentioned but they don't do because it's very easy on the ears of investors to hear the simplified version.

I'll go into the nuances of what really is being done. Validation methodology. So in a robotaxi, you collected sufficient data to convince yourself that you can go driverless in a city. Now you want to expand to another city. How much data do you need to collect, right? This is part of validation. Economy of scale and how -- what would the sensor setup look like 4 years from now. We have certain ideas of how to reduce cost by reducing the sensor suite. I'll mention that later. And then more importantly, in a robotaxi setting, there is this teleoperations, which everybody knows that exists but nobody talks about, right? So if you have 1 teleoperator per 1 vehicle, you don't have a business. You haven't done anything, right? So you need to aspire to have 1 teleoperator over many, many cars and eventually asymptotically to have no teleoperations, right? That's how you build the business.

So Christian mentioned 100,000 vehicles. So we don't want 100,000 teleoperators in the back office or even 50,000 or even 10,000, right? So how do you handle this asymptotically? You can start with 1 teleoperator per 1 vehicle at the beginning. But later, you need to convince yourself that you have a line of sight on how you kind of reduce this considerably. Then when we look at the ingredients of autonomy. The first 2 left columns have to do with how do we take vision language models. So just to remind, vision language model is a transformer where the input is both images and text and the output is text, right? So how do we harness VLMs in a way that makes sense? So kind of the caricature approach. I'm not saying naive approach because nobody does it. So it's not naive. It's a caricature so what is a caricature? Is you have pixels coming in, you have a network in the middle, say, transformer network, a VLM and trajectory, the commands coming out.

Now nobody does this because there are so many reasons why this is wrong. First of all, these networks hallucinate. We know that from ChatGPT, Gemini language models, they hallucinate. So it means in our world, how do you give safety guarantees when you have something that can hallucinate? Second, there are issues of sample complexity. Sample complexity is the amount of data that you need in order to generalize, right? So the sample complexity of perception is much, much smaller than the sample complexity of planning because planning is a multi-agent, so there's a compounding effect in sample complexity. So putting them together doesn't make sense from a sample complexity.

So when you look at research, I'm talking about recent research, recent academic papers, also blogs of actors in the space, they have heads that create sensing state. Sensing state is the recording of all objects of all relevant information around the car because that will reduce sample complexity. So all of a sudden, now you need to label data. How do you -- so it's not just input in and commands out and you have a driver that gives you kind of the error signal, it is you have to now label all the objects around you. So this adds another nuance, another complication.

So the caricature approach is just a caricature. In reality, it's -- there are nuances, which I [ want to ] get in. What people have noticed and it started from robotics actually, is that when the problem is, you want to decipher a scene, which is a very complicated scene, a VLM could be a very interesting tool because it has been trained on all the data of the Internet. Imagine trained on all YouTube clips. So it has a very strong sense of scene understanding. But the demand for scene understanding is sparse. You don't need to do that at 10 hertz. You don't need to do it at 10 frames per second. So what has emerged is a concept called fast and slow. So it has emerged both in academia, academic papers like from Li Auto, a system called DRIVE VLM. It has emerged from Waymo's blog. They also mentioned fast and slow. And in robotics, like Figure AI Helix system.

So what does that means? You have the fast route, which is 10 frames per second that is responsible for all the safety layers. And then you have the slow system, which is 1 frame per second, 2 frame per second, which does the deep scene understanding when it is needed. So now how do you put them together, this fast and slow? The third column is policy. Policy is the planning, is the decision-making. The sample complexity of policy is very, very high, again, because it's a multi-agent compounding effect, right? The host vehicle performs an action in the world and this action affects other road users. Therefore, there's this compounding effect. So when you have something of a very high complexity, you need a lot, a lot of data. Now the data coming from real world is limited. Even if you have millions of cars sending data, it's still limited. You can do -- you can run on a simulator, say, photorealistic simulators and do training over a simulator but then you are compute bound because then compute becomes a bottleneck.

Running over a photorealistic simulator requires a lot, a lot of compute. So say the target is to train over 1 billion hours of driving. So this is not realistic for real data and not realistic for photorealistic simulator because of compute constraints. So how to do that? And there are interesting innovations there. The last one is end-to-end. End-to-end is important because what end-to-end does, you back propagate from the commands back to the input of the system and you optimize what really matters because you could have perception mistakes that don't matter or you have perception mistakes that could kind of accumulate. And when you do this back propagation, you ameliorate this accumulation. So end-to-end is important in that aspect. That's why you build a system, you want to do this last fine-tuning of an end-to-end in order to optimize what really matters.

Normally, this requires that all your components are differentiable. This is how you do by propagation but this is not necessarily the case and I'll mention that. So those are the ingredients. Now let's put these ingredients together into an architecture. And the pieces in blue are the pieces that I want to expand. So let's start from the left. The left, you have sensors, cameras, LiDARs, radars, all the modalities. And you have also high-definition map. Again, I'm talking about the robotaxi world, right? There's no robotaxi that drives without a high-definition map.

So you have maps and you have -- could be -- have other data, telemetry data. All of that goes into an end-to-end perception network. Even this is not true, right? I put here end-to-end perception network. But in reality, in machine learning, there's a concept called shortcut learning. What this means? If you have 2 sources, one with a high sample complexity and another one with a low sample complexity. For example, LiDARs have a low sample complexity because they record 3D data. Cameras, they don't record 3D data. You have to infer 3D from 2D images. So sample complexity of cameras is much, much higher sample complexity of LiDARs.

What happens if you do this low-level fusion, you feed them into 1 network, the network will tend to do a shortcut and rely only on the LiDARs. So you have to be a bit more sophisticated there. Just assume you have a network for the cameras, the networks for the LiDARs and then you have a network that does this fusion. But let's not go into that resolution. Let's call this an end-to-end perception network. This network now outputs a sensing state. Sensing state is the recording of all the relevant information around the car, where the vehicles are, the type of vehicles, the lanes, the traffic lights, pedestrian crossing, whatever, all the relevant information, that's the sensing state. And as I mentioned before, you have to output sensing state, even those that talk about only pure end-to-end, they also output. They have a special head to output sensing state because you need to reduce the sample complexity. This is backed by academic papers. It's backed by blogs of actors in this space and so forth. So you output a sensing state.

Now this sensing state goes into this blue box ACI, which I'll mention in a moment, that is responsible for the driving policy, for the planning. So it receives that input, the sensing state, it receives as input also the slow route, which is the VLM. We call it VLSA is for semantic. So it's a Vision Language Semantic Action model, which I'll expand in a moment. And it makes decisions. It goes into a safety layer where the safety layer receives an input, the commands and also the sensing state and it employs RSS and PGF, stuff that I talked about last year and outputs the commands.

So now the innovation -- there are 2 innovative blocks here, which are in blue, which I now want to expand. First of all, is the ACI. ACI is something unique to Mobileye, although there are academic papers that talk about this as well. So let's start with ACI. So what we want is, we want to train over simulated data. Now it's not photorealistic simulation. It is the sensing state simulation. That means you have a map and on the map, you can place agents, which are cars, buses, pedestrian and so forth and you start to simulate. You can generate as much data as you want. And you can also generate interesting data because when you talk about real-world data, most of the data is boring, right? You want to inject edge cases at a much higher density.

So a simulated environment will allow you to do that. And because the input is sensing state, it's not photorealistic. So we can first generate much, much bigger volumes of data. Second, we're not compute bound because it's not a photorealistic simulator. Now what is the inspiration? Inspiration is the AlphaGo and AlphaGo Zero, a concept called self-play. So just briefly to mention, 2015, DeepMind introduced a reinforcement learning system to play Go. Go is a game much more difficult than chess. And what they did, they imitated humans gameplay. So they had millions of games played by humans and they created an imitation of the humans and they achieved impressive results. It was really amazing.

A year later, they introduced something which is really mind boggling. There was no human data at all. What the system has done, it's -- it played against itself called self-play. It could then train on much more data because it's not limited by the amount of data you have from humans and it was much better than AlphaGo, and they called it AlphaGo Zero. So this concept of self-play is the inspiration of what I want to show here. So we call this ACI, Artificial Community Intelligence. It is -- idea is to use self-play in order to train planning. So what do we have here? We have -- we take a map and now we can leverage our REM maps. We, basically, we have the entire world map mapped. We have all of U.S., all of Europe, right? So maps, we have plenty.

You take the maps and now you place agents on the map. Now when you are training driving policy, one of the things -- one of the challenges is that you are training your driving policy. You want to be a safe driver but you need to make assumptions on the driving policy of the other agents, which are not necessarily safe drivers. So you cannot assume that other agents are using the same driving policy as you are using because you are a safe driver, right? So now there's the question, so what's the driving policy of all other agents? In this type of simulation, you are creating all possible driving policies. That means, first of all, you have a kinematic profile. So kinematic profile of a pedestrian, of a truck, of a car is different.

Then you have reward weightings. You can have rewards for reckless driving, for fast driving, for slow driving, for someone who's violating traffic rules. So you can create a lot of behaviors. You can create hundreds and thousands of different behaviors. And then you have augmentations. For every agent, it could be -- you are doing abrupt stops, you can violate the traffic laws or not violate traffic laws. So in this way, you create a superset of all possible behaviors. And now you are doing reinforcement learning where the goal is stay -- reach your destination and don't collide. No collisions, okay? And what is our unique aspect in this because we didn't invent self-play. We didn't even invent this concept of training on a simulator.

There are 2 aspects. First, we're leveraging our REM maps. So we're not just taking some city which we have data of and training on it. We can train on data on the entire world in terms of maps. Second, the big issue, you trained on a simulator but you want to move it to the real world. It's called sim-to-real. So you need to understand the noise model of your perception engine because your perception engine is not perfect. How do you kind of capture the noise model of your perception? And we developed very, very sophisticated techniques on this sim-to-real transfer in order to transfer the policy learned on the simulator and bring it to the real world. So let me now just show you what this means. So we have here agents, this -- the 12 agents you see here in the cars. The circles are destinations. So when an agent reaches the destination, the agent disappears from this simulation. And then you have crosswalks and stop signs here.

So now if we look after tens of hours of training, so there are 12 agents, none of them reached the targets and there were 6 collisions. So you look at this, it looks terrible. They didn't even stay within their lanes, right? So nothing is happening after tens of hours of training. If you look after 140,000 hours of training, all agents reached their destination but there were 2 collisions. Let's have a look at this. You're seeing here, that was a collision. If we train after 2.8 million hours of training, all agents reached their targets and 0 collisions, okay? So we are not training millions of hours. We're training billions of hours and we have a cluster where all this billions of hours are trained overnight. So overnight, you can train with the amount of data that no real-world data can match to. And this is very important. So we do that also for training the driving policy and also for validating the driving policy, which I'll mention in a moment, okay?

So what we -- the ACI provides first policy training. We can train with much, much more data that you can imagine from real-world data or from photorealistic simulation. And also, we can validate. So imagine robotaxi, where now we want to expand to a new city. We have our HD map of the new city because we have HD maps all over the world. We take that HD map of new city and then we train 1 billion hours to validate that there's nothing in this new city that is unfamiliar in terms of the map. For example, there could be a very special 4-way stop with multiple lanes that you haven't encountered or you haven't trained when you did your 1 billion hours because it also depends on the type of map. So in this way, you can validate overnight that there is nothing in the map that creates an unfamiliar situation for your driving policy. So this also provides policy validation.

So this was 1 blue box. The second blue box is how the vision language model, the slow path. So this is the kind of scenes. When you look at this, you ask yourself, what is happening here and what should I do? So normally, in a robotaxi setting, what you do in this kind of situation, you ask your teleoperator. This is why you have teleoperators. And the teleoperator will tell you, you have to -- on the left hand, you have to yield to the policeman, ignore the traffic light and yield to the policeman. On the right-hand side, it tells you your lane is blocked and you have to turn either right or left depending on your destination, right? Now the demand for this is not 10 frames per second. This is why it's a slow thinking. So this is the slow versus fast, the slow.

The fast is the safety layer, the 10 hertz, 10 frames per second decision. It goes through safety layers like RSS and so forth. And the slow is to get very deep understanding of the scene when the scene is very, very complicated. okay? And the vision language models are a very good candidate to help you understand complex scenes because they are trained all over the Internet, all YouTube clips, you can imagine. So they have very strong sense of understanding of the scene. And in the context of robotaxi, this provides you a way to think asymptotically how to remove teleoperators completely. So what we have -- we built a system in which the output of the VLM is not a trajectory. So the fast system output is a trajectory, what are the commands. The slow output is a script. Think of it as an accompanying adult accompanying a young driver. So the accompanying adult doesn't have control of the steering wheel, doesn't have control of throttle and brakes. It talks. It talks to the young driver and gives them some guidance. So this is what this slow system is doing. It talks.

And talking is very natural to language models. It's much more natural than outputting a trajectory. So you see on the right-hand side, it's like filling a table. So action type, static interaction, target box, you see this red, the target box. The passibility, you cannot pass, so it's no. Then there's action type navigation, right? You need to do something. What is the direction? Turn right. Command? Turn, right? So it's kind of filling a table, which is very natural for a language model to do. Or in this case, the action type, temporary ignore traffic light. All right? The target box, you have your target box of the traffic light and of the police. And also action tied dynamic interaction. What is the offset? You don't need to do any offset but the relation, yield, you need to yield to the policeman.

So this is what an accompanying adult does, and this is what this VLSA does. It outputs semantic information that goes into the ACI, into the planning, which receives both the sensing state and the information from the slow system. Okay? So just to recap, this design, we have the ACI, which we can do massive scale training, way more than you can do with real data or photorealistic simulator. And we have the VLSA, the Vision Language Semantic Action model that can act as an accompanying adult and gradually replace teleoperators, okay?

So last point, what about supporting end-to-end? So as I mentioned before, the end-to-end is important as the last stage of training where you want to propagate errors and optimize what really matters, which is the end result, which are the commands. So normally, when people talk about end-to-end, they assume that the entire sequence is differentiable. Networks are differentiable. And here, you have nondifferentiable elements like the sensing state, like VLSA. But the solution to that is being done in language models. Language models have tools and these are nondifferential elements. So they do this end-to-end training using reinforcement learning and this is something I'm not going to get into resolution of that but this is something we also do. So we can support end-to-end even though we have nondifferential elements along the track.

So I hope it wasn't too complicated. I just wanted to give a sense of going away from the caricature because nobody really does this caricature. Even people talk about it but nobody really does that. It doesn't really make sense. And you need to design a system, how best to harness the best of modern AI. And I show an example of our design with some innovative elements like the ACI and VLSA. So now I'm going to put this together into how we see the future, how we see the next 5 years. So I put here the advanced product, the Level 2+ plus the SuperVision. This is an eyes-on -- hands-off, eyes-on system. The challenge here is just cost reduction.

The driver is responsible. You don't need to get into MTBFs that are relevant for eyes-off. I need to have a good system that's comfortable and the driver is responsible. So when the system makes a mistake, the driver can take over. And the challenge here is cost reduction. We -- our first system coming out of 2027. We have now designed a cost reduction of about 40% cost reduction for 2028 and we'll continue this path of cost reduction such that it will become more and more affordable. Next comes the Level 3 product, which we call Chauffeur. What we see there, the evolution of this line of product is to go from eyes-off to mind-off. So in an eyes-off system, the driver is now not responsible in the ODD, say the ODD is highway driving. But the system can ask for the driver to intervene. It's not instantaneous intervention but you say you have 10 seconds to intervene and if you don't intervene, the car will stop on the site.

Now nobody talks about what is the frequency of intervention. Say it's every 10 minutes, you ask the driver to intervene. It's not that comfortable. Mind-off says that I can give you a guarantee that I'll not ask you to intervene. So this is where this VLSA comes in. This is where this fast and slow system comes. So in robotaxi, you want to reduce teleoperators, in the consumer car, you want to reduce this intervention rate, the intervention frequency. And we call that mind-off. You can call it moving from Level 3 to Level 4, the same. And robotaxi, the current status is, the system is commercially deployable at small scale. This is the status of the industry today. And where we want to reach in the 2030s is scale. As we mentioned, Christian here talked about 100,000 vehicles by 8 years from now. So there are 2 things. You need to cost reduction in terms of sensor set and compute.

We believe that we developed our imaging radars by thinking long term that -- we think that the second generation of robotaxis can rely on cameras and image radars alone. Maybe you'll need a front-facing LiDAR and that's it, compared to the -- all the sensors that I mentioned in the first generation. And second, lower the teleoperators, number of vehicle ratio. Asymptotically, you don't want any teleoperators, right? So this would allow for scale. And this is why we developed this VLSA.

So now let's look how all of this goes into our technology stack. So here is the EyeQ7. It's already sampled. It's ready for production. It's called PPAP, ready for production next year, Q3 '27. I even have here -- I'll do kind of a Jensen thing and show you. And what you see here is the chip, the EyeQ7 chip, right? Again, it has 22 accelerator engines, 12 cores of CPU. It's very, very powerful. But to understand the power, rather than talking about TOPS and things like that, I want to talk about what it can do. So first, a reminder about EyeQ6 High, which is coming out with the SuperVision, with the Surround ADAS, all-in in 2027 and compare it to the Jetson AGX, the [ Orin X ], which is a competing chip. And we took here 2 types of workloads, which are very, very relevant to all what we're doing. One is convolutional net. This is the ResNet-50. And second, the vision transformers. So it's a transformer that accepts images as input and creates a representation by going to the transformer.

And there are latencies. We want to measure latency. So on the right-hand side is the latency that NVIDIA reports on these 2 benchmarks. So it's not our assumption. It's not our measurement of NVIDIA's chips, it's theirs. So you see here that 0.64 millisecond compared to our 0.5 millisecond on the convolutional net. On the vision transformer, the gap is even bigger. We are looking at a 9 million parameter vision transformer, 1.5 millisecond compared to 0.5 milliseconds, right? So that means that our chip is really purpose-built for these types of workloads, convolutional nets, vision transformers. So now what would these chips, both the EyeQ6 and EyeQ7 do with the VLSA, the VLMs? So with EyeQ6 High, we can run a 3.8 billion VLM at 2.5 hertz. Again, we're talking about the slow system. So 2 frames per second or so. 3.8 billion. So 3.8 billion parameters is quite a decent size language model. It can do quite a lot. EyeQ7, we can run a 15.6 billion parameter.

Now our vision of these VLAs -- of these VLSA, that in the world of robotaxi, since we're talking about this slow system and since it's outputting text and not a trajectory, that some of it you can run on the cloud. So we see 3 layers. The first layer is on chip and you see what kind of networks we can put on chip. Second layer, talking about a 70 billion parameter network is running on the cloud. And the cost of that is not big. For robotaxi, few hundreds of dollars per robotaxi, that's way, way, way cheaper than having a teleoperator. So that's running 2 hertz on the cloud. And then we can have 1 trillion parameter on demand. So it's not that it's running every second. Is the system asking, just like you ask a teleoperator, I don't know what to do. Tell me what to do. So now you have a Gemini 3 telling you what to do, okay, or Gemini 3 like.

So the beauty of this slow system, it's not only that you can run it on board, you can also run it on the cloud, which gives a lot and a lot of flexibility. How does it look like in terms of the product portfolio? So again, on the left-hand side is our SuperVision to EyeQ6 High. We don't see any need to move to EyeQ7 or so. Here, the purpose is just cost reduction. And EyeQ6 High, the cost is between 1/5 to 1/10 of competing chips. So this gives us a lot of flexibility of reducing cost. Then we have the Chauffeur, the Level 3 is running on 3 EyeQ6 and 2 boards because you need redundancy, hardware redundancy for fail operation. And then the mind-off, the Level 3 is now add EyeQ7 or even EyeQ8, the next chip that we are designing to, do the VLSA, right? It's not something you want to send to the cloud because we're talking about consumer vehicles. And as you saw, EyeQ7, you can run a 15-gigabyte network.

It's very, very -- it's a very impressive size of network. And with the same sensor set. So it's just, you add another chip. And this is also convenient because we don't need to revalidate the system. It's the same EyeQ6 with the Level 3. You just add another chip that does this slow route. With the Drive, the robotaxi, the first generation is 4 EyeQ6 and one of them is doing the VLSA, As I said, the 3.8 billion parameter VLSA and the rest would be in the cloud. Second generation, towards the end of the decade, 2029, there will be an EyeQ7 or EyeQ8 that will do much more onboard, less on the cloud and also reduce the sensor set to just cameras and imaging radars and perhaps 1 front-facing LiDAR.

Okay. So now I'm going into the next aspect of physical AI. So Mobileye is an AI company working in physical AI but only in one aspect of physical AI, which is autonomous driving. So the difference between physical AI and AI that most of you -- all of you are using, the AI that you are using is in the digital space, starts in the digital space and ends in the digital space. Autonomous driving or physical AI is that the AI, the decision-making is in the real world. And when you think about what works in the real world, there are 2 things, cars and robots, right? And Jensen last year called it physical AI, which I think is a great term for it. And Mobileye wants to expand its scope to all aspects of physical AI because there are a lot of synergies in terms of the technology layers, both systems, they use fast and slow, both systems use VLMs or VLAs, both domains, they use simulation a lot, a lot of simulation and sim-to-real. So there's a lot of synergy and lots of overlap from the technology part.

And now you have a new TAM. You have another growth engine, which is very, very interesting, right? So I would say that the difference between the 2, autonomous driving and robotics is, cars operate in a structured world. Structured doesn't mean it's easy but it's structured, where as the robots operate in an unstructured world. There could be some structured special cases but in general, it's an unstructured world. Imagine a robot in a home use, right, in the house. So all houses look different, right? So it's an unstructured environment. And also the number of tasks is open-ended. Right? When you're looking at asymptotically, if you want to do whatever a human can do, the amount of tasks are open-ended, whereas the amount of tasks that the car does is really narrow. Narrow doesn't mean it's easy but it's really, really narrow.

So let's look at what Mentee Robotics has developed. First, just simple specs. This is their third-generation robot. It's a human height, 175 centimeters, 72 kilos. It can pick up 25 kilos, which is very important if you want to work in fulfillment centers, you need to move stuff which are heavy, so 25 kilos. It has a swappable batteries, so it can work 24/7, just need to swap its main battery while it's active. Also, the -- everything is designed in-house. So it's really a vertical integrated company. All the actuators are designed, the gear is designed in-house. Electronics are designed in-house. All the software is designed in-house. It's running on 2 [ Orin X ] NVIDIA chips. And it has a very interesting -- hands is -- hands are -- maintaining human-like dexterity of hands is very, very challenging in humanoid robots.

The kind of the default is to use tendons just like with humans and then use sensors on the tips, which complicates manufacturing and increases the price. What Mentee has done, they did rigid links. So you don't need any sensors on the tips because it's connected directly to the motor. So you have kind of a backward feedback. And let's look at the AI capabilities of it. So the first core principle is sim-to-real. You are training -- you are doing a reinforcement learning training on a simulator and then moving to the real world using technique called sim-to-real. And many demonstrations of humanoid robots that you see out there, it's teleoperated. What I'm going to show you here is no teleoperation. It's all full stack AI. So let's look at this clip. It's also on the website. You have here 2 robots working for 18 minutes straight, clearing up all the boxes from one side to the next. And what you see here in the text is all the internal thinking of the robot, both the computer vision, the navigation, the instruction following.

Let's give it another few seconds. So on their website, you have this full 18 minutes, you see this end-to-end. And now we're just simply fast forwarding it. So all the boxes from one side moving to the other side, okay? So this was the -- first is instruction following. So next clip, there is a full end-to-end. There's an instruction, understanding of the instruction, navigating, recognizing. So let's have this.

[Presentation]

So see on the left-hand side, it has a cognitive map, so it knows where the kitchen is. It's now also -- you'll see in a moment, it's recognizing all the objects and putting a green bounding box on the relevant object. And this was part of the instruction of bring me another coke can and he showed him the coke can. And then going back, navigating back to where they were. So it's all kind of end-to-end, from instruction, instruction following. And this was not learned or trained by imitation learning. They are kind of out-of-the-box capabilities of pick, move, drop that is common to many industrial settings that the robot comes out ready just by following instruction.

Here, now I'm showing something that is teleoperated. The point I want to make here is to show the dexterity of the hands. The hands are very, very interesting. So in terms of the accuracy, now taking a drill and using the drill, moving this from hand to hand. And these are hands that are really manufacturable at really low cost because they're not based on tendons, okay? Next, I want to show something that nobody has shown before, right? So when you talk about humanoids, I think there are 2 stages for humanoids. The first stage, let's call this low-hanging fruit. Again, it's nothing that's easy. Low-hanging fruit is when you're working in a structured environment, so you are working in an assembly plant or you are working in a fulfillment center moving boxes or moving items. So normally, a customer would not buy 1 robot, would buy scores of robots, dozens, hundreds, thousands of robots.

And the customer would be interested in a set -- predefined set of tasks, so you can customize your software to the need of the customer. But now look at the second stage. Second stage, you would like to operate the -- robot to operate in an unstructured environment like home use. So a home would buy 1 robot. It doesn't make sense to start customizing a robot for every customer, right? So now you need the robot to be able to continuously learn, to generalize. Now you want it to be able to learn from the customer. So the customer would show you a task. So you, the robot are watching the customer. The customer can talk to you but also show you the task and you would like within minutes to be able to repeat that task, right? So we call this real-to-sim-to-real. So the robot is watching, sending the clip to the cloud, in the cloud, we have a foundation model that we built that moves this video into a simulated environment.

Inside the simulation, there is an RL loop, reinforcement learning loop to learn the task. And then it moves back to the robot, this is the sim-to-real part. So real-to-sim-to-real. And let's look how this looks. So first is the observe. What the task is going to be here is to swap batteries. So the human is showing the robot what it's doing. On the right-hand side, on the small [ inbox ], it's already has been moved into the simulated environment. But in reality, the clip will be sent to the cloud. And in the cloud, it will create a simulated environment. And now so -- and now comes the training. Now the training isn't in a simulator. So this is after 100 iterations of training, after 500 iterations of training, after 1,000. So right now, this training takes about 3 hours. But by the time we go into production, this will take minutes.

Now this is the real part. So now the robot is performing the task that the human has showed him. So -- and this is why the company is called Mentee, the idea of mentoring. The idea is that human will mentor a human, which is a layman human, not a professional, just a customer, showing the robot a new task and the robot will be able to do it. So where we are in terms of the road map. In 2026, start POCs with multiple customers. We have already signed agreements with AUMOVIO, they -- it's Conti, they changed their name to AUMOVIO. They'll be our production partner. 2027, we'll start manufacturing the first batch, experimental batch and to be ready for 2028 deployment -- commercial deployment for the first stage of robots, which is for fulfillment centers, assembly plants, production plants, which are structured environments.

And then towards the end of the decade, 2030 to go into the home use. And this is where this real-to-sim-to-real becomes very, very crucial because the robot needs to continuously learn and not just come with predefined capabilities because the predefined capabilities can be useful but not sufficiently useful in order to cover all the needs of a home. So -- and as I mentioned before, there's a lot of synergies. It makes a lot of sense because there is a overlap in the kind of tools that are being used. For example, the sim-to-real, the real-to-sim-to-real is very useful. Mobileye is also doing sim-to-real, as I mentioned, with the ACI block. We train in a simulator and then we learn the noise model of our perception and move it back to the real world. So there's often a lot of know-how in this sim-to-real. This is shared among both companies. So whatever Mentee is doing can help Mobileye, whatever Mobileye is doing can help Mentee, makes a lot of sense to do this -- to join forces.

And it offers Mobileye a new growth engine, which is a very important growth engine. I'm very optimistic on humanoids. I believe that 10 years from now, there will be millions of robots. There's a labor shortage out there. In fulfillment centers, in many of them, the turnaround is 100% per year. It's very boring task. People get injured, people get bored and there is labor shortage. Also labor shortage in health and home. You can think of elderly home care. There's lots and lots of potential if you have a useful robot. And I believe that the technology, AI that is moving so rapidly forward, the technology is ready.

And just a matter of productizing it and being smart and thoughtful about how you put this technology together into a product. And this would be a -- this will be a magnificent growth engine for Mobileye. So I think this is a very, very exciting new stage for Mobileye, we call this Mobileye 3.0. So 2.0 was after the acquisition of Intel. Now we're going 3.0. As you saw, robotaxi is on its way and we have a very good vision of scaling. At the end of the day, scaling is what's going to matter, the ability to scale. We have the product portfolio of consumer cars and now a new growth engine. The AI stacks are really exciting and the entire field is exciting, where AI is moving fast forward.

So I think I'll end.

All right. Thanks for joining us, everyone. Very pleased to continue the day here with Mobileye. We have Dan Galves, the Chief Communications Officer from Mobileye, who I'm sure many of you are familiar with us. Dan, thanks again for joining us at this year's conference.

Thanks, Joe.

So a lot to talk about here. I think the Mobileye story is almost a sort of continual sort of evolution, right? You have the base ADAS products, which is sort of the core competency of the business. That was evolving more into SuperVision, which started to launch and then sort of for reasons we don't sort of need to go in here today sort of took a little bit of a backseat. And then you sort of have the surround ADAS sort of come in, which becomes maybe a more cost-effective solution, something that your customers think they can sort of implement a little bit more broadly. But then on the other side, right, you have the drive or the robotaxi business, and that seems to be where there's a lot of at least market and media enthusiasm for that business as we start to see some businesses scale. So I guess I just want to sort of to start the sort of conversation and talk to you, like how does the -- where is the company sort of focused right now?

Is it really more on getting these surround wins with customers sort of across the finish line? Is it more on the robotaxi side? And then even on the robotaxi side, is it really sort of technological development or operational and process development that needs to get things out?

Yes. No, I mean I think that's a very good question. So I think from a business development perspective, a sales perspective, you tend to focus on the area where you have pull from whoever your customers are, other ecosystem players. And there's definitely a lot of pull right now for a next generation of ADAS systems for mass market vehicles that will bring vehicles to kind of the next level of safety that creates safer vehicles, but also meets future test criteria, which are getting more and more difficult, provides a system where consumers can take their hands and feet off the controls on the highway in a very kind of predictable, wide-ranging manner, automatic lane changes to create convenience and tech-forward systems, but those systems want to be much lower cost and more scalable than what are on the road today.

So I think that the OEMs feel like that combination of sort of a have to do safety improvements combined with something that they believe will be attractive to consumers, consumers would pay extra for this technology, but not have to pay too much extra seems like kind of a very attractive place to be, and we're seeing a lot of traction there. And so there's a lot of effort going into marketing and getting design wins for that technology. I think on the robotaxi side, the pull -- I mean, the pull is certainly from investors, but the pull is also from demand generators like Uber and Lyft and Bolt as well as like public transit operators who see the potential for dealing with driver shortages and a more efficient transportation system.

So you have a lot of pull from those players. and that's creating a lot of activity in that space. I think in that midrange of SuperVision and Chauffeur, which are kind of premium products for premium vehicles that don't really have like a half to do component to it is probably less traction than the other 2 right now. I think -- but the main priority of the company is not necessarily to get more design wins right now, it's to execute the ones we have because we do have production programs with Volkswagen Group for all 4 of these categories. And our view is if we do our job and kind of execute the products and have the performance that we're expecting that the OEMs are expecting that, that will drive more business for these technologies like across the board. So that's really, I think, a core focus of the company over the next year to 2 years.

So let's go back to sort of the surround ADAS. You obviously have the initial Volkswagen win, which starts in '27, right? And then last quarter, you sort of announced the nomination for sort of another surround ADAS. And I think there was sort of the implication that there's sort of -- that has sort of been opening up more and more conversations. So maybe you could just give us an update on sort of where some of those conversations stand. And look, I understand that these are large important decisions by these organizations. But like what is key? Like what are they really looking for to sort of get over the finish line here?

Yes. So I think that there were already like a number of active discussions before the second design win, but that one is I think it's a great demonstration of why the traction is happening to begin with, like this particular OEM believes that hands-free on highway driving is going to be a critical feature to offer to consumers, and it shouldn't be only on high-priced vehicles. This particular OEM has an existing sort of Gen 1 system like this that because of cost, never really could scale to lower-priced vehicles. And so their direction is that this technology would be on all of their cars eventually. I think that what we were told by them is compared to the kind of original plan, which was to have like an EyeQ6 Light for ADAS and then kind of go to the next generation of this Level 2+ system they had because of the cost efficiency of surround ADAS, because of the fact that you don't need as many sensors is that they were planning, we're saving them $1 billion.

So for us, it's a very large kind of uptick in revenue with this particular OEM. But for them, it's a $1 billion cost savings, and it's kind of lowering -- relative to what they were planning to begin with, and they believe the performance will be better, and it will be kind of like a differentiating factor for their cars.

And do you think that dynamic where it's a revenue generation for you, but a cost savings for other -- for automakers can exist at other potential customers?

I think it definitely can because the original kind of direction for these Level 2+ systems was essentially split the -- create control for the OEM by splitting the supply chain. supply chain, source radars from one company, cameras from another company, chips from another company, certain software, DMS, but you end up with 4 or 5 different ECUs, too much cost and a group of technology providers that aren't really all on the same page. So this more of a vertically integrated system creates cost savings. So you're looking at a big revenue benefit to us, but a cost savings for the customers.

Okay. I guess then on the -- switching gears to the robotaxi side. I think one of the challenges here is, I think, that investors sort of want to understand not only where the technology stands but sort of what the business model ends up looking like. Because I think one sort of view here, right, and I know we were having this conversation a little bit earlier is that you could have great technology. But at the end of the day, it does seem like you're somewhat beholden to your customers in terms of sort of getting that technology out on the road.

So maybe you disagree with that, maybe you don't. I'm curious to sort of get your views. But how do you sort of -- I guess, why did the business model evolve that way, A, B, like is there an opportunity to -- as there is sort of more demand for robotaxis and there is, I would still say, supply availability like I was just out on the West Coast and the Waymos are -- there are more of them there are everywhere. But like they are -- I would still say, probably not fully scaled, they're somewhat limited, and that's because, right, they are beholden to getting the Jaguars or the Zek, eventually the Hyundai sort of out there. Is there an opportunity for you to sort of take some vehicles, a small fleet, right, but just sort of show what you can do and almost force their hand to sort of move the needle forward with you?

It's a complicated question. I mean I think that it's easy to oversimplify this business and say, okay, you need somebody to produce the car, you need someone to develop and deploy the technology that makes it self-driving. You need someone to operate the car and you need a demand generator. I think where we're seeing the most uncertainty or like the most flux is in that kind of own and operate section because it's easy to think of it as like you finance cars, you put them into a fleet, you make sure that there's facilities to maintain and clean and charge them, and that's it.

But like what about insurance? What about what about orchestration of the vehicles? I mean utilization is going to be the driver between profit and loss in this business. So how do you create software and orchestration to kind of optimize the utilization of these vehicles. Customer service, right? You need like a pretty massive customer service support system.

So -- and then there's like questions of liability, like who's responsible, even like the cleaning solution. So you have a vehicle that's operating perfectly, but because the cleaning solution on one of the sensors broke, it gets in an accident. So is that the OEM who produces the car? What about if you upfit the vehicle at a kind of secondary center, who takes responsibility for all of that?

And I think we're seeing OEMs be a little coy about kind of how -- do we just want to sell cars? Do we want to own and operate cars? What do we think about liability? That determines whether the system is integrated on the assembly line or somewhere else. The ride-hail players originally were just like bring us your cars and we'll put them on our network. Now they seem more open to that kind of owner-operator role. So -- and then like once you get the technology ready, I completely agree with your point, like you could be in a situation where the technology is ready, it's performing well, but because of logistical issues or being with a partner that missed a few things like that could really slow down your scaling. So -- and I don't think we want to be in a position where we're relying on 1 or 2 players to do that right.

So we do think about -- I don't think we would ever be the owner-operator of the vehicles, but could you do something without OEMs where you just buy cars, you upfit them and then you take on a different partner that may be more of a neutral player, maybe not be a ride-hail player or an OEM but has these types of skills in orchestration and customer service and deep insurance ties and is willing to take a role in the integration of the system. So I think that, that's something that you can probably expect from us.

But even for -- whether it's you or whether it's one of your partners, right, even like getting 10, 20 vehicles out there. It's like not that big of a commitment, but can make a meaningful perception difference, right, and getting those vehicles out there. So I guess, like what is sort of the -- I mean, I know -- well, maybe you could just remind us, I guess, of the time line you sort of expect for sort of the initial sort of rollout of some of these vehicles, taking the driver out. And let me take a step back. Maybe you could just sort of, again, tell us where you think the technology stands right now? Is it ready? When will it be ready to sort of get to what, let's say, a Waymo can do today? And then what's sort of the process for getting cars out there, having -- doing rides with the safety monitor, taking the safety monitor out and scaling from there.

Right. So just to level set, we have about 150 vehicles operating in 6 or 7 different cities around the world. They all have someone behind the wheel, but different -- in different cities, there's multiple cities where there are like closed user groups, invite-only user groups that are taking rides. So it's more than just car with safety driver driving around, finding problems, reporting them, this type of thing. Our -- when we started the program with Volkswagen to integrate our drive system into ID. Buzz, it was about 2 years ago, and there was a time line and glide path that was intended to create milestones of performance. And the meantime between failure, I mean it's an upward sloping curve because you want long, long periods of time between failures.

So we've been hitting these milestones consistently. Volkswagen is a tough customer, and they're evaluating it all the time. And we're on a glide path to get to the point where we have both agreed that you remove safety drivers when you hit this number by the middle of next year. I think we would do that first in a relatively kind of simple environment, validate the on-the-road system and then bring it to a market like Los Angeles where we have an engagement with Uber. So that's the goal is kind of by the middle of '26, have safety drivers removed in one area and go into more scaled commercialization by the end of the year. That's in the U.S. In Europe, there are actual kind of certification requirements for removing safety drivers, and it should be about a 1-year process and Volkswagen is at the point where they can start to undertake that. The technology is at a point where they can start to undertake that process now and try to complete it by the end of '26, and you could start to commercialize in Munich, Camburg, Berlin, Oslo in early '27.

I know this is a difficult question because there's limited available data. But I guess, like you're talking about sort of mean time between failures and you see the Waymo is out there, you see Tesla sort of trying to do something in Texas and maybe some other areas as well. Is it as simple as you think that Tesla and maybe even Waymo are just operating at a different level of risk tolerance than Volkswagen and maybe even yourselves? Or is there an actual delta in the technology at this point?

I think Waymo is the leader, right? It's not debatable. They spent a long time, had kind of long-term operations in different cities. And have been reported -- disclosing data that would indicate that their cars are experiencing less incidents than human-driven vehicles. So it's very impressive. It's still only 2,500 cars, approximately maybe it's 3,000 now. So it hasn't really scaled. They've been doing a good job expanding cities. It's impressive, and we think it's been a big benefit to us in terms of like activity and demand and these types of things.

So I don't -- I think that they have a high -- I think that their risk tolerance is in a good place. I think Tesla is we don't really know what they're doing. I was in Austin a few weeks ago and talking to the team there, they were basically saying we like rarely ever see these Tesla robotaxis around. The public-facing data for FSD is -- has improved quite a bit, but it's still maybe 1,000 miles between critical intervention when our view is you need to be in kind of the 10 million mile range to be at human safety levels.

So I think -- but they still have safety monitors in the vehicle. So I can't really speak -- so I think it's another point of we want our risk tolerance to be balanced and I think working with OEMs, like you could get yourself into a situation where you feel like you're ready, but the OEMs not decide they're not ready. So we also want to try to solve for that. We don't think that that's going to happen, but having multiple partners, maybe having your own fleet in order to at least create a catalyst where you're proving that you're on the road, no safety drivers operating because that's not just for investors, but for other customers, for other potential partners.

Right. Yes. I think it's interesting because like you did say Waymo is, again, still a relatively small amount of vehicles, but they -- it has -- in terms of even vehicles in certain locations, it has recently, I think, hit somewhat of an inflection, right, in terms of areas where they're either operating or will be operating soon. I'm curious if you think -- is that -- in some respects, like when they're entering these markets, right, they're helping pave the regulatory path. Does that make it easier for you as you sort of come to market with your partners to sort of be able to maybe not deal with some of the lumps and growing pains and costs associated with entering markets?

Yes. I mean we have a WhatsApp thread that is essentially just packed with whatever people can find out about Waymo because there's like a lot of lessons learned here. I mean even in -- I think there was some story about these kind of random things that you wouldn't...

Hopefully, not one way ran over the cat.

No. I mean you don't want to run over pets for sure. I'm kind of spacing that there was a story recently that was just an unusual thing for people to get annoyed about, right? So how do you -- so preparing for these things. So I think paving the regulatory environment, paving the expectation of consumers seeing that consumers are seem to be gravitating to these vehicles despite the fact that the cost is not lower, the time from point A to point B is not faster. It's probably slower.

The wait times are longer, but people are -- they're still generating a lot of market share because there's benefits here. So yes, I think it's beneficial. And we feel like we have scaling advantages over Waymo that assuming the precision and the performance can get there along the time line that we're expecting, we believe that there's ways that for us to compete very well -- us and our partners to compete very well with Waymo in the coming years.

What about just -- I think this used to sort of come up a little bit more, but I'm curious whether this still comes up in some of your conversations and sort of what the company response is now just to sort of the CapEx question, which is that you look at what Tesla is spending. We don't really know what sort of Waymo is spending, but they obviously have sort of the parent company sort of balance sheet to sort of help with that.

And you guys are relatively CapEx Light. So do you still not see that as an issue towards sort of progressing the technology? I mean I did note -- I think like last week, maybe there was that sort of AWS announcement you had. So I don't know whether that was new or incremental in terms of sort of data center space or usage. But maybe just sort of talk again about your sort of strategy there on the technology front and spending front.

Yes. I mean we're progressively growing the -- what we spend on cloud compute for training purposes. We're consistently spending $40 million, $50 million a year on GPUs for our on-prem data centers. We just don't think that money is the constraint here. We don't think it's -- training compute is a constraint on us.

We don't think people -- ability to hire people is a constraint. I think on the robotaxi side, I mean, you could definitely have more test vehicles. But again, like generally, like the interventions and the difficult scenarios you're running into are -- the bulk of them are happening in specific scenarios. So having more vehicles out there driving, trying to encounter corner cases, we don't really think helps that much. So we don't see a big inflection coming in the CapEx or OpEx.

I should note to those in the audience, there's a QR code on your table. If you snap that, you'll be able to ask a question. It will show up here on my iPad, and I could ask Dan on your behalf. So please, if you have questions in the audience. feel free to submit them. Maybe let's sort of go back a little bit to sort of the here and the now. So last quarter, you raised the low end of the guidance. I think it implied about, call it, 8 million units for 4Q. That's a little bit -- or that is below, I think, sort of what you've been running on a sort of evenly annualized basis. Now I know there's been a lot of sort of seasonality and you sort of talked about maybe a little bit of destocking that sort of was occurring. But 3 weeks left in the year, sort of how are you feeling about that number?

Yes, we feel good. Like we haven't seen anything that would impact kind of the business that we thought would occur at the beginning of the quarter. We did -- the volume number in Q4 is lower than the rest of the year, but we're proud of kind of the tools and processes we put in place to really very closely monitor stock levels that are out in the channel, and we feel like we'll be shipping below demand in Q4, and that will lead to kind of a tight inventory situation to start 2026. And as we start to see like Q1 orders come in, I think that will be borne out.

Okay. So I guess, as you start thinking about '26, I know we'll sort of get your guidance here in a couple of months. But if you look at just sort of global production, I think SMB has it down a little bit. I know you sort of more reverted to your sort of top 10 customers because obviously, that global number includes a bunch of business that's sort of not really available to you. But I think also probably flat to sort of maybe down a little bit seems reasonable. And I think you're sort of still saying even in that environment, you think you'll be able to sort of outgrow mid-single digits or so. Is that still the algorithm we should think about?

Yes. Yes. We still -- we have market share kind of like embedded market share growth with certain of our customers because of programs we won a couple of years ago. I think that we probably wouldn't forecast it, but like the Chinese OEMs, like export volume has been growing. We've got a good kind of position there, especially for vehicles exported into Europe. I think India remains like a very...

When does this start to get meaningful? Because it's obviously a big opportunity. You've got some good alignment with customers there, but...

Yes. I mean, I think '27 is a year where you could start to see like multiple hundreds of thousands of units of growth. It's still smaller than that, but the planning for vehicle launches that happen towards the end of '26 and over the next couple of years is that ADAS is standard in some of those vehicles. So we do feel like that's a big growth opportunity for us. So yes, I think like that's similar to what happened this year was we grew about 5 points faster than our underlying customer production. And I think that's a reasonable way to look at the future as well for volume specifically.

And then on the SuperVision side, I think you've, again, sort of been fairly volatile. I think you sort of talked about maybe 12,000, 13,000 units a quarter. Is that still a reasonable baseline to think about?

Yes. Yes. We kind of closely look at kind of end market demand or volume for the vehicles that we're on because there's only a few of them. So it's like fairly easy to track. That set of vehicles was doing like 12,000 a quarter throughout most of 2025. It's actually gone up in the last few months to more like 15%, 16%.

So maybe there's some upside in kind of what we did this year. But I think the positive is that all of those vehicles are doing pretty well in the marketplace. And so I think that business should stabilize.

Okay. When does -- is it '26? Or is there sort of more '27 when the mix of chips on sort of the base ADAS side of the business sort of starts to really kick in a little bit more and help?

Yes. So right now, the bulk of the volume is still EyeQ4. EyeQ5 will never be like a super high-volume chip, but I think this year, maybe it's 6% of the volume. And next year, it's going to be 10% or 11%, something like that. And then EyeQ6 Light will start ramping kind of very rapidly. It's already in production, but it will.

It's short cycle to EyeQ5, is that?

EyeQ5, like I wouldn't -- EyeQ5 was really developed for the first-gen SuperVision, but we created -- and that's the high version. We created a mid- version as well. It was never really intended to be a high-volume chip, but there's a few programs there. They're typically higher ASP, but also the cost is significantly higher. So anyway, I think that -- but EyeQ6 Light is -- the intent there was to maintain kind of typical pricing and typical margin, but for more performance for better features.

And so EyeQ6 is not going to create higher ASP. EyeQ6 Light will not create higher ASP or higher margin. The margin on EyeQ5 is lower. So as that ramps up a little bit more next year, it creates a bit of a gross margin headwind. But yes, I think I wouldn't be expecting significant upward movement in ASP next year.

In '27?

In '26. '27, it will -- we will have some of that. Yes, '27 should be very good because you've got the initial kind of Porsche, Audi SuperVision launches happening kind of at the tail end of this year. Even if they got pushed a little bit, you'd still have significant volume in '27. You have the initial surround ADAS volumes in the back half of the year. You have initial Chauffeur volumes in the middle of the year. And if things go well with robotaxi this year, you could be getting into the thousands of units in '27.

And that's just with the MOIA Uber one? Or does that include the Lyft deals or...

I mean just with MOIA Uber, it could be in the thousands. But I think you also -- you have to consider the European volumes because there's really good demand in German cities in Oslo. There's good demand for the whole on shuttle as well. So you could start to see volumes there.

And remind us what the Lyft portion, when is that?

The Lyft portion is also targeted for the end of '26, but we haven't named an automaker yet. So I would expect maybe that gets pushed a little bit. But that's -- we're doing the work with Lyft now to create the interfaces between their network and our drive system, and that would be consistent for whatever vehicle it ends up going in.

Okay. You mentioned in Europe, but it also seems like on the robotaxi side in Europe, the competitive environment is a little bit different with the Chinese way more aggressive there of late. So how do you view that competition? And I guess, the Mobileye positioning versus the Chinese even from a geopolitical perspective?

Yes. I mean it's a very good question. We don't have great visibility on the Chinese technologies. We know that there's a decent amount of vehicles operating within these zones in Chinese cities. We know that there's been several announcements in the Middle East and in cities in Europe of launches, right?

But like calling like putting 5 vehicles on the road with safety drivers a launch is not really -- that means we've launched in 10 cities starting 4 or 5 years ago. So I'm not trying to downplay it. I think we have like a lot of respect for the skill sets of these Chinese players. But beyond even geopolitical concerns, which nobody is really talking about in Europe yet, but we think that, that will ramp up at some point. I think that they've probably got a long way to go to develop technology that works in Europe. Even ZEEKR, which has a very good operating like supervised hands-free system in China, continues to use our SuperVision system for vehicles sold anywhere else besides China.

One more sort of nearer-term topic that's sort of become a little bit more -- a little bit louder, let's say, in some investor conversations over the past couple of days is just reports about potential memory shortages as the memory makers sort of shift their capacity to AI and higher-margin products. And I didn't mention ADAS is sort of one of the areas that could feel a little bit of a pinch. I mean what -- can you just sort of tell us sort of what you use, how you use it, are you sourcing it?

Is it sort of the Tier 1 who's sort of sourcing it? And what's sort of your outlook? Like how real is sort of this fear? Could there actually be volume shortages? Or is it maybe -- does it maybe result in just some higher prices for the memory?

Yes. So in our core business, we sell a chip to Tier 1s and the Tier 1s put that chip on a circuit board and kind of install the camera. It's the circuit board that has the DRAM on it. So it's the Tier 1.

You're not responsible for sourcing.

We're not responsible for sourcing that. We have a good knowledge of kind of the entire design of the system. So we know kind of where DRAM is coming from. And we did start to hear about this a few weeks ago, got in touch with Tier 1s. We're setting up war rooms, task forces, like the industry is very good at doing that.

Mobilizing for a crisis.

Yes, mobilizing for a crisis, like there's been a few. What we're being told is they don't expect production impacts. like as we start to see Q1 orders come in, they're kind of in line with basically what we were expecting. So we don't see any impact in Q1, at least so far.

Okay. So -- but is it something that could become an issue in '26? Or is it maybe more '27? Like what type of insight do you have into that market in terms of when there might be some.

I think we don't know enough yet. I think we don't know enough yet. I think pricing is -- yes, like cost increases because it does seem like -- I guess what we've heard is there is enough capacity to support this business. But you may need to resource certain components to other producers of it, and that could take some time and maybe there's like pricing opportunities for those memory suppliers that could create some cost pressures, but it shouldn't affect us because we're not the source -- we're not sourcing it.

Maybe just to close here, Dan, I'm curious what sort of insight you could give us in terms of the conversations you're having with potential customers as they evaluate the competitive landscape in terms of willingness to look to competitor meaning auto competitor solutions.

So obviously, Tesla and Elon has sort of talked about a willingness to sort of license out before. I think that has been an element of sort of a positive thesis for Tesla in some respects. I think you sort of -- Elon may have even recently sort of downplayed that a little bit saying like the automakers just aren't doing that. But then you still have other companies like Rivian saying like we're open to partnerships and. So how do you view that sort of slice of the competitive buy?

Yes. I mean I think it's always -- there's always a lot of challenges for OEM competitors to cooperate on kind of technology that's seen as strategic and advanced. I think like Elon's recent comments about, well, we did get companies reaching out, but they only wanted to do like a couple of thousand units, like that's actually kind of what we've been seeing, too. So some of these kind of discussions on SuperVision or Chauffeur that started out with we're going to put this on 5 or 6 platforms, and it's going to be hundreds of thousands of units a year.

The decision kind of at the end of the day is like, well, we want to do more of a pilot project that would be on 1 vehicle, maybe 10,000 units over a few years, and that's not interesting to us. It's certainly not interesting to Tesla. But I think that this kind of lack of conviction in what to do with like significantly advanced premium ADAS systems is real. Yes, I don't -- I think that we're not seeing appetite by our customers to commit any real programs to like one of these start-ups that are out there or another OEM.

Could you see things like with Nissan and Wayve at extremely small volumes a few years from now on a small scale? Yes, but it's -- I think we're at the point where we believe our products are ready for scale, and that's kind of what we're pursuing.

Great. Well, I think we're out of time. So thanks again for your time. Look forward to seeing you at CES again this year, and thanks for joining us again.

Thanks, Joe.

Take care.

Greetings, and welcome to the Mobileye 3Q '25 Earnings Call. [Operator Instructions] As a reminder, this conference is being recorded. It is now my pleasure to introduce your host, Dan Galves. Mr. Galves, you may begin.

Thank you. Hello, everyone, and welcome to Mobileye's Third Quarter 2025 Earnings Conference Call for the period ending September 27, 2025. Please note that today's discussion contains forward-looking statements based on the business environment as we currently see it. Such statements involve risks and uncertainties. Please refer to the accompanying press release, which includes additional information on the specific factors that could cause actual results to differ materially.

Additionally, on this call, we will refer to both GAAP and non-GAAP figures. A reconciliation of GAAP to non-GAAP financial measures is provided in our posted earnings release. Joining us on the call today are Professor Amnon Shashua, Mobileye's CEO and President; Moran Shemesh, Mobileye's CFO; and Nimrod Nehushtan, Mobileye's EVP of Business Development and Strategy.

Thanks. And now I'll turn the call over to Amnon.

Hello, everyone, and thanks for joining our earnings call. Starting with the results. Q3 revenue of $504 million was up 4% year-over-year. The driving force was 8% EyeQ volume growth significantly outpacing the 1% growth in overall vehicle production among our top 10 customers in Q3.

Operating cash flow was again a highlight. We generated $167 million of operating cash flow in Q3, well above net income. On a year-to-date basis, we have generated nearly $500 million of operating cash flow, up around 150% year-over-year. This reflects the cash generative nature of our business and our continued discipline in managing working capital. The core ADAS business is performing well, with volumes in very healthy range for the last 5 quarters and expected to do so again in Q4 based on our updated guidance.

We again raised the midpoint of our full year outlook this time by 2% in terms of revenue and 11% in terms of adjusted operating income. Compared to our initial 2025 guidance, these increases are even more pronounced, 7% for revenue and 27% for operating income at the midpoint. Overall, we expect volumes to come in about 2 million units higher than our original guidance. This outperformance reflects a combination of stronger-than-expected launch activity, ADAS adoption growth and better than accepted results in China both from our shipments to Chinese OEMs and from the performance of our top 10 Western OEM customers in China.

If we adjust our inventory digested in 2024, our 2025 volume growth is expected to outperform the production of our top 10 OEM customers globally by about 5 percentage points. We see continued momentum as EyeQ6 light is generating ADAS program wins at a high rate. So far this year, we already have been nominated for programs with future expected volumes well above our full year 2025 volume.

We have added a customer in Volvo. The growth potential in India is becoming increasingly clear as strengthening adoption trends and supportive regulatory environment. Additionally, we are seeing continued traction in adding REM to front-facing camera programs further reinforcing our base business.

On the advanced product side, our position is differentiated in the fact that we are an OEM neutral platform that is cost-efficient and scalable with a credible technology path to eyes-off autonomy in both privately owned vehicles and robotaxis. All 4 of our advanced products surround ADAS, supervision, Chauffeur and drive share common building blocks including the EyeQ6 high inference chip, substantial portions of our perception policy AI tax and [indiscernible] crowded crowdsourcing driving intelligence and robust safety frameworks and the company's comprehensive data and validation infrastructure.

EyeQ6 high-based surround ADAS systems continue to develop as the next generation of standardized driving assist on high-volume vehicle platforms. This system addresses multiple objectives in a cost-efficient package. It's designed to meet stricter late-decade safety standards enables highway hands-free performance for lower cost and current systems and supports OEM goals to consolidate ECUs and to integrate technology on a single SoC.

We have meaningful traction with a number of OEMs and very recently, received confirmation from a leading Western OEM that were nominated for a high-volume EyeQ6 high-based surround ADAS program across mass market vehicles. We continue to pursue a number of promising supervision and Chauffeur opportunities, although timing remains difficult to predict. The best way to ensure eventual new customers is to focus on execution of the supervision and Chauffeur production programs with Volkswagen Group, where we are first movers. Near-term execution includes major software drops in the coming few months that embodies significant innovation in AI.

A few weeks ago, we received the first silicon sample of our next-generation SoC, the EyeQ7 high and all initial tests have been successful. EyeQ7 and its successor EyeQ8 now in design stages are designed for upgrading eyes-off autonomy to [indiscernible] autonomy. Eyes-off systems is what mobilized breaking to production in early 2027 and also describes the current technological state of robotaxis. In both cases, there is a human either the driver or a teleoperators, that can resolve issues when needed. In the minds-off system, which we are targeting for 2029 and beyond, there is no human to resolve issues, and therefore, the driver can sleep and the robotaxi no longer needs teleoperatorster to intervene. This transition from eyes-off to minds-off where EyeQ7 is going to play a meaningful growth. More details will follow in the coming months.

Turning to robotaxi. Our engagements are expanding through both Volkswagen and Holon, a division of Benteler. Once we remove safety drivers in our first U.S. city in 2026 and secure type approval for self-driving vehicles, the Volkswagen ID.Buzz and Holon organ shuttle in Europe, we anticipate geographic expansion. VW and Uber in Los Angeles, Lyft and Holon in the U.S. and multiple commercialization initiatives with Volkswagen and public transport operators across Europe.

Additionally, we continue to get closer to naming an automaker and vehicle platform to complete the [indiscernible] any value chain. That will enable preparations for commercialization in Dallas and other cities to accelerate. On the robotaxi technology front, we continued to outfit more of the ID.Buzz test fleet with a full EyeQ6 high-based production hardware and have successfully completed the first closed-loop testing with the Holon production vehicle.

The MTBF performance is tracking well against the KPIs that are required to remove safety drivers in 2026 and begin commercialization. Everything continues on track. In summary, the opportunity set in front of us today is larger, broader and more urgent than it was when we went public in 2022. Near-term volumes remained strong. The demand for higher performance at lower cost is intensifying and eyes-off capability whether for personal cars or robotaxi is no longer seen as experimental science project, but as an achievable and commercially viable reality. This is exactly where Mobileye excels.

I'll now turn the call over to Moran.

Thank you, Amnon, and thanks for joining the call, everyone. Before I begin, please be aware that any comments on profitability may refer to non-GAAP measurements. The primary exposure in mobilized non-GAAP numbers in amortization of intangible assets, which is mainly related to [indiscernible] acquisition of Mobileye in 2017. We also exclude stock-based compensation.

Our Q3 results exceeded the color we provided on the Q2 2025 earnings call in July, with revenue up 4% year-over-year versus our prior outlook of roughly flat. The upside was a combination of EyeQ volume, which came in at 9.2 million units compared to the outlook of 8.7 million to 9.3 million, and SuperVision volume, which was higher than 20,000 units in the quarter, a meaningful uptick versus Q2.

Just a quick note on SuperVision. Volumes were higher than prior quarters, but should not be viewed as establishing a new higher run rate. We now expect around 50,000 units this year. This full year number is significantly higher than our original expectations and a good reflection on what we see as a sustainable run rate heading into next year for the current first generation programs applied to ZEEKR export volume and Polestar 4. These programs remain relatively small within our overall business and quarterly volumes can fluctuate as they have this year.

Our gross margin declined by just over 100 basis points year-over-year basis. EyeQ ASP was down about $0.50 year-over-year. This was primarily due to higher volume of Chinese OEMs, where pricing remains a significant headwind, as we've discussed before. Another factor was higher volume of ADAS program based on EyeQ5 which carry lower gross margin due to higher costs. EyeQ5 currently represents about 10% of volume and is expected to peak to around 15% next year, creating some continued pressure. Beginning in 2027 and the more profitable EyeQ6 light significantly ramps up EyeQ5 share will go down, providing a tailwind to margins.

Operating expenses were up 4% year-over-year, which was a bit higher than what we expected due to timing of engineering reimbursement. We continue to expect overall non-GAAP operating expenses in 2025 to be up about 7% to just below $1 billion. As Amnon mentioned, operating cash flow was $489 million through the first 3 quarters of the year. This is primarily due to strong cash flow from the core business. However, we've also managed tight control over the working capital accounts, particularly our balance sheet inventory which came down by about $100 million year-to-date. We are now well aligned with our 6-month target for balance sheet inventory, and we expect working capital to be more cash natural going forward.

Turning to full year guidance. We are increasing the revenue midpoint by 2% and the adjusted operating income midpoint by 11%. Our full year outlook is based on EyeQ volumes of 35 million to 35.5 million, up from 33.5 million to 35.5 million. Earlier in the year, we maintained an unusually wide range to reflect macro uncertainty and ensure conservatism. Without condition now better clarified, we have greater confidence in narrowing the range and increasing the midpoint. Given 27.3 million units year-to-date, the implied outlook for Q4 is 7.7 million to 8.2 million. At this point in the year, we expect that Q4 volume will end at the higher end of the guidance. We retained a small buffer to account for any unforeseen year-end logistical issues or OEM production constraints to stay cautious.

In terms of understanding the current run rate of volume, we think it is best to look at the full year. This is particularly the case in 2025 where normal seasonality was affected somewhat by tariff timing and expectations. Typically, global production is stronger in the second half than the first, but that pattern did not hold this year. Bottom line is the lower Q4 volume compared to Q3 and Q2 should not be interpreted as a trend. It simply reflects an alignment of supply and demand across the full year to ensure customers enter 2026 with lean inventories.

As noted earlier, SuperVision volumes are tracking ahead of expectations, and we are modestly raising the outlook to low 50,000 units at the midpoint versus prior outlook of around 40,000 and original outlook in the low 20,000. We expect full year gross margin to be right around 68%, implying a slight uptick in Q4 versus Q3. The full year is expected to be up about 30 basis points year-over-year, pretty consistent with our July commentary. Operating expenses, as noted earlier, are expected to be up 7% year-over-year to just below $1 billion, in line with our original outlook.

Thank you, and we will now take your questions.

[Operator Instructions] Our first question comes from the line of Aaron Rakers with Wells Fargo.

I guess the first question is, you mentioned a Western OEM design win that you've achieved. Can you just remind us again, is that additive to the prior kind of development engagement status that you've outlined previously? Is that reflective of the prior Western OEM that was on that list? And just anything around timing of volume contribution? And then I have a quick follow-up.

So to be clear, the confirmation for a nomination that Amnon referred in his remarks, is for a second [indiscernible] program. We have announced previously in the year our first surround ADAS program. This is the second [indiscernible] a second OEM, a leading Western OEM with significant volumes with multiple vehicle models, and we expect this to be a significant portion of that OEM's vehicle lineup in the future. And we will disclose more details on this in the next few weeks.

Yes. And Aaron, this is Dan. You might be referring to the IR Day, the Investor Day slide from last year. It is one of the OEMs that was on that chart for surround ADAS.

Yes. Dan, I appreciate that. And then talk a little bit about gross margin. You highlighted the fact that EyeQ5 volumes at 10% would go to 15%, and that would continue to be a headwind to gross margin. As the EyeQ6 volumes start to ramp. How do we think about the delta or the gross margin inflection as we think about 2026 between those platforms?

Yes. So just to highlight that EyeQ5 volume doesn't have a lot of running programs or production programs, and we don't anticipate a new programs with EyeQ5. So all the new launches we have with EyeQ6 light. The profitability is not that different between them. It's just that it has a bit lower profitability than on EyeQ4. As for EyeQ6, that again is launching for a new program, the profitability is, of course, the gross margin is higher than EyeQ5 and very similar to EyeQ4 that we are currently selling. So it's not some significant headwind in just a matter of platforms or specifically -- specific vehicle launches, mix between products, sometimes some of the projects ramp more. Some -- it's not something you can anticipate, but it's not very dramatic in terms of gross margin fluctuations.

Exactly, yes. I don't think we're going to specify like the impact. But the bottom line is EyeQ5, like percentage of total will be the highest next year, around 15%, so not too meaningful versus this year and then start to go down.

Our next question comes from the line of Edison Yu with Deutsche Bank.

This is [indiscernible] on for Edison. My first question is on the 4Q expectations. Just curious if there's any other factors that you're factoring into new market or reasonably, you heard about the chip issue, whether you're digging that into the outlook? And then the second question is wondering if you can provide a bit more details around the list and demo program? The launch time, the economics, et cetera.

Yes. So I think on the Q4 volume, I think the point we're trying to make in the script is that you should look at the full year volume of kind of around 35.5% as the right number. Like when we're looking at '26, we are -- the starting point is 35.5%. It's not Q4 volume times 4. So there's really nothing going on specifically besides seasonality was different this year due to really, we think, because of the tariffs and trying to pull ahead some production into Q2 and Q3. So this is kind of exactly within expectations for us, shouldn't be seen as a trend.

And I think that we see the trend of around $9 million a quarter. And that's -- there's -- in terms of [indiscernible] we have not received any indications of request to reduce production, reduce shipments at all. In fact, if anything, it's the opposite. It's a very new situation. We don't -- in talking to customers, we don't expect any material impact in Q4, but we do have a bit of margin in the guidance versus the high end to account for that if there's a bit of lower production, but we don't expect it to affect Q4.

What was the second question, if you can repeat it?

It was on the [indiscernible] program, you can talk about -- the evolution of that win, the launch time and the economics? So if you refer to the Lyft robotaxi program. So we are working with Lyft and [indiscernible] on a robotaxi program in the U.S. We disclosed the first city will be in Dallas Fort Worth. We are now in advanced testing stages of this program, and it follows the leading program we have with Volkswagen Group for robotaxi activities in the U.S., and it's been tracking well. And the launch date will be disclosed in the near future.

Our next question comes from the line of Joshua Buchalter with TD Cowen.

I wanted to ask about the metric you gave about normalizing for inventory. I think you said you grew volumes 5% more than your top 10 customers. Is this sort of a rule of thumb we should be using of your expectations for sort of normalized unit growth in '26 and beyond as you see ADAS market and attach rates develop and then we layer in ASPs more. I'd just be curious to hear if that's like sort of a normalized growth rate you think is the right level for us to benchmark to on a unit basis?

Yes. I think the key here is that investors and analysts should focus on kind of the expected volume of our top 10 customers, which are mostly legacy OEMs and has been a bit below kind of the overall vehicle production for the last few years, not meaningfully, but a bit below. And then we would expect to grow faster than in volume and revenue, we would expect to grow faster than that level because of things like ADAS adoption growth because of things like growing share within some of those customers because of things like emerging markets like India. So this year, we consider the performance pretty good. We grew about 5 percentage points faster than the top 10 OEMs, which were down 2% to 3%. And we're not going to put a precise number on what we think it should be going forward, but something in that range is probably fair.

Got it. Dan, I appreciate the color there. And for my follow-up, so it sounds like you're speaking to engagements for eyes-off continuing to move forward. And it does seem like there is a good amount of momentum across the industry from both the robotaxi side and in consumer passenger vehicles for eyes-off features. I guess what do you think the OEMs need to see that gets them across the line in these engagements and I guess, any time line you would expect to be able to -- a reasonable time line to expect where you think you will be able to announce some additional Chauffeur or even SuperVision wins?

I think the focus now is execution. We are with the SuperVision, the hardware is on the C sample stage, which is very advanced from a production level. We have a number of meaningful software drops in the coming 6 months. So I believe that somewhere in the first half of 2026, we'll be in a very good position of being very close to production ready with the platforms of SuperVision and Chauffeur. And that should enable us to get more exposures to new program -- new program wins. So the focus of the company in that area is execution also in the robotaxi with a driver is execution. So really 2026 is an execution year and not necessarily focusing on bringing new business, at least not in the first half of the year.

If I can add color to what Amnon said, in the last 1.5 years, we've been working simultaneously on SuperVision, Chauffeur robotaxi for execution in the past, let's say, 8 months, we've added also surround ADAS production program that execution process. Right now, we are on track with the original time lines of all of these programs, and we are in B sample or C sample hardware and stable platforms, which is an important achievement in order to maintain the original time line. And now we're focusing on software iterations, AI innovation and integrating the latest AI stack into these platforms. [indiscernible] simultaneously is a significant achievement for us. And we believe that now within the next few months of showing a very mature production platform that uses production hardware with the latest AI technologies that shows meaningful performance improvement compared to what exists today in the industry is the next big thing for us to show in general to our customers and also for new engagements, and that's expected within months from today.

Our next question comes from the line of Chris McNally with Evercore ISI.

Thanks so much, team. Amnon, I wanted to dive in on the surround and congratulations on the big win. It sounds like there's a different technology path that maybe you and the industry had thought a couple of years ago where supervision would sort of be the kind of a walkway to higher forms of Chauffeur and eyes-off and it sounds like given the industry has been a little bit slower on that front, maybe how much they would charge for it, et cetera, it seems like surround is now that technology gateway. And I just would love to understand from your standpoint, is it sounds like a must for the OEM to hit 2029 regs, meaning they really can't do this with an internal solution or even the solution that you've been providing them originally with the $50 chip. So is this sort of -- is this one of the reasons you're seeing so much traction this is the logical step up of your Level 2 customers to surround?

Okay. So I'll let Nimrod take the first answer and then I'll complete if necessary.

Chris, so I think surround ADAS is a very important category for OEMs because it's not just about new user experiences, but also adhering to emerging regulation in developed markets. And it's a very, very cost-optimized product segment. because it's designed for high-volume vehicles and for pretty much $20,000, $30,000 vehicles and above, it requires very, very efficient design and a close software hardware integration. Mobileye is known to have a very efficient chip and very efficient software, and we can achieve pretty much, we think, the most competitive price point for this product category. And from an OEM standpoint, thinking about whether you want to buy or build a product in that category, it's not just about understanding AI or different software technologies, can you get to such a level of efficiency on vehicles that are in tens of millions per year, if you fail, you may jeopardize your core business.

And so forcing an existing available solution that is very mature is the safe choice. And maybe if you're into in-house development, you can focus this on the higher end of applications and smaller volumes, maybe 1% of your cars [indiscernible] failed or significantly delayed, there is no damage done to the core cash cow of the company. And that's where we see their interest just yesterday, GM announced their Level 3 eyes-off development. That is designed for a very specific vehicle category. I think they disclosed the type of vehicle, and it's a very high price point. As you all know, GM sells cars in $30,000, $40,000 also. Obviously, that solution is not appropriate for these vehicle price points. And it may make sense for them to find a proven, reliable, trustworthy high-performing, cost-efficient solution for the vast majority of volumes, while they focus on the high end.

And I think I'll continue. I think likewise going into Level 2 plus with 11 cameras, our SuperVision, we are working very diligently on very innovative cost reduction schemes. So looking into 2028 time frames and beyond, we can offer significant price reduction on SuperVision. The next level that OEMs are considering our eyes-off systems. And there, as I said before, execution is the key. If we launch an eyes-off system, and that's what we are planning to do in 2027 with Audi, that will be kind of an inflection point. Seeing such a product at work passing through all the regulation, the sub certification and regulatory approvals, passing the MTBF bar that is needed to have eyes-off, this is a significant achievement. Once we pass that bar, I think that will be a big inflection point.

Yes. I think that there is no question beyond -- or about the fact that eyes-off driving and later mind-off driving is the ultimate value proposition for consumers. And we think that most OEMs are very interested and very bullish on this sort of proposition. The question is, is now the right time given the maturity of the technology and the available system is what costs for them to go all in with a partner. Today in the industry outside of China, there is no other technology provider that is working closely on the entire system, hardware, software, silicon, AI -- receiving approvals for testing and going through the ropes of homologation and all the necessary check boxes other than Mobileye with Audi. And all eyes are on us. And hopefully, within months, we'll show more and more evidence of the maturity of the technology getting there. And as Amnon said, we believe that will be a significant inflection point for that product.

Nimrod, my only clarification or a summary of what you said is super helpful. So is it logical then to think like your first customer, which is VW, that basically your target audience it's not going to be 100%. But your target audience for surround is existing basic ADAS customers that now need to convert for 2029 and so your second customer, as an example, upgrades ADAS into surround. And then there's a future path beyond that to eyes-off?

That's exactly the case in that nomination we disclosed. It's an upgrade from EyeQ6 to EyeQ5. That's -- that's essentially the decision that OEM made. And so that's the right summary of how we see things.

Our next question comes from the line of Mark Delaney with Goldman Sachs.

[indiscernible] can double-click on the drive opportunity with [indiscernible]. I believe the company said today, I think they can name the OEM partner for that engagement soon. So should investors assume that the OEM partner is already effectively finalized? Or is there still uncertainty as to which OEM Mobileye is going to partner with? And if there is still uncertainty, what would the time frame need to be to line up the OEM partner in order to meet the 2026 start-up operations objective.

So there's no uncertainty who that OEM is, but it's not finalized yet. So I cannot say with 100% guarantee that it will be finalized, but it looks on track and it looks good. I would say that this is in parallel to our existing activity with Volkswagen of the ID.Buzz and with the whole loan platform with Benteler. So we have quite a lot of scale opportunity with the existing relationships. We have -- this is why I mentioned before that it really focuses execution. Scale and business will come once execution is there. You want to add something?

Just one more clarification, Mark. That vehicle provider, it's not that we will just start working with that vehicle OEM once we will finalize the agreement. In the past almost 18 months, we've been working closely between Marubeni, that vehicle OEM and Mobileye on creating multiple prototype vehicles and then working on the actual vehicle platform itself, integrating our self-driving system with the sensors, and it's all pretty much -- we have numerous vehicles that are equipped with all the sensors and all the compute infrastructure needed as if it's like an ID.Buzz we do in our leading program, and so we are starting -- we're hitting the ground running once it will be finalized and disclosed. It's a natural transition into serious development towards commercial launch.

Yes. And I would just add one more point here is that if there's a bit of wait and see on consumer-owned high-end eyes-on and eyes-off technology, robotaxi is exactly the opposite. There's so much activity, but the key is removing the safety driver and starting to commercialize. And that's our main priority right now. And the core technology is the same, no matter what the vehicle platform is. There's integration work to be done, but it's the core technology that's being worked on. And once we kind of can remove the safety driver, then we can start to commercialize and scale.

That's all very helpful. My follow-up question is also on Drive with VW and the ID.Buzz, you mentioned you're tracking to be driver out next year. Could you just speak a bit more on what still needs to happen in order to take the driver out next year? And any sense of when within the year that milestone may occur?

Well, there are a number of milestones. One is the readiness of the vehicle that should be ready in the next weeks or a few months. And second is the MTBF KPIs, which we are tracking, that are on track. And we believe that in the first half of 2026 we can start removing the safety driver in one city in the U.S. and to prepare for a commercial deployment later in the year and beginning of 2027.

Our next question comes from the line of Joe Spak with UBS.

Just going back to the surround ADAS nomination. I was wondering if you guys could provide a little bit of color as to sort of what got the customer over the line, maybe a little bit of detail on the level of integration that you were doing versus maybe the automaker is DXP involved? And then just in terms of the rollout, is this a case where we need to wait for new model launches? Or can this product be placed on refreshes?

I will take this, Joe. So we think the driving forces behind this decision to upgrade from EyeQ6 light to EyeQ6 high was about basically, the standard sensor set for that OEM is at minimum 5 cameras and 5 radars. And in the older days, these -- some of these cameras was used just for top tier visualization when you do parking, for example, for the human driver. And just the front camera was used for ADAS and the radars were used for ADAS and that created a very inefficient design. And that OEM decided to create a more consolidated ECU architecture in which you can think of this as somewhat of a -- like simplifying the architecture and then routing all the sensors to the EyeQ, the EyeQ6 high is powerful to process all of these sensors and creates a much richer sensing state, a much richer user offering.

For example, hands-free driving in highways, supporting all of the most cutting-edge advanced ADAS requirements in [indiscernible] and so on, as we've mentioned. So it's about system simplification consolidating efforts and routing all sensors that already exist in the car to a more powerful chip, they can process them more intelligently and offer better user experience. The added cost for that OEM was very reasonable compared to the added value as evidenced by their decision. So it's not like a 10x more expensive chip in order to get that value, it's as we've disclosed in the past, it's 2 to 3x more expensive for that silicon component generally speaking. So I think it makes -- as they said it, it makes too much sense not to make this transition.

We are also discussing about potential future -- that with other areas but potential future consolidations like with parking applications and driver monitoring system and ADAS and hands-free driving all can be processed and provided by Mobileye on the EyeQ6 high chip with a very attractive cost and I think that's a compelling proposition for OEMs. And regarding your question about vehicle launches, it's a new architecture. And in parallel, we work on other, let's say, more existing architectures as well. So it's a combination of the two.

And then I guess just as a follow-up, I think in the prior update there were maybe 3 or 4 other sort of advanced engagements about surround. I was wondering if you could just get an update there. And maybe going off of some of the commentary on Chris' question, like have you seen sort of more initial maybe SuperVision move more towards surround in the near term?

So I wouldn't say it's SuperVision engagements moving towards around. It's more about base ADAS engagement expanding to surround from what we're seeing. The first 2 design wins we have for surround ADAS are examples of OEMs with high volumes, the sell cars at relatively low vehicle price points that have, in the past, sourced a front camera solution and these examples decided to source surround ADAS solution for the same vehicle category. So I think that's the evidence that we're basing our assessment on. And yes, we have a lot of engagement. Mobileye works with pretty much all of the OEMs on a recurring basis on what we have to offer, and we see a lot of interest. And I think we will -- we prefer to disclose when we have nominations and it's concrete like we did today. But we remain confident in the strength of our outlook.

Our next question comes from the line of Shreyas Patil with Wolfe Research.

Maybe just to follow up on the earlier one. How do you think about the competitive landscape when it comes to surround ADAS? I think there are 3 or 4 other major suppliers that are trying to win awards here as well. And maybe just to give a little bit of background on the bidding process that went into securing this award, the second award that you just announced.

We're talking about a very highly competitive in terms of cost, cost optimized product. So all the high-performance chips that you hear the price point is not relevant for such a product. Our chip, the EyeQ6 high is both the core chip for our AV, for example, in SuperVision, we have 2 of those chips and [indiscernible] we have 4 of those chips and in Chauffeur 3. But one chip is highly cost optimized, and we can meet the cost desires of OEMs with our system on chip that can process all those multiple sensors, 5, 6 or 7 cameras and radars and ultrasonic and so forth. So we're talking about the game in which cost is highly, highly critical. So we have first mover advantage. So we were the first to receive nomination with the Volkswagen Group on surround ADAS. And the new win of today shows that we are successful in leveraging our first-mover advantage. So it's all about here cost and performance, but cost is critical because we're talking about high-volume vehicle categories.

Just so it's not that there is no competition, but we do have first-mover advantage, and we are showing that we can leverage that.

Another aspect of efficiency that we are -- should be considered beyond just the price, our EyeQ6 high chip does not have any limitation in deploying pretty much all of the state-of-the-art AI architectures while being very efficient in power consumption. This offers us -- this allows us to offer a solution that is [indiscernible], for example, and does not require liquid cooling. So this is a small technical detail, but for the OEMs, it's a big difference in overall system cost in the complexity of the system, combustion engine vehicles may not have liquid cooling available, for example. Other competitors are trying to rely on more -- because the underlying product requires sophisticated processing of sensors and using some of the AI technologies, other competitors may try to use a high-performance chip for that product category. And beyond just the price disadvantage, it also complicates the overall system. So this is another element that is hard to compete with the low power consumption of our chip without compromising performance.

Yes, I'll just give a color in terms of performance of the EyeQ6 high. Our internal benchmarks running on both [indiscernible] transformers show that we are on par and in many cases, better than the [indiscernible], which is now the choice of competition when OEMs are considering the Level 2+ systems. But at a price point, which is less than 25% of it.

So this gives us a great advantage. On one hand, we have a high-performance chip, which is on par with the latest high-performing chips in the automotive industry and on other hand with a cost structure and power consumption constraints that are way, way more appealing.

Okay. That's really helpful. And then maybe, Amnon, I think last quarter, you talked about drive potentially becoming a more material revenue contributor by 2027. Wondering if maybe you can expand on that a little bit. And you've talked about in the past at a relatively high upfront revenue stream, maybe $40,000 to $45,000? How should we think about the rate at which you can bring that down, especially as robotaxi operators are looking to bring down vehicle costs to improve overall unit economics?

Our economics from every robotaxi is comprised of both onetime fee and revenue sharing on price per mile. And over time, we will kind of change the equation, maybe reduce the onetime fee, increase the cost per mile. So we both will have a recurring revenue and also the onetime fee of the system. With the robotaxi, it's really just execution because the current contracts that we have with the existing partners they talk about many tens of thousands of vehicles [indiscernible]. Just to give you a proportion, today's very, very successful Waymos based on 3,000 vehicles. So it's really just a matter of execution. If we execute, we execute our plans of removing the driver during the first half of 2026 prepare this for commercialization beginning of 2027, the volume is there and in very, very big numbers.

And in terms of the upfront cost, we have we have the room to switch around upfront cost versus recurring revenue because we have low costs in general. So we can stay profitable on the upfront cost down quite a bit and kind of replace that with more recurring revenue.

Our next question comes from the line of Luke Junk with Baird.

I want to stick with robotaxi. We've talked a lot about U.S. robotaxi driver out this morning. hoping we could get an update on private storage driver out in Europe as well, especially relative to the different regulatory homologation there and just maybe Europe versus U.S. dynamics in general right now for Mobileye?

Yes. So there are differences between Europe and the U.S. U.S. is mostly a self-certification process which we're doing. And we have today, close to 100 vehicles driving in the 3 locations in the U.S. for testing and working with additional carmakers or additional robotaxi platforms as mentioned, expanding debt further. In Europe, we have an equivalent, let's say, volume of vehicles. The process for launching commercially is around -- is more about homologation and engaging with regulatory bodies in advance before you can commercially launch. We are doing this process alongside Volkswagen Group that are directly engaging with the German government.

Just to add color on this, in the last IAA conference in Munich, we had a chance to meet the German chancellor who visited the Mobileye booth, and he also participated in a test trial with our ID.Buzz vehicle in Hamburg, and he was very impressed, he made his remarks. It was covered all over the German media, and he made a very interesting remark about how Germany wants to be the leading country in Europe for autonomous driving, that he believes the time is now that our funds by the German government that should be allocated to accelerate this as much as possible. And he was very happy to see the successful collaboration between Volkswagen and Mobileye on this, and it's all covered in the German so there is nothing confidential in what it I just said.

And we think that we have good tailwinds to engage with the German government, specifically, and we plan to launch first in Germany. In Munich, in Hamburg and Berlin are the first 3 cities and we have good support, good alignment working with Volkswagen on that. We think it raises the entry barrier for other competitors when they want to enter the European market. So I think, again, as the first mover advantage by collaborating with an OEM at Volkswagen that has good ties with the German government, which really helps us in this situation.

And then for my follow-up, Amnon, you mentioned just a wrinkle in terms of more OEMs adding ramp to front-facing programs, just curious your thoughts on that. Is it mainly around increasing the data collection? Could it be maybe a precursor of something in terms of future advanced product engagements with those customers as well?

So it's both for data collection, what we call harvesting. We have more OEMs using ramp for harvesting and that's a precursor to also using ramp for hands-free driving. Maybe Nimrod, do you want to add more?

Yes. We recently signed with a new OEM for one of the bigger OEMs in the world that has significant volumes, and that's designed to provide significant volumes for harvesting globally and also use the REM database to improve the performance of the front-facing camera. It is for us today, the strategic value is mostly about expanding the OEM pool that uploads data. This data is important for us not just to use [indiscernible] database explicitly as one of our moats, as we mentioned many times in the past, but also it's -- we use this very elegantly in our AI training and development. And that's something that maybe we'll share more details on in the future. But it is a very significant competitive advantage and expanding by being able to offer much better performance with not significant price increase OEMs of [indiscernible] cameras, getting more data from multiple OEMs in millions. I think today, we have more than 7 million vehicles globally uploading data, more than 2 million vehicles in the U.S. alone. Europe is a similar number, also in Japan and Korea and soon in India and so on. So it's really a good global coverage but it is also diverse in terms of the type of OEM, type of vehicles, a number of OEMs. So we feel very comfortable with the strength of harvesting.

Our next question comes from the line of Dan Levy with Barclays.

I wanted to just follow up on the prior question first and on the driver out, specifically in the U.S. for some of the programs you had, I know you said it was a self-certification process, maybe you can just go into maybe what are some of the gating factors that you need to see from a technical side to get comfortable to actually pull the driver? And what is the expected timing on driver out?

In terms of expected timing, we said it's going to be the first half of 2026 in one city in the U.S. In terms of the technical milestones, now we have a very elaborated safety program. It's called the PGF that we talked about in the past IR meetings and at the CES and we made that public also in terms of an academic paper that we published around it. And also in terms of mileage driven in order to prove to ourselves what is the MTBF. So together with ADMT which is the daughter company of Volkswagen, who was responsible for this program, we have agreed on MTBF milestones per type of accident. So it's not just one number. There are a number of different types of accidents.

What is the MTBF, and we're tracking those numbers. And the trajectory that we see gives us confidence that we can achieve that by the first half of 2026.

Great. I have a follow-up. On Tesla's call last night, [indiscernible] talked quite a bit about efforts with AI, their AI chip and all of the efforts in design and the strength of the performance, really emphasizing this, I think, is a key advantage. So -- and I think you've talked about this on the call here and more broadly. But as it relates to EyeQ6 and eventually EyeQ6, can you maybe just remind us to what extent the engagements with customers are looking at the strength of your SoC design? How that stacks up versus some of the other players? And for those that are maybe SoC agnostic, bring your own silicon -- to what extent you're seeing customers actually leaning into your solution because of the SoC?

Well, I think the first question that you need to ask is why do you need more compute? And Tesla's approach and Mobileye's approach are different. Tesla is relying on a single sensor modality, which is cameras, omni cameras. And therefore, there are this bias variance in machine learning, you want to lower the variants, you need more and more data and more and more compute in order to bring the variance down to a point in which the MTBF is high enough. And the MTBF of Tesla's SSDs, just based on the public record of the FSD tracker, is orders of magnitude away from the bar that you need to pass in order to be unsupervised. Therefore, the requirement of significant more compute and significant more data is very, very intense and hence, the AI fight.

Mobileye's approach relies also on redundancy. So we are doubling down on computer vision, on cameras and bringing the camera processing MTS to be very high, but we are also relying on imaging radars, relying also on a front-facing LADAR for eyes-off. So when you have redundancy, it's a different equation. And this is our PGF framework for creating profusing redundant subsystems.

So then in our context, the question is, if with the EyeQ6 high, we are going to launch eyes-off at a very, very cost-optimized platform, right? They show for [indiscernible] EyeQ6, and it's very, very -- it's highly cost optimized. The drive for EyeQ6 is highly cost optimized, especially compared to platforms of robotaxis of today. So then the question is, why do we need more compute? Right, we replaced the EyeQ6 with EyeQ7. It was not going to reduce for what purpose? Is this to reduce price, not necessarily. So in our view, you need more compute to move from eyes-off to minds-off, and this is something that the industry are not talking about at all. All the targets around eyes-off, what is the bar to reach an eyes-off system. We are thinking of 0.29 and above on mind-off. And mind-off means that you need an AI that has very, very strong see understanding capabilities maybe it can work at lower frame rates. It doesn't have to be, let's say, 10 frames per second. Maybe it can be slower frame rate doesn't have to be doesn't have to replace the safety mechanism. So in our view, the EyeQ7 and EyeQ8 and all the additional compute comes on top of EyeQ6 and does not replace it. So it's a different concept. It's not that we have an EyeQ6 generation, and now we're replacing the EyeQ6 with EyeQ7 generation and then we'll replace EyeQ7 generation with EyeQ8, actually does not make sense when you are thinking about eyes-off systems because the validation process that is required to remove the driver is huge, is huge, right?

So now you did all this validation process and now you are replacing your chip, and you have to do all the validation process again, doesn't make sense. So we have a completely different view of where computers need it, where the added compute is leading.

This will be our last question.

Okay. Our final question is coming from Colin Rusch with Oppenheimer & Company.

I just have a quick one around the cadence of your own chip design. Given some of the tools that we're seeing out there and the potential for accelerated time frames, are you seeing meaningful opportunities in terms of accelerating some of those development time lines and really being able to validate some of these more simplified designs that you guys are talking about?

Our chip portfolio covers both the very, very low end, like EyeQ6 Light and the very high end. So EyeQ6 High, which is in production, is equivalent to, say, [indiscernible] in terms of looking at running programs like convolutional net ResNet, Vision transformers, and the like. And EyeQ7 would compare to [indiscernible] in terms of its strength. And EyeQ8 is going to be -- which is now in the design stages and will be ready for 2029 production, is going to be 3x, 4x stronger. And again, the EyeQ7 and EyeQ8 are responsible for the minds-off. It's not -- for the eyes-off we are set. We have the EyeQ6. We don't need to replace EyeQ6 with a more powerful chip to reach eyes-off capability or to reach a robotaxi -- robotaxi with teleoperators in the back office.

The idea in robotaxis to remove the teleoperators. This is why we want minds-off and the idea with consumer operated cars is to enable the driver to sleep while using the system. This is where we need more compute. And the cadence is once every 2 years. So this is the cadence and which is sufficient for the speed of where the industry is going and where technology is going.

Colin, thanks for the question. I don't think we have time for a follow-up. I want to respect everyone's time.

This now concludes our question-and-answer session. I would like to turn the floor back over to management, Dan Galves for closing comments.

Thanks, everyone, and looking forward to the Q4 call in January. Thank you. Have a good day.

Ladies and gentlemen, thank you for your participation. This does conclude today's teleconference. You may disconnect your lines, and have a wonderful day.

Greetings, and welcome to the Mobileye Second Quarter 2025 Earnings Conference Call. [Operator Instructions] As a reminder, this conference is being recorded. It is now my pleasure to introduce your host, Daniel Galves, Chief Communications Officer. Thank you. Mr. Galves, you may begin.

Thanks, Maria. Hello, everyone, and welcome to Mobileye's Second Quarter 2025 Earnings Conference Call for the period ending June 28, 2025. Please note that today's discussion contains forward-looking statements based on the business environment as we currently see it. Such statements involve risks and uncertainties. Please refer to the accompanying press release, which includes additional information on the specific factors that could cause actual results to differ materially. Additionally, on this call, we will refer to both GAAP and non-GAAP figures.

A reconciliation of GAAP to non-GAAP financial measures is provided in our posted earnings release. Joining us on the call today are Professor Amnon Shashua, Mobileye's CEO and President; and Nimrod Nehushtan, Mobileye's EVP of Business Development and Strategy. Unfortunately, our CFO, Moran Shemesh recently experienced a death in her family and will not be joining the call today. I'm sure everyone listening joins me in wishing the best to Moran and her family. For today's earnings call, I will essentially take on Moran's role. Thanks. And now I'll turn the call over to Amnon.

Hello, everyone, and thanks for joining our earnings call. Starting with the results. Q2 revenue was up 15% year-over-year as demand for the IT was strong across regions and OEMs. Adjusted operating income was up 34% and adjusted operating margin rose 3 points to 21%. Q2 was a good display of the strong operating leverage created by our business model. On a year-over-year basis, more than 40% of the revenue growth converted to operating income compared to Q1, nearly 70% of the higher revenue dropped to operating income. Operating cash flow was again a highlight over $200 million for the quarter and over $300 million for the first half, about 33% of revenue.

Our ADAS business is highly cash generative, and we are maintaining strong working capital discipline. The core ADAS business is performing well, with volumes at or above $8.5 million per quarter for the last 4 periods, and we are raising our full year revenue outlook by 4% and our adjusted operating income outlook by 14% at the midpoint. Our core ADAS business truly illustrates that mobilizes an execution machine. EyeQ light will be the future high-volume chip for this segment and the ramp-up of that new system has been seamless. Only 1 year after the first SOP, we already have EyeQ6 light-based systems on the road in North America, Europe, China, Japan and India.

On the advanced product side, we are the only OEM neutral platform that is cost-efficient and scalable and has a credible technology path to eyes of autonomy in both privately owned vehicles and robotaxis. All 4 of our advanced products surround ADAS, supervision, Chauffeur and drive share common elements, including the EyeQ6 high inference chip, major portions of the perception and policy AI stack, REM crowdsourced driving intelligence, our safety frameworks and the company's data and validation infrastructure. This common backbone creates many synergies for us and our customers. enables us to develop and execute all 4 solutions simultaneously and leaves us agnostic to whether the market moves faster in one way or another, whereas a couple of years ago, OEMs were primarily focused on supervision. We are now seeing broad momentum across our portfolio from next-generation ADAS to full pojnt A to point B eyes on hands-free to Level 3 systems and to robotaxi.

The EyeQ6 high-based surround ADAS system continues to develop as the next generation of standardized riding assist on high-volume vehicle platforms. This system addresses multiple objectives in a cost-efficient package. It's designed to meet stricter late-decade safety standards, enabled highway hands-free performance for a lower cost than current systems and supports OEM goals to consolidate ECUs and to integrate technology on a single SoC. In recent months.

We have seen growing demand from OEMs to shift away from already sourced single camera programs toward our multi-cameras around ADAS Mando. Overall, opportunities to substantially grow content per vehicle in the ADAS space have improved over the last 6 months. Supervision activity remains robust, but lack of competitive pressure is enabling OEMs to continue to take their time with decision-making. Meanwhile, Chauffeur has generated multiple new OEM prospects that sees eyes off on highway as a breakthrough feature that allows drivers to reclaim their time during commutes.

The central question around eyes off consumer AV programs is simply technological maturity. How likely is it for a technology provider like Mobileye to execute a system with human label safety and an expansive ODD. In this context, our 4 production programs with Volkswagen Group are significant strategic assets. They showcased our rapid progress in transforming our core technologies into scalable products. our ability to demonstrate these products to other customers, including production level hardware and software associated KPIs is an important proof point that our competitors do not have and will drive increasing competitive pressure as we approach launch.

Turning to robotaxi. Waymo's achievement of 25% market share in San Francisco despite offering now time or cost advantage of our human-driven alternatives is very encouraging and has reignited industry enthusiasm. When you look at the robotaxi opportunity, 2 pillars are critical, safety and scalability. On safety, this means reaching the mean time between failures that exceeds human statistics as well as other critical safety standards. Safety has long been a strength of Mobileye, supported by foundational innovations like our RSS model, which we developed in 2017 and the PGF, our recently published framework for using multiple subsystems.

Once safety goals have been reached, the second pillar is scalability. The name of the game here is how fast can you scale. This should be evaluated across 3 different vectors. The first scalability vector is geographic. How fast can you expand from city to city. REM is a huge asset here. The second is cost. What are the all-in operating cost of the system. Our in-house design compute, imaging radar, efficient AI supply chain synergies. These all combined to create significant cost efficiency advantages relative to the competition.

Finally, scalability also entails production capacity and business model. The fact that we work in partnership with OEMs that produce vehicles where our system is integrated during the mass production line rather than being uplifted in a different facility after the vehicle has been produced is very important. This approach allows us to be capital light, but it's not just about capital light. It also allows us to scale fast.

Even if we had all the capital to go and purchase 100,000 vehicles and then build production plans that would uplift the self-driving technology, it would have slowed us down. We are working with Volkswagen, of course, but also Holland, which has a production facility underway and have advanced engagements with other high-scale OEMs. On the operations and distribution side, we have arrangements with Volkswagen's Mobility arm MOIA and Japanese fleet manager Marubeni, to provide operations and the customer-facing technology.

Finally, we have announced reengagement with demand generators, Uber and Lyft in the U.S. and public transport operators in Europe to provide the demand platform. All of these actors have skin in the game, which is also important to drive scale. So Mobileye with the kind of partnerships that we are building is in a very good position to scale rapidly once we start commercial deployment in 2026. In terms of a technology update on robotaxi, we recently successfully transitioned into our full production hardware inside the ID buzz test vehicle. The mean time between failure performance is tracking well to the KPIs that were laid out at the start of the program. We expect to reach our KPI goals by the end of 2025, then start adding tele operations and then remove the driver in 2026. So it's all on track.

In summary, the opportunities set in front of us today is larger, broader, deeper and more urgent than it was when we went public in 2022. OEMs are indicating increased clarity in planning and decision-making. -- near-term volumes are strong. The demand for both higher performance and lower cost is intensifying, and eyes of performance, whether for personal cars or robotaxi is no longer seen as a science experiment but as an achievable and commercially viable product.

This is exactly where Mobileye thrives. I'll turn the call over to Dan to cover the finance section.

Thank you, Amnon. Before I begin, please be aware that all my comments on profitability will refer to non-GAAP measurements. The primary exclusion in mobilized non-GAAP numbers is amortization of intangible assets, which is mainly related to Intel's acquisition of Mobileye in 2017. We also exclude stock-based compensation. Our Q2 results significantly exceeded the color we provided on the Q1 2025 earnings call in April and were slightly better than our pre-release numbers from earlier this month.

Revenue was up 15% year-over-year versus the outlook of plus 7%. The strength was due to several factors. Multiple OEMs, including China OEMs, showed modest outperformance, which taken together, contributed to significant overall gains. Supervision volume was also a bit stronger than expected as production of the vehicles we are on is running better than expected year-to-date.

I'll spend a minute on inventory as we continue to monitor it closely. Based on our discussions with customers, inventory was relatively tight entering the year, and there was some direction from certain OEMs to increase safety stocks due to the volatile macro environment. Even so, a variety of analyses we run on a regular basis indicates that shipments were relatively consistent with demand on a year-to-date basis. To frame it, EyeQ volumes in the second half of 2024 were 17.8 million units and inventory ended the year at a low level. Volumes in the first half of 2025 were $18.1 million in what we believe was a comparable demand environment to the second half of last year. We continue to believe that inventory at our customers remains well aligned with underlying demand.

Turning to gross margin. It was down slightly year-over-year and versus Q1. Gross margins are stable by product and by region. The exact results, however, depend on mix of China volumes in the ADAS business and supervision. Each of those segments carries gross margins somewhat lower than the corporate average. Supervision in particular, was a higher percentage of revenue in Q2 versus Q1 causing a bit of a gross margin reduction.

Operating expenses were up 7% year-over-year and flat compared to Q1 versus the prior outlook that indicated Q2 OpEx would be slightly higher than Q1. As Amnon mentioned, operating cash flow was over $300 million in the first half. This is primarily due to strong cash flow from the core business, However, we've also managed tight control over the working capital accounts, particularly balance sheet inventory, which came down by about $90 million in the first half. We're now well aligned with our 6-month target on balance sheet inventory, and we expect working capital to be more cash neutral in the back half.

Turning to the full year guidance. We are increasing the revenue midpoint by 4% and the adjusted operating income midpoint by 14%. On the last call, we noted that the implied step-down in second half 2025 revenue versus the first half did not reflect any specific indication of production weakness but rather a cautious stance given the elevated uncertainty around automotive tariffs at the time. Since then, while tariffs remain in place, the actual impact on production and consumer demand appears rather limited and third-party forecasts have risen since April.

Our current outlook releases some of the conservatism in Q3 as visibility is high at this point. That said, visibility into Q4 remains more limited as is always the case in July. and we believe it's prudent to maintain a cautious stance and a wider-than-usual range for that Q4 period. To be 100% clear, the business is performing very well. We are not seeing any tangible headwinds and we've not received any indications from customers that Q4 volumes will weaken. We are simply choosing to remain conservative beyond the very near term.

Our full year outlook is based on EyeQ volumes of $33.5 million to $35.5 million, up from $32 million to $34 million previously. As noted above, supervision volumes are running better than expected, and we're modestly raising the outlook to about 40,000 units at the midpoint versus the prior outlook in the low 20,000. We expect gross margins to be up about 0.5 point year-over-year in 2025. This is slightly worse than our prior outlook, but this is simply due to supervision and China EyeQ being a bit of a higher percentage of revenue.

Adjusted operating expenses do not typically flex according to revenue and remain in line with our prior expectations. We continue to expect an increase of about 7% year-over-year to slightly below $1 billion. Looking at the balance of the year, we would expect Q3 to be somewhat higher than Q4, consistent with historical seasonality.

Turning to Q3. We expect to deliver approximately $8.7 million to 9.3 million EyeQ units and for our revenue to be roughly flat on a year-over-year basis. We expect gross margins to be slightly below the Q2 levels and for operating expenses to be seasonally higher in Q3 versus Q2, aligned with previous expectations.

Thank you, and we will now take your questions.

[Operator Instructions] Our first question comes from Chris McNally with Evercore ISI.

Maybe we could just double click Amnon your comment around sort of the higher momentum at Chauffeur, maybe a little bit of slow momentum on supervision decision-making. How much do you think this is sort of OEMs having more of a question around their own pricing ability to pass through sort of a Level 2 plus product versus something else because I think we've all seen this sort of delay in implementation, and there is some fear that we're seeing these products given away almost for free in China, a lack of clarity, let's say, for how OEMs would price such a product? I would love your thoughts on that.

I think there is lack of competitive pressure for these systems in Europe and the U.S. You see these systems a lot in China. And in the -- outside of China, it's only the Tesla FSD, and the OEMs have seen the test FSD for more than a decade. So we need more competitive pressure to kind of bring OEMs to a sense of urgency. I think the last news about penetration rates of Tesla FSD are encouraging. It's more than 25% take rate. and it looks like it's climbing. So I think the news are good in terms of public interest in these kinds of features and willing to pay for them.

But regardless, OEMs are still in planning stage because it's not only the Level 2 plus, the supervision, there is a Chauffeur. They want to be part. They want to have skin in the game in robotaxi, not just produced cars. and just sell them to the likes of Waymo and others. They want skin in the game in the robotaxi domain. So it's all part of planning. There is around ADAS, whether they should -- it should take over the front-facing camera or just be a premium product. There's a lot of planning to do. But the more we deep dive into it. I think that planning phase is coming to a close. So we see a lot of activity by OEMs talking about supervision, but in addition, also surround ADAS and Chauffeur and with a number of high scalers OEMs also about robotaxi.

If I may add to this I just may add 1 comment. We recently started inviting OEMs to see our Generation 2 supervision system. -- which is now operational in various locations and shows our EyeQ6 platform with new technologies, and we've seen increased interest and pretty much a lot of excitement by OEMs to see the demonstrations and it's kind of -- it's another positive momentum around supervision. So it's not just the competitive pressure is also seeing kind of more evidence to our Generation 2 system and how it performs in the field that now is available, and it's been some in the past few days and so far, it's been very successful.

That's really helpful. And just my quick just follow-up. Is it fair to characterize or power phrase as sort of the flag slide that you showed at December is more of a an implementation delay rather than a full pause on supervision and that you still see supervision as essentially the stepping stone for a lot of these OEM programs into Chauffeur given the software overlap and just obviously additional hardware needed for Chauffeur.

Nimrod start, and I'll complement if necessary.

No, I don't think that necessarily we have suggested that supervision is a prerequisite for Chauffeur. I think that what remains true is that there is a consensus at least from our perspective, amongst OEMs that Chauffeur is a very compelling value proposition for consumers. And as Amnon said in his opening comments, it's a question of whether or not the technology is mature and at what price and which time line. And we are making consistent progress not just in Chauffeur directly but also through robotaxi, which is showing a lot more about our robustness and the maturity of our technologies for ISO for no driver, which requires very high precision levels. And the more we're making progress, the more we are convincing OEMs that this is a technology that is for here and now and not for the next 5 years.

And therefore, we see some OEMs that are considering going straight to Chauffeur. For the, let's say, 2027, '28 time frames. So I think what we have learned is that the OEMs are a spectrum of needs and interests and planning strategies. And our strategy is that our products are playing on the complete spectrum of solutions. And so we can offer the entire product portfolio like we with Volkswagen. We can offer parts of it. But what's important is that we're progressing towards SOPs towards launching these products in the market, regardless of how OEMs are kind of thinking about their planning. And the more we make progress, the more we can convince them that the technologies are mature, which product makes the most sense for their segments and so on.

Yes, I'll mention that we have a start of production 2027 with Audi on Chauffeur and it's on track, and there's also a number of homologation steps that also we have passed. So as time goes by, the maturity level of this system is now becoming more and more evident and that should bring OEMs to the table and get convinced that the maturity level is good enough to start thinking about the production program for Chauffeur.

Our next question comes from James Picariello with BNP Paribas.

Just starting with supervision, the guide for 40,000 units, a near doubling of the expectation for the full year. Can you just speak to what's driving that, how the relationship is trending with ZEEKR? And then just looking ahead, any thoughts on the timing for next year concerning the portionalitylaunches for supervision and Chauffeur.

Yes, we took a conservative stance on supervision volumes for this year. Since then, what we've seen is ZEEKR 009 for export markets has been selling more vehicles than we probably expected. Polestar 4 production and end demand has been pretty good as well. I think key here is that any ZEEKR vehicles that are being shipped outside of China are still using the supervision system, which kind of indicates the maturity of our system for kind of non-China markets. But yes, I think it's just a reflection of kind of conservative start of the year and kind of production of these vehicles running better than expected.

Yes. As for the portion Audi, the standard production is the end of 2026. So the effect on revenue should be seen in 2027. We see 2027 as really an inflection year in terms of revenue, where supervision by Porsche and Audi, and we believe more would come out robotaxi will start generating revenue as well because we are removing the driver mid of 2026, and we have a very strong plan of scalability. So in 2027 is really the inflection year in terms of revenue.

Yes. And if we look at the kind of the consensus expectations for Supervision in '26, it can be almost exclusively covered by the current vehicles in production.

Got it. That's helpful. And just my follow-up. In regard to the recent secondary offering tied in tail stake, how should we think about any future intentions there and the potential time frame?

Dan, do you want to take this and I'll complement?

Yes. No, I think Intel shown quite a bit of patience with their stake in Mobileye. They hadn't sold any shares for 2 years. They still maintain more than 80% ownership. I think they've made public comments that they have kind of a very strong view of the kind of the potential of Mobileye and want to participate in that upside. So we weren't really surprised that they'd want to sell some shares after the next couple of years, but we can't really speak to any future plans that they have.

Our next question comes from George Gianarikas with Canaccord Genuity.

I'd like to concentrate a little bit on robotaxi. I think you sort of characterized the interest as accelerating from OEMs and deploying your solution. Can you just help us understand a little bit about what you're seeing in the marketplace, the potential for new wins? And what the competitive set looks like when you're offering your solution to OEMs?

Well, we have a relationship with the Volkswagen on the ID buzz where MOIA is operator and customer facing. There's also deals with the Uber regarding this platform. The volume expectation towards the end of the -- until the end of the decade is very substantial. There is a hold on with a platform called Mover. We already have prototypes equipped with our system and testing. It should come out 6 months later, also volume productions projections are very high.

In addition, we have relationship with Marubeni. We are working with additional OEMs to supply vehicles both from MOIA and also for Marubeni and hopefully, we'll be able to update the market soon about additional OEMs. But Volkswagen alone is a very high volume opportunity for robotaxis.

And if I may add to this -- sorry. I just wanted to add maybe a little bit more color on what we're seeing in the market and the competitive environment. I think that there is a -- we need to distinguish between the U.S. and Europe in that regard, which are our 2 primary markets for the first launches. In the U.S., of course, there is Waymo and Tesla that has been making statements about this. Beyond these 2 as a technology provider that can provide the full self-driving system, which includes the hardware, the software, AI technologies and so on in a scalable way in a cost-efficient way that we'll leave enough room for all players involved to generate a profit.

We're seeing Mobileye as kind of a unique company at this stage. So Waymo and Tesla, of course, have their own business model being vertically integrated at this stage. And for the OEMs that want to basically build a business of producing robotaxis in a serious production fashion and then sign a business model with the demand generators. We're the primary, if not the only candidate at this stage at least from what we're seeing. And in Europe, we are -- I think that we are in the pole position in the way. And just recently, the German chancellor took a test drive with the ID buzz vehicles mobilize technology in Germany, which is kind of putting a lot more public attention and some -- let's say, political attention into enabling robotaxis in Europe.

In that sense, being partnering with Volkswagen is hugely beneficial for our interest.

Okay. Just as a follow-up, can you just -- on the robotaxi does help us maybe understand a little bit more about the business model opportunity, the price per system and also particularly the potential for you to participate in the revenue per mile as you deploy these systems and if that can be replicated across OEMs.

Yes. So we receive revenue for the system and we received also recurring revenue or the cost per mile. We have both. Maybe in the future, we could reduce the cost of the system and add more in terms of the contribution of per mile, but even the current setup is very good in terms of the revenue potential, the recurring revenue potential over time.

Our next question comes from Dan Levy with Barclays.

Wanted to just first start with a question on the near-term EyeQ shipments. And maybe you could just give a bit more color on where the strength is coming in from and specifically, the trends within China, which had obviously been quite weak in 2H. But it seems like the last couple of quarters has been pretty good. What's the right run rate to think of now from China, both from the domestics and from the multinationals there?

I can start on that. So I mean, I think from a kind of from an overall comment, it was difficult to analyze the EyeQ volume growth the last year or so because of some disruptions on inventory in China, now you're starting to really be able to analyze it. So in Q2, if we adjust for inventory digestion last year, volume grew around 13% year-over-year in Q2 for EyeQ volume. And our top 10 customers were minus 3%. So significant growth over market. If you look at the Q3 outlook, it's for growth around 5% year-over-year. Our top 10 customers are minus 2. So this kind of comparison to our top 10 customers, it's starting to show up as kind of very favorable for us.

On China, the China business has been running better. We did slightly over 1 million units in the back half of last year. We did -- we thought we would do around 1 million in the first half. That was the outlook. We did more like 1.5 million. So there was some upside there. We're not assuming that type of volume for the back half just because we don't have as much visibility and we want to stay conservative. but it does look like that's a fairly stable run rate for us.

But yes, I think overall, the revenue outperformance has been pretty broad-based. If you look at kind of all of our top 10 customers, for most of them, there was at least a bit of outperformance. It added up to a bigger number. There was outperformance in China, there was outperformance in supervision. So it's all pretty broad-based.

Okay. Great. The second question is, as you're ramping on your efforts in Drive, I wanted to get a sense of the type of resource allocation. And I go back to the CMD you had last year where I think you gave the pie chart of your spend that 11% of your spend is on Drive. It seems like your efforts are accelerating here. Can you just give us a sense of how extensive the resource requirement is on Drive and what this could do on the OpEx in the next couple of years?

Well, our OpEx grew substantially in 2023. also grew in 2024. We see the OpEx as more or less flattish in the coming years. That means all the growth to prepare for Drive and Chauffeur and supervision to the transition from Tier 2 to Tier 1 on some of the programs like with the Porsche and Audi. All of that account for the growth that we have already experienced. So we don't see a substantial growth in the near future in terms of OpEx growth.

Our next question comes from Samik Chatterjee with JPMorgan Chase.

This is MP on for Samik Chatterjee. So I just wanted to double click on the Imaging RADAR deal, which you did during the quarter. And like how should we think about the size of that particular business? And like will you be open to doing more similar deals in the future where you will be selling individual components other than full systems? And I have a follow-up.

Yes. So the imaging radar for us is a strategic sensor. The deal we had with that particular OEM is just for the sensor. It's a very reputable OEM, and we thought that this would drive credibility because the RFQ phase was very lengthy and all competitors of imaging radars participated and our radar was shining through. So we sold it as a separate sensor, but we do not expect to do that in the future. It's part of the bundle of eyes off systems on Chauffeur and Drive, for example, the ID bus has 5 of our imaging radars as a front-facing imaging grade and corner imaging radars. And we believe that all future Chauffeur programs will have our imaging radar because it allows you to get the speed that you need in terms of highway driving. You need to see hazards very far away, more than 150 meters away and the sensor that we have can do that in a very high resolution and high dynamic range and it's simply an enabler for eyes off systems at scale.

So it's part of a bundle. We don't see it as another source of business as a sensor business.

Okay. Got it. And another question which I had was regarding the 2027 ramp. So you will be ramping on supervision Chauffeur and Drive in that year. Any way to understand like which will be the biggest driver of those? And how will you rank out of those opportunities in '27?

So we have the 2027, the Chauffeur and supervision. We mentioned in the past as more than 19 car models coming out with those systems. We're not yet ready to make a guidance for 2027. In terms of Drive there is a significant plan of expansion to multiple cities starting from end of 2026, both in Europe and in the U.S. So it should drive substantial growth. We're not at a position to put a dollar number to it right now.

Exactly. But I think what's new here is that we do now expect Drive to be a significant contributor in 2027, and that's a reflection of the confidence we have in commercial deployment during -- sometime during 2026..

Our next question comes from Vijay Rakesh with Mizuho Securities.

Just a quick question on supervision. Obviously, a nice upside here. You raised 40,000 from ZEEKR Europe. And sort of how Calendar '26 should shape up in terms of units for supervision, especially some of the newer ramps, and I have a follow-up.

We're not really ready to talk about specific expectations for 2026. Like I said, we're essentially kind of marking to market the end production of the vehicles that have supervision today. This still doesn't include any U.S. volume for Polestar 4. They did start producing that vehicle in Korea. So there's a tariff in Korea from vehicles produced in Korea, but it's not 100% like it is from China. So they should be able to launch in the U.S., and we think that, that will create some growth for next year as well. The export volume of ZEEKR has been probably a little bit better than expected this year. We would expect that to grow a bit next year as well. So it should be some growth in '26 from the existing vehicles, and we'll have more to say about kind of the overall supervision volumes like when we get to 2026.

Got it. And then on the -- just a quick housekeeping question on the inventory side. I know EyeQ, you raised from like midpoint of $33 million to like $34 million here. So definitely seeing some improvement. But if you were to look at the inventory level, just to -- I know it's tough because every OEM has a different inventory level. But if you normalize that, how does that inventory level compare now towards last quarter or last year, just to get some idea of where the levels are.

Do you want to start Nimrod?

Yes. So I don't think that we can disclose like the inventory levels that the OEMs are keeping themselves as like the safety stock. But in general, we've been in line with what is a -- we can consider modest compared to historic periods. So the way we are kind of analyzing this is in multiple ways, we have direct information coming from our Tier 1 customers. that they get direct information from their OEMs. We also cross-reference this with third-party analysis on the vehicle -- the overall industry vehicle production compared to our sales. And so we keep a very close track of this. And we are -- we'll keep our eyes on this on a weekly basis.

Yes. No, I think that's right. We -- the finance and sales teams have done a great job of kind of developing tools to as well as kind of direct feedback from the Tier 1s and everything looks like it was pretty flattish from the end of 2024 until now.

Our next question comes from Adam Jonas with Morgan Stanley.

Thanks, everybody. So Amnon I'm just looking at your CapEx here, $28 million for the first half of the year. If I annualize that, that's obviously down substantially year-on-year. But even if it's flat, and I think consensus has you guys spending around $100 million, maybe $110 million this year. Your CapEx is basically not moved. It's declined over a number of years, and it really makes Mobileye stand out as for a physical AI hard tech company really in the thick of so many exciting programs, collecting data, growing the fleet, how do you do it? Like where is your CapEx spending on compute? How much compute do you have? Because it would strike me that your compute needs and therefore, your AI CapEx needs would somehow scale at least proportionate to the amount of data that you're feeding into your simulation and data centers. And so tell me where I'm wrong there. Why -- how are you able to do it? Or is your message, we just don't need compute like others and that guys like Elon and Jensen are [indiscernible]? And then I have a follow-up.

Well, we need compute. We have compute both on-prem and also on the cloud. Our cloud spending has slightly reduced in -- and I cannot reveal those numbers, but it's in the tens of millions, a large number of tens of millions. And in favor of on-prem, more GPUs. But we have a different philosophy on how to spend money on compute than what you hear from our competitors. And -- we have very good systems, very good performance. Our EyeQ6 generation, our generation 2 supervision and Chauffeur and Drive is really top-notch. If you look at our drive vehicle, there are more than 100 ID. Buzz. There has been a lot a lot of demonstrations to journalists, both in Europe and also in the U.S., as [indiscernible] mentioned just last week, the Chancellor of Germany drove the ID. Buzz performance is very, very good. And we know how to train the models in a way that is more efficient.

Okay. Appreciate that. As a follow-up, what is your simulation stack? What does it look like? How much synthetic data are you using to reduce costs for training on edge cases because that also seems to be for the problem that you're solving, and we talked about humanoid, but even in autonomous cars, very, very important. Love to hear your views there.

Yes, it's a very good question. When we think about simulation, there are 2 types of simulation. There is a photorealistic simulation. We used photorealistic simulation in order to simulate age cases, for example, putting a car on the road.

Amnon, you went on mute.

Just when it was getting good.

Were trying to reconnect them. I think you dropped out. So we just need to reconnect him.

There are 2 types of simulation. One is photorealistic in order to emulate edge cases. Say you have a cart falling off from a truck on the road, right? These things -- you don't need to wait until you find them in the physical world. You can put them in a photorealistic simulator, and we have very powerful photorealistic simulators. Another type of simulation is to simulate the driving policy. This is a piece of technology. Maybe we'll talk about it more at next year's CS that we developed, it's kind of, we call it ACI, artificial community intelligence, where we have a simulator, not a photorealistic, a synthetic simulator simulating hundreds of road users over billions of miles of simulation. We run billions of miles overnight. And we use that in order to train the driving policy. So that amount of compute that you need there is way below the amount of compute if you would train it on photo realistic simulations. And it's much more efficient. So those are the 2 types of simulations we use.

Our next question comes from Shreyas Patil with Wolfe Research.

Maybe could you guys talk about the typical lead time between securing awards and surround ADAS and launching programs? I believe it's typically 2 to 3 years. So given the timing of the new ADAS standards in Europe, which I think is 2028, that would suggest OEMs need to secure contracts in the next 12 to 18 months. So is that the kind of time line we should be thinking about in terms of potential awards?

Yes. So when we're talking about Western OEMs, a time line is typical 2 years -- 2 to 2.5 years from nomination to startup production.

Okay. That's helpful. So if the standards are coming on in 2028, it would suggest they would be needed to secure these awards in 2026, something like that?

Yes. Nimrod, do you want to add?

Yes. I think the majority of the RFQs that we have and we have RFQs with multiple OEMs, and the majority of our customers are engaging with us in this solution. These are things aimed for '27, '28 SOPs. So that's the current plans that we're seeing.

Okay. And then on robotaxis, there are a large number of AV developers in the space and some of the rideshare operators such as Uber are striking agreements with multiple players for their platforms. So curious how you gain confidence in the number of vehicles that Mobileye will be supporting either on an Uber or a Lyft type of platform.

Nimrod, do you want to take this?

Yes. I think that First of all, it makes sense for companies like Uber, who face pressure and questions from investors about their strategy for robotaxis as a potential threat for their business to maximize their chances of being one of the winners in this space. I think that we're at the beginning of the adoption curve. And longer term, we believe that winning solutions will be the most cost efficient, geographically scalable with the highest performance, highest availability rates and we believe that our products are inherently with the pole position in the axis. And so today, there might be some announcements and statements with pretty much everything that can be a potential contender but we think that within not a lot of time, it will -- there will be a separation between a very selected few companies that will have advantages in these economic scalability, geographic callability, the availability of the service and the others.

Because if you think about this, we're still not at the stage of even thinking about, for example, how many charges they need to do per day for the vehicle. It doesn't play any factor. And our system is roughly 20% in power consumption compared to Waymo's. So these are just small things that today don't play a role because it's still a question of can you do it or can't you do it? We're at the cusp of getting to how well can you do it, how efficiently can you do it? And our system is designed to be excelling in these parameters. So that's where we get the confidence that ultimately, we will be 1 of these 2, 3 companies that will see the highest volume of robotaxi services.

Our next question comes from Joe Spak with UBS.

Maybe just sticking on Drive and robotaxi line. If you could you give us some things here on sort of what commercial deployment looks like you have to finalize the vehicle, then there's [indiscernible] and there's the remove the driver in '26. So I was wondering if you could add any more color in terms of maybe where you first see that happening, in U.S. or Europe? And then -- and also like maybe how much input do you really have here in terms of things such as the size of the geofence, the number of vehicles, like when the drivers actually moved? Like what -- how does that relationship with your partners work.

The leading partner is Volkswagen ADMT division. And we have a very tight cooperation. We work very well together. The driver would be removed in the first city, it will be in the U.S., I'm not at liberty to say the name of the city. but it's -- there is very concrete plans in terms of how the driver would be removed, the design of the tele operators. We have a very unique design of tele operators that allows for scale. So going from, say, 1 teleoperator per vehicle to quickly going to 1 to X, 1 operator for X vehicles to scale that very fast using technology, certain cloud computing technology that will enable us to scale it. And it'll start in middle 2026 with that first city in the U.S.

Okay. And then the second question is, there was a report this morning that Volkswagen is looking for capital at their autonomous unit and offering a minority stake in the subsidiary that they're searching for strategic or financial investors. Like, I'm not asking you to comment specifically on that potential offering. But is a strategic investment in a partner something Mobileye would consider here as you look to scale drive.

Yes. I think it's a very good development. I think also, Google did that for Waymo, even though Google has deep pockets and confront Waymo without any external funding. I think it's a very good development we supported, and we will seriously consider also participating as an investor.

Maria, this will be our last question, the next one coming up.

Okay. Our next question -- our last question then is from Colin Rusch with Oppenheimer.

Given the leverage that you're seeing off of the compound platform. Can you talk a little bit about the cadence of learning that you're seeing, put some metrics around it, potentially talk about the reduction in hallucinations that you're seeing in the system at this point?

Yes. We don't have hallucinations. Hallucinations is a metric for large language models. What -- our KPI is mid time between failure. and that is very important in Drive because that's the only way to remove the driver that you reach an which corresponds to very strict KPIs, and we are on track. All the indications are that by the end of this year will be at the [indiscernible] that will enable to remove the driver. And then for the next 6 months until we actually remove the driver, would be working on the teleoperation technology and then start removing the driver. But all our KPIs for MTBF and other safety measures are all on track.

And just the last one here is around potential for reduction on cost of the perception suite. As you look at not only your own internal reduction in cost, but sourcing other elements. Can you talk about how quickly you can start driving some cost out of the system as you get into '27, '28 and start seeing some incremental volumes ramp up?

Well, the cost of our system, we are talking about Drive. The cost of our system is already very lean. We have cameras, which doesn't cost much. We have our ECO with 4 EyeQ6 high, it doesn't cost much. We have imaging radars, which we produce, it's hundreds of dollars overall. We have LADARs that are supplied by Innovis, also very reasonable cost. If you look towards the end of the decade, there is a possibility with just having 2 layers of redundancy cameras and the imaging radars and reducing the number of LADARs or reducing LADARs altogether. But this is something that's too early to say. That could be another cost reduction. Another cost reduction towards the end of the decade is moving from EyeQ6 to EyeQ7 -- that will be another element of cost reduction, but it's not really a very meaningful cost reduction. But we are already starting with a very lean cost platform.

We have reached the end of our question-and-answer session, and I would now like to turn the floor back over to Mr. Galves for closing comments.

Thanks, Maria, and thanks for everyone -- thanks to everyone for joining the call. We will see you again at the Q3 earnings call in October. Thank you very much.

This concludes today's teleconference. You may disconnect your lines at this time. Thank you for your participation.

Great. Welcome back. Next up, we have a company I think everybody is quite familiar with Mobileye. And we have Dan Galves joining us, Chief Communications Officer. We also have former DB alum. You made it.

That's right. It's good to be back.

Welcome back to the mothership.

Hope they let me out of the building.

So a lot going on in the industry, we'll spend very little time in video on tariffs. So I don't know if that's a good thing or a bad thing. For those of you who do not normalize the leading supplier of advanced ADAS autonomy solutions, acting both as a Tier 1 and Tier 2.

And maybe we can start off in order of what I like to think of technical complexity, most complex being robotaxi. You now -- Mobileye now has a partnership actually with Uber, Lyft, [Marubeni], VW. I think there's a couple of more I missed in there, but I think people get the point, you have quite a few potboiling. What's the ultimate goal here with having all these partnerships. And if we think about that by region and by time line, how do you see this playing out?

So I think, first of all, like one of the big decisions Mobileye made quite a while ago was that we don't really know how the AV world is going to play out, whether AVs are going to be deployed on privately owned vehicles or in fleets of robotaxis.

So we made the decision that we wanted to be agnostic and be able to win in either scenario, which means that you have to have kind of a very cost-efficient system that could fit into the cost of a privately owned vehicle. It means the vehicle needs to be able to kind of scale geographically quickly because you can't really sell a car to somebody and say, "Okay, you can only drive in California or you can only drive in this state. And so I think that, that's something that's unique about Mobileye is we're taking kind of a very sort of efficiency based approach to the technology.

Now what we've seen in the last 1.5 years since Waymo kind of was successfully removed the safety driver from their vehicles is some really positive consumer acceptance data points, like they're doing more than 25% market share of the ride-hail business in San Francisco. Today, despite the fact that they're not offering lower pricing, they're not offering faster routes. In fact, we think like maybe they're at a disadvantage to human-driven ride-hailing at the moment, but it's showing that consumers are kind of really accepting this technology in a lot of cases, preferring it.

So this is -- has generated like a ton of interest from a variety of players in the industry. What we're trying to do in terms of our go-to-market is put together vehicle producers that can produce vehicles with quality and at scale, put them together with demand generators that have the user bases that would -- are going to be the ones to utilize these vehicles.

So our main kind of activity right now is with VW Commercial Vehicles. Their strategy is essentially to own and operate the vehicles and kind of insert them into demand-generating networks. In the U.S., that's mostly kind of Uber and Lyft. In Europe, it's quite different. In Europe, they have a massive kind of public transportation infrastructure with a lot of issues, like a lot of like large buses route -- large buses on routes that are underutilized, major shortage of drivers. So there are cities and municipalities around Europe that are extremely interested in creating more efficient transportation.

And we think [AV] is the best way to do that. So VW would be the owner operator of the vehicle. They would kind of insert those vehicles into demand generation networks and essentially like that car would become the driver. So Uber or Lyft or the public transit agency generates their revenue from the ride. They pay the driver, which would be VW and then VW we've negotiated with them to get a kind of a piece of the revenue per mile. So somewhere around kind of $0.20 per kilometer.

So our business model would be to sell the self-driving system, including software and hardware to the automaker. We would generate a good gross margin on that sale. But I think the main kind of revenue potential is around the operating revenue per mile that we would have a piece of.

And I think the demand generators are really excited about it as well because, number one, I think it's probably a threat. But then there's also a lot of opportunity that if the price of transportation goes down, you could start generating much more activity than the human-driven ride-hail is today.

And I think that at our cost point, there's good potential to kind of reduce that to undercut that human-driven ride-hail cost. So yes, we're excited about it. Like our main milestone is we want to remove the safety driver by the middle of next year with the VW program we're getting kind of tracked. There's 50, 75 vehicles in testing. That data is being kind of monitored by VW for different types of interventions. That's all being tracked to the glide path that we at the beginning, and we're hitting our milestones. So we're excited about it.

A lot there. A lot of exciting developments there. How do we think about time line? Is this 2, 3 years away? I think there's some kind of no earlier than dates, I guess, in practice.

Yes. So I mean part of the plan is you want the hardware and kind of the -- you want the system to be integrated into the vehicle during the vehicle design process and you want that to be installed in the regular factory production line. ID. Buzz, which is kind of the vehicle that VW is targeting, I don't know, maybe it's $125,000, $150,000 a year. Our robotaxi will be kind of an option within the factory.

So if they wanted to scale up from whatever 2,000 to 10,000, you could do that relatively easily rather than having vehicles kind of built, delivered, you upfit the system into the vehicle kind of some third-party location. It doesn't have -- it doesn't do much for you kind of on the scalability side. So the scalable production of this vehicle is supposed to start in late '26.

Like I said before, we're looking to remove the safety driver in the middle of '26, so you could start doing kind of paid rides, start the commercialization process. You start kind of learnings around apps and things like that long before that. But yes, it's hard to call the time line, but we feel pretty confident that we'll be able to kind of get to the mean time between failure rate that would support commercialization sometime in the middle of next year.

In terms of the operational aspect, I know you also Marubeni is a partner, the servicing, the maintenance, the support -- who handles -- who is, I guess, responsible for them? What are the economics for them?

Yes. No. I think that that's going to take maybe a variety of configurations, right? I think like with Volkswagen, it appears that they want Moya, which is their ride-hailing, kind of their mobility division to be the consumer-facing brand. So like I said before, they would sort of be the owner operator of the vehicle like on the Uber network, whether they contract someone to do the maintenance and the charging or whether they do that themselves, like I'm not sure it will probably vary by region.

We have interest from rental car company. Is there other kind of entities like that, that have a lot of local assets that could be useful that maybe they -- that would be an advantage for them to own and operate the vehicle.

So I think it could take lots of configurations. But the key is to have the vehicle production, the technology that works like obviously very key and then kind of the user base, the demand generation platform. I think that there's a lot that each of those kind of 3 parties can provide in terms of value.

On the cost, one interesting point you brought up earlier is that it's not cheaper than Uber or Waymo is not cheaper than Uber in the market and then we got 25% market share. Is that -- do you think that's kind of capped as in you'll never really be able to take more than a certain amount of share at price parity in order to really -- let's say, I'm going to make like 80%, 90% share 1 day in the future. You actually have to be like significantly cheaper? Or how does one think about the consumers and the price?

Yes. I think it's hard to say. I mean car ownership varies like greatly depending on where you live, and it seems like ride-hail costs or kind of correlated to that in some ways. I think with -- the reason we like our approach is that if you are only kind of looking at operating a service -- let me start this way. So it's like when you think about it conceptually if these cars can operate 16 hours a day, like you're replacing 2 drivers.

So maybe you're replacing 2 bus drivers that are each making EUR 100,000 or maybe your replacing 2 Uber drivers that are each making $45,000 all in. So when you think of -- and you're replacing 3 years of that, right? So when you think about it that way, you're like, okay, if I can get the cost to $100,000 or $150,000, like it works. Our approach was to say, if you were to put a system like this in a regular privately owned vehicle like what would people pay, maybe $10,000, $8,000, $9,000. So you really have to get to this kind of $4,000 or $5,000 BOM cost to support that business.

There's extra kind of hardware that's going to be in robotaxis, so we're not going to be that low, but not too far away. So all of a sudden, that gives you a lot of room to think about like what is the car going to cost divided by 300,000 miles, let's say, it can kind of operate that long. You do that division and you get to like a pretty low kind of depreciation cost per mile. That's well below the $3, $4, $5 a mile that Ubers are typically costing.

So I think you can -- yes, you can get to the point where you can start to make people think about like, should I own a car or not? Or should I own 1 instead of 2.

At the same time, so you mentioned earlier you kind of agnostic to robotaxi versus personal autonomy. You had some OEMs pivot more towards, I guess, personal economy, maybe perhaps trying to imitate the Tesla model to some extent or the old Tesla model. Is that -- I guess, what's the -- how do you see the opportunity being different? And are those discussions being accelerated because of some of the things that are happening now?

Yes. I mean I think whereas you have -- you're starting to have like really good kind of demand proof points in the robotaxi space, like you kind of lack demand proof points in the call it, navigate on pilot, right? So you don't really have any kind of L3 consumer on the road. You've got a couple that are kind of very constrained, more like traffic jam systems. So there's not really a lot to go on.

China is a market where there's hundreds of thousands of sort of Tesla FSD like systems on the road, but nobody is being asked to pay for them, right? So it's like -- it's not like they're saying we won't pay for them. They're just not being asked. And so you're sort of lacking these like what will consumers pay for this.

I think the automakers feel like there's a -- in our opinion, feel like there's a very clear value proposition for an eyes-off system. Your -- maybe your commute is 80% highway. You can kind of disengage like catch up on e-mails, while your car is driving you, like that seems to be like a clear value proposition. Eyes-on hands-free, there's probably more questions about it. You've got companies like BYD sort of pushing cost down, focusing more on highway hands-free with eyes-on. So I think that there's an appetite for sort of a more efficient, lower cost system.

So I think we're -- what we're seeing kind of recently is more some new interest in Chauffeur because I think there's a view that if you get there, and we think we're the only supplier in the world with a real production program for L3 eyes-off, that's with Audi, launching late next year. If you get there, it's one of the few things that you could say like, okay, this is a real differentiator in a car. So we're seeing kind of incremental interest in Chauffeur.

We're seeing a lot of incremental interest in surround ADAS. We're still seeing kind of good progress on supervision, but it's just -- it's taking a while. So I think there's still a lot of appetite, maybe there's a little bit of a kind of an increased focus on eyes-off as opposed to kind of eyes-on all ODDs.

That's a good segue to SuperVision and Chauffeur. The big launch you mentioned Audi, what should we look out for next? I think there's a lot of moving parts just given the size and given the nature of it are we feeling good? Are we feeling ready that this will launch on time at kind of expected volume?

Yes, we feel good. So just to recap, we have production programs for SuperVision that's being led by Porsche, but it will be on Porsche, Audi, Bentley, Lamborghini models. We have a production program for Chauffeur that's being led by Audi. It will be on Audi and Porsche models, and then we have kind of the production program for Drive which is the robotaxi.

So yes, like we're like on time, on progress for SuperVision and Chauffeur, like the big milestone was getting the EyeQ6 high ECU converted into the test vehicles replacing the EyeQ5 high ECU with EyeQ6 high. EyeQ6 high has kind of 10x the processing throughput as EyeQ5. It was designed for a very kind of different software stack than EyeQ5 has with a lot more kind of transformers and foundational model type of stuff in there, a lot of new software engines that are to like deal with specific categories of errors that we're seeing with the new kind of LiDAR net and RADAR net processing as well.

And so now the hardware is in all the prototype vehicles for SuperVision and Chauffeur with the EyeQ5 software. It's up and running on the road. Now we add the kind of new software engines over time and start watching the MTBF progress happen, which we're seeing in the lab, but now we need to kind of prove it out on the road. So I think we're being tracked like day by day, like that's the value of having these production programs because it's -- Volkswagen holds you to a very high standard, is tracking you kind of time line, et cetera. And so everything like looks good.

We'll start to be able to put people like normal people like you and me into nonautomaker employees into the cars starting the second half of this year, hopefully at IAA. And so kind of we're looking forward to that because I think people will see a difference. You're not going to drive it 500,000 hours to see if you have a failure, which is -- or 500,000 miles. But I think in a half an hour of driving, you're going to see like a kind of a very robust system.

Well, I will be at IAA. I would hold you to that. There's a couple of other automakers out there. You have GM kind of integrating cruise stuff. Rivian has become much more vocal about it recently. I guess perhaps intends to kind of not rely on you. What are your thoughts on that? Is that wise?

on Rivian?

Yes. Or on both.

I lost my train of thought, Tesla and Rivian?

GM. GM cruise..

Yes. Got it. So yes, I mean, I think we've seen a lot of kind of ambition for in-house software development across the industry for the last like 5, 6, 7 years. We haven't seen a lot of success stories. We've also seen sort of a desire to kind of exert control, design control of a program by utilizing 5 or 6 different suppliers for different components. We think that that's really going the other direction now because kind of cost has become kind of very primary focus, ECU consolidation, like not having 4 or 5 ECU within the same system.

And I think we're one of the few suppliers that can provide a full stack solution. So we do think the pendulum is swinging back towards like a lot of openness to a supplied solution, but we're the solution, we're not the product right?

So like -- for Porsche and Audi, like they're not going to call their solution SuperVision or Chauffeur, they're going to call it their own name, and they should because they're the ones that taking that technology and turning it into an actual product deciding kind of how the vehicle should act in certain scenarios, how cautious, how aggressive, being able to kind of show visualization on the center screen. So there's a huge amount of work that the OEM needs to do and can rely on our core technology.

So like I think our view has always been like they shouldn't try to get into like the core technology of perception or driving policy, they should focus on taking the outputs system and creating like a product that people will love. And I think like we've been able to provide new tools for the automakers to do that. And I think it's working well.

On the question of cost, I think it's a good pivot to surround ADAS, which you have a big win on recently with the VW mass market brand. And based on the chart you had in Munich, there was quite a few others on there. What's the ultimate kind of trajectory of surround ADAS? And I ask in the context of -- it seems as if it's a very cost-effective solution to get actually a lot of performance that a few others can match internally. So does this become eventually the new norm maybe for developed markets?

Yes. We think it's a possibility. So we're -- there's a couple of -- there's a few market drivers that are happening that we feel like are creating tailwinds for this product. One is that the safety ratings tests every 2 years, the kind of safety ratings groups will kind of publish to the automakers like this is what you should expect 4 years from now.

And this year, for 2028 tests or last year, kind of all the new criteria that they're putting in like a lot of kind of side risk side dangers basically make it so we don't believe you can get to a 5-star crash rating with just a front camera. So it's forcing multi-camera systems. So I think like future-proofing, safety rating compliance, there's new rules in the U.S. as well that are kind of doing the same thing.

There were already parking like these kind of short-range wide-angle parking cameras on vehicles. You can utilize those. So like it's not really that much extra hardware costs. So anyway, future-proofing, safety compliance around ADAS, we'll deal with that. Getting to a highway hands-free like Level 2+ system that is cost-efficient compared to like what I was talking about before, these kind of multi-supplier sort of like OEM-directed systems with too much -- too many ECUs, too much cost. We feel like surround ADAS is kind of a vertically integrated single SoC, single ECU.

If you look at the Volkswagen announcement on this program, that's what they were focused was hardware and software sourced from the same place, single ECU, single SoC. This all reduces cost. So that's another kind of market driver. And then I think like -- I don't know if I mentioned it already, but like the BYD God's Eye announcement saying like, hey, we're going to put highway hands-free system on $20,000 cars for free that really got the attention of the industry because it starts to look like this is what consumers are going to expect in the future.

So yes, we have seen a lot of incremental interest and traction for this product. And we do see potential that it could be sort of like the new ADAS of the future. And it's -- considering it's triple the ASP for us, it would be a big revenue driver for Mobileye.

I want to spend the range kind of more on the base ADAS side, production schedule visibility. I think you've been pretty conservative or you guys have been pretty conservative on the last earnings call. Has that improved at all based on some of the tariff stability?

I mean you've done a good job like up-to-date on every action OEM has taken, and there hasn't been really a lot of kind of disruption that we've seen. So at the time that we reported in -- I think it was April 24, it felt like the world was going to end.

And we were seeing stability in Q2, and we kind of gave an outlook for Q2, it indicated that we would -- based on the midpoint of our full year guidance, it indicated that first half would be a lot higher than the second half. That was to account for uncertainty, like there's still a lot of uncertainty, but I would say that kind of incrementally, we feel better about kind of where the industry is, and we've seen fairly stable schedule. So nothing to report really. But yes, it feels like compared to what we were thinking in April, things have become more stable, less risky.

One area of outperformance, I think you have called out in both 1Q and 4Q, China JVs actually performed better. Is that something -- were those kind of one-offs? Or do you think that's actually holding up?

We were actually referring to the China OEM.

I think, yes...

Foreign JV production in China has been a pretty big headwind for us the last couple of years just because like those companies are underperforming in the market in general. China OEMs yes, we needed to kind of reestablish kind of a new way of doing business with the China OEMs. We see a lot of opportunity to help them with their export business where you really need to -- you've got very tight kind of safety rules, recall rules.

And we have good kind of solid kind of business with them for their export, but it takes some resource, right? So we want commitments on domestic volume as well. And so kind of we were able to get to kind of a level that I keep on saying stability like that's kind of where it is. Like since the third quarter of 2024, we've been $2 million to $2.5 million annualized run rate with that group of OEMs and kind of that's where we were a little bit.

We had kind of forecasted $2 million for the year. We -- our run rate has been higher in Q1. We were expecting a higher run rate in Q2. And if that continues in the back half, you could see some upside there as well. So yes, no, I think things are pretty -- going pretty well.

Then on Europe, we have seen some rumblings recently of some call volatility. It seems to not be that meaningful, but I wanted to check in with...

Yes. I think there's going to be volatility like we've been sitting in like for the last 3 quarters and then based on the Q2 outlook, we've been sitting in this kind of $8.5 million to $9 million per quarter range. And we feel like that matched up with end demand. The second half kind of to the midpoint of the guidance would be more like low 7s, right?

So -- no, no, like a little bit more than 7.5% per quarter. We haven't seen anything stark enough that would cause that type of a headwind, but you still have a long way to go for the year, right? Like there's still kind of risk for the consumer out there, what's going to happen to pricing. So it's we're being very cognizant of the market, understanding kind of where looking at end demand very closely. And so far, we've been encouraged, but we'll talk more about it in July.

Switching on the cost side, I think last quarter is also pretty solid cost execution helped by the LiDAR restructuring. Can you remind us how we're thinking -- how to think about the OpEx trajectory for the year?

Yes. So we 2/3, a little bit more than 70% of our OpEx is related to advanced products that aren't generating a lot of revenue yet. So we're definitely like if you looked at the base ADAS business stand-alone basis, it would be a lot higher margin than we're showing right now as a total company. I think that's a positive potential for the future.

And I think we've increased R&D a lot for a several year period, but feel like we're kind of at a steady state right now where we have the right resources in place to support these 4 new products that we're bringing out to support the production programs with the automakers, and we would expect like kind of comp inflation to be the only real growth area over the next few years, so 5%, 6%, 7%. Comp inflation is higher now than it used to be.

On CapEx, I think it's been asked you in the past a few different ways. But if we take into account everything we just talked about on autonomy, robotaxi, your CapEx has been actually fairly low. Do you, at some point, consider increasing that to either accelerate some of these advanced efforts, own more of the stacks. I think you've been asked before like some on the GPUs, do you need to own more to kind of do the training aspect of the gen AI. Is CapEx kind of -- is that something that could move a lot? Or would you consider moving that a lot?

So a lot of our CapEx is related to AI training infrastructure, and we've been buying $50 million, $60 million of GPUs for a long time. So we've got a very kind of solid base of kind of processing capability. We have 300 petabytes of data that we're using for training. So it's like a very, very large data set. So I think that we don't feel constrained by compute.

We also leverage AWS, also significant spend on AWS, like $40 million, $50 million a year inside of our OpEx. So we don't feel like we're constrained at all on training infrastructure. We're trying to stay capital light. That's one of the reasons why we're pursuing kind of the business model. We are in the robotaxi business and not wanting to own fleets. So we don't see a lot of risk that CapEx needs to go up maybe modestly, but not any kind of step change.

All right. I think we have time for one question, if anyone in the audience [indiscernible]. So back in Munich, we obviously grow in cars, and it was impressive because we have a couple of other cars. What do we sort of expect next? What are we expecting next in terms of is the focus on robotaxi with all these partnerships? Is it on the SuperVision customer because I think the Japanese one has been a little bit shaky. Or is it huge volumes on surround ADAS? What is the priority amongst all these things that we talked about?

I think like the main priority for us executing the VW programs, like I think me as a sell-sider back in the day. It's like you get a design win and it's like 2 years from now, you're going to start generating revenue, but it actually takes like a ton of work to remain on time and actually produce a really good product. So we feel like the VW programs with SuperVision, Chauffeur and Drive are like the ultimate proof points for us, right? And so we have to execute those. We have to remove the safety driver middle of next year, so we can start to ramp the robotaxi business. We've got to get to L3 highway so we can prove out that this can be done in a safe kind of transparent, explainable way. So that's the main priority for the company.

Obviously, like we would love to have like another 2 or 3 VWs where you've got this sort of like synergies between the different programs and a company sort of aligning behind your portfolio like we think that, that will happen. So we're definitely looking to kind of close convert deals with new customers, but we can't keep our eye off the ball with the VW programs because that's going to be the kind of the ultimate proof point for the industry that you can do these kind of low cost efficient, high accuracy, high precision, high availability systems, primarily with cameras, but also with RADAR and LiDAR.

That's it. I know we're just about out of time. Thank you, Dan.

Thanks. Good job. Thanks.