Greetings and welcome to the Mobileye Fourth Quarter and Full Year 2024 Earnings Call. At this time, all participants are in a listen-only mode. A Q&A session will follow the formal presentation. If anyone should require operator assistance during the conference, please press star zero on your telephone keypad. As a reminder, this conference is being recorded. I would now like to turn the call over to your host, Mr. Dan Galves. Thank you, sir. You may begin. Thank you. Hello, everyone, and welcome to Mobileye's Fourth Quarter and Full Year 2024 Earnings Conference Call for the period ending December 28, 2024. 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 CEO and President, and Moran Shemesh, Mobileye CFO. Also joining today for the Q&A session is 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, Q4 was closely aligned with our expectations. EyeQ volume was a bit better than expected and was up 9% versus Q3. The upside was largely related to higher-than-expected volume from Chinese domestic OEMs, who continued to order above the levels we talked about back in July. ASPs, gross margin and operating expenses were aligned with our view at the beginning of the quarter. Operating margin of 21% was almost five points higher than Q3. Operating cash flow was robust in 2024, finishing up at $400 million. That was flat compared to 2023, despite significant year-over-year revenue and earnings declines driven primarily by the previously disclosed inventory digestion period we experienced in the first half. Operating cash flow was approximately double our non-GAAP net income in 2024. Moran will cover the guidance in more detail in a few minutes, but I will first set the stage. Bottom line is that if we use indications from our customers for the full year, our guidance would be higher. But we prefer to take a conservative approach that accounts for the risk that uncertainties negatively affect earnings. Our top 10 customers were assuming global production volumes meaningfully worse than that assumed by third-party forecasters. With Chinese OEMs, forecasting remains difficult due to less visibility we receive. Volumes appear to have stabilized in the 2 million-plus annualized range in the second half of 2024, higher than we had expected several months ago. But we're assuming a deterioration from that level simply to account for the low visibility. On SuperVision, we're assuming about half of the current run rate of end-market demand for the vehicles we are on. On ZEEKR, we are electing to account for the risk that ZEEKR could choose to go with their in-house system on ZEEKR 009, which is currently running at 2,000-3,000 units per month. We have no indication that this is planned, but we are unwilling to be surprised again. On Polestar, we're assuming volumes of Polestar 4 to remain at current levels, despite their stated plans for further geographic expansion. Turning to seasonality, as a percentage of the midpoint of the full-year revenue guidance, our assumption for Q1 revenue is 25% of the full year. Typically, it would be lower, so this gives us further confidence that the full-year outlook is achievable, even if macro conditions deteriorate somewhat. Turning to the commercial and new business side, we continue to win new ADAS business with our core customers at the same very high rate we have for years and are seeing good opportunities with some new customers. We recently won a multi-million unit REM data harvesting and cloud-enhanced driving assist program from a very key customer, in parallel with continued due diligence for SuperVision. This deal strengthens our global data harvesting with another leading OEM with significant global volumes. This data plays a key role in our EyeQ6-generation AI stack. Additionally, we move forward with an Indian OEM on REM data harvesting for this important growth market. A number of upcoming launches from this customer will have cloud-enhanced driving assist capability. On the advanced product side, we have the customer engagements in place to drive a steady cadence of announcements over the course of 2025, which is consistent with the messaging we delivered at the December Capital Markets Day. While our expectations for advanced product design wins remain intact, the exact timing of those announcements remains challenging to predict. The decision for an OEM on what path to take towards autonomy is very strategic, a very long-term decision, and this is why the due diligence process is so intense and takes so long. But it is clear that the customers believe in our approach and that the path to outstanding products runs through Mobileye technology. In the past few months, we have revealed through our AI Day back in October, the Capital Markets Day in December, and the CES early this month, a lot of technological advancements underlying our stack. The common theme is efficiency of design, efficiency in our silicon design, and we'll be revealing very soon detailed benchmarks of EyeQ6 High versus competing high-performance chips. Efficiency of our AI, for example, the AI Day, how to build a transformer network, which is a factor of 100 more efficient than the existing transformer architectures used by practitioners. This efficiency in design is also translated to the amount of resources, data, and compute which is required to train our AI stack. Efficiency in our mind matters and goes against the dogma of brute force development. Brute force development is a signature of our competitors, as reflected by the massive investment in compute and data pipelines. Mobileye has a long tradition of excellence and efficiency. Pre-production vehicles powered by our EyeQ6 High stack show great promise for a substantial leap in performance and precision of our next-generation supervision and Chauffeur. More updates will come in the course of 2025 as we get ready for standard production during 2026. Thank you, and I will 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 primary exclusion in Mobileye's 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, as well as the goodwill impairment that occurred in Q3. Our Q4 results slightly exceeded the Q4 outlook implied by the full-year guidance we provided back in October, largely due to higher-than-expected volumes from Chinese OEMs. It was down 23% year over year. There isn't much insight to be gained from that comparison, as a portion of the meaningful inventory build-up that impacted the first half of 2024 occurred in Q4 of 2023. Q1 will again be an apples-to-oranges year-over-year comparison, given the inventory digestion that occurred in Q1 of 2024. Beginning with Q2, the comparison will be more relevant. Gross margin was consistent with expectations and up slightly versus Q3 due to a lower percentage of SuperVision revenue in Q4 versus Q3. Operating expenses were down somewhat versus Q3, as expected. This is related to the initial impact of the LiDAR unit wind-down, an increase in military reserve duty refunds, some adjustment based on revaluation of benefit accruals, and other items that are largely timing-related. Turning to guidance, we provided 2025 revenue and adjusted operating income guidance in today's earnings release. At the midpoint, we expect $1.75 billion of revenue and $217 million of adjusted operating income. This represents approximately 6% year-over-year revenue growth and more than 10% growth in adjusted operating income. The revenue guidance is based on EyeQ volumes in the range of 32-34 million units. Amnon mentioned we are assuming the vast majority of supervision units this year are for Polestar 4 and expect overall volumes in the low 20,000 unit range at the midpoint, assuming no expansion into the U.S.. I note that for the time being, we don't plan on expressly addressing supervision volumes in the near term. Until we begin launching this system on more products with Western OEMs in 2026, we don't expect it to be significant enough to call out, and we want to set expectations low enough that any variances to downside are not material. On the EyeQ volume side, based on our analysis and information from Q1, we believe that customer inventories are currently at normal seasonal levels. To give you better insight on our guidance, we thought it would be helpful to provide a bridge from the second half 2024 annualized run rate of 35.6 million units to the midpoint of our 2025 volume outlook of 33 million, which is somewhat below the indications we currently have from our customers. We don't plan to provide this level of detail on a quarterly basis, but think it is helpful as context for our initial 2025 outlook. First of all, some pull forward of volume into Q4 is typical given annual price changes and volume bands. We estimate this added about 500,000 units annualized in Q4 2024, and we are not assuming this occurs again in Q4 2025. We assume that 2025 overall production of our core OEM customers, which represent about half of industry volume, will be down almost 7% versus 2024 levels, similar to the decrease we saw in 2024 versus 2023. This is meaningfully lower and more conservative than IHS projection of - 4% and would account for about 2.3 million units of reduction. Partially offsetting the assumed production declines, we do expect share gains and ADAS adoption growth from these customers to drive approximately 1.5 million units of growth in 2025, or about four points of growth over market. These are not generic expectations, but rather relate to new programs and additional markets with specific OEMs. Regarding seasonality, second-half production at our top customers is typically 2% higher than first half. This would represent almost 1 million unit annualized reduction versus the second-half run rate. Finally, regarding the China OEMs, we are assuming approximately 500,000 unit decline as compared to the second-half run rate. This is to reflect the volatility we've seen over the last several quarters. Turning to gross margin in 2025, we are assuming about 1.5 points higher than 2024, primarily due to the lower percentage of SuperV ision-related revenue. In terms of operating expenses, we expect about $250 million per quarter during 2025 on average, which is consistent with our comments on the Q3 call, where we said the run rate at the same time would be sustained through 2025. Versus that Q3 run rate, we will see savings from the wind down of the LiDAR unit. We expect this reduction to be offset by typical employee compensation inflation, as well as our expected military reserve investment. I remind you that headcount-related expenses represent well above half of our OpEx. Other areas of growth, such as AV testing and customer support and cloud-related expenses, are largely offset by efficiencies within our data labeling activities, as well as expected higher year-over-year engineering reimbursement on production program spending. In terms of Q1, we expect revenue to be down about 11% versus Q4, which reflects typical seasonality and a bit more than 80% growth year-over-year against the inventory digestion we experienced in Q1 of 2024. Our revenue expectation for Q1 implies about 25% of the midpoint of our full-year revenue guidance. We expect overall gross margin about 100 basis points higher than Q4 level and for adjusted operating expenses to be at or slightly lower than the $250 million per quarter I indicated earlier. Operating cash flow in 2024 was well above adjusted net income, even after taking into account capital expenditure of $81 million. Cash flow generation in 2024 was consistent with 2023, despite substantially lower adjusted operating income, reflecting strong management and control of working capital in a variety of areas. Additionally, we anticipate continuing to reduce our strategic reserve of cheap inventory on our balance sheet in 2025, which should support another year of delivering operating cash flow above adjusted net income. Finally, we expect the full-year effective tax rate to be approximately 20%, similar to 2024. Thank you, and we will now take your questions. Thank you. At this time, we'll be conducting a Q&A session. If you'd like to ask a question, please press star one on your telephone keypad. A confirmation tone will indicate your line is in the question queue. You may press star two if you'd like to remove your question from the queue. For participants using speaker equipment, it may be necessary to pick up your handset before pressing the star keys. In order to allow for as many questions as possible, we ask that you each keep to one question and one follow-up. Thank you. Our first question comes from the line of Shreyas Patil with Wolfe Research. Please proceed with your question. Okay. Thanks so much for taking the questions. Maybe just at a high level, as you're talking to legacy OEMs on adopting advanced automation like SuperVision, I'm just curious what kind of timelines they are considering for this kind of autonomy because, Amnon, you mentioned uncertainty by OEMs as it relates to making that decision. So are the legacy automakers still viewing this as a 2027 or 2028 kind of launch for real adoption, particularly in markets like North America and Europe? I'll start, and then maybe Nimrod would add. During 2024, we did a lot of build-up with the potential customers, building and development vehicles with our stack, working with them on very extensive testing. We believe all of that will bear fruit in 2025. Exact timing in 2025, it's difficult to pinpoint, but all the indications that all what we did in 2024 are going to bear fruit, both in SuperVision and in Surround ADAS. Surround ADAS is also a very interesting category that is being built right now. 2027 seems like still looks like the sweet spot in terms of introduction of these kinds of technologies. The work of Tesla is really creating a sense of urgency with our OEMs. We still believe that 2027 is really the right timing for introduction of these systems. Nimrod, do you have anything to add? I agree and support. The majority of the RFQs that we have and the engagements that we have are addressing or aiming for the 2027-2028 timeframe, and especially for, I think, Surround ADAS, which is kind of a new category that has been picking up in the past year, I think that there is an even more expedited sense of urgency, let's say, due to also regulatory drivers, not just the competition, so that still remains intact, and we did not see any shift in the timeline so far. Okay. Thanks. So maybe just on that point, because I think typically you've talked about OEMs would have to kind of secure a win maybe two to three years ahead of deployment. I know it's hard to pinpoint timelines, but just trying to get a sense of your confidence level on some of these awards, particularly the chart that you had provided at the Capital Markets Day, which showed a number of OEMs that seemed fairly closer to crossing the finish line. So just trying to get a sense of your confidence on those. Nimrod, do you want to? Yes. Yeah, yeah. So again, I think we've been maintaining steady progress in these engagements, and we kind of continue to do so. And as Amnon said at his remarks, we cannot really predict the exact timings of decisions, but we maintain steady progress since the Capital Markets Day in these activities. We believe also that any start of production until middle of 2027, if the nomination is within the next four or five months, we meet it, especially if we're talking about the carryover of what we're doing with the Porsche and Audi on SuperVision. And there's also a possibility to introduce EyeQ7 in that timeframe as well. Yeah. And just maybe to add one comment on this, if I may. I think what is also happening in parallel is that we continue to mature our next-gen products within our programs with Volkswagen Group that do address or have a concrete plan to start production in 2026 onwards. So our work in maturing the technology, the hardware, software, next-gen AI technologies is continuing as planned. So it does give some added confidence also to our future opportunities that we can meet even shorter timeframes. If and when decisions are going to maybe take a little bit longer, they can maintain the same SOP plan because Mobileye is continuing to improve and kind of mature its products in parallel. Okay. Great. And maybe just a quick one. How do we think about gross margins for 2025? Thanks. Yeah. I believe I mentioned that we expect a slight increase in gross margin of 1.5% as a result, mainly of SuperVision being lower volume than 2024. But within EyeQ, it's pretty flat within the year, so no issue with gross margin on 2025. Thank you, Shreyas. Okay. Thank you. Let's move to the next caller, please. Thank you. Our next question comes from the line of Chris McNally with Evercore ISI. Please proceed with your question. Thanks so much, team. Just maybe around the ADAS Qs, Amnon. I think one of the questions that we always get from investors is, if you were to lose these, what would be the reasons that you were to lose them? And I kind of always think about three buckets. The first bucket being sort of timing or push-out. OEM just is unclear which trims, which vehicles they want. They're constantly deciding, and that changes year by year. Sort of the second, which is very clear within the legacy OEMs, is there's no action, and they just decide to go with their basic level two, and they pass on advanced solutions. And then, obviously, the third is sort of more Mobileye losing, which would be an OEM decides to go to in-house. So I don't want to hold you to percentages, but I'm just curious where you think the risk would be in those three buckets: timing, OEM taking no action, or an in-house solution. Thanks so much. Well, I think the third bucket, in terms of in-house development, we don't see a trend there. We don't see anything that is serious about in-house development of such advanced products. So it's mostly related to the first two buckets, which is all about, now, power train. So sales of EVs were below expectations. So car companies are kind of going back to the design board and putting more emphasis on combustion engine models. So issues of power train, so that delays also driving assist or big driving assist decisions. But in terms of in-house development, we don't see any significant traction there. Nimrod, do you have anything to add? I think that it's a little bit hard for us to answer this question because we're not aware of all the behind-the-scenes considerations. So from our perspective, it's obvious that there are attempts in the market for in-house development, and GM recently talked about their plans and so on. So we are aware of those. I think for the kind of opportunities that we are now pursuing, it's a combination of these buckets that you mentioned that can play into this. Mostly kind of whether or not the OEMs are ready for such a product at this timeframe, what will it take from them, and some perhaps more like a vehicle lineup considerations. Longer term, we cannot really predict what will be the outcome of these in-house developments, but we've been working with customers that did have these attempts in the past, and we did manage to build a very successful business in parallel because eventually we executed, so. Very helpful. Just a quick follow-on. In SuperVision, the launches you have in 2026 that are sort of known and the 2027 ones that are in our queue, are they mostly or the majority, are they standard or take rate? Thanks so much. So it's not decided yet, to be honest, and it's something that we're discussing with our OEM partners, different models of how to sell these products to consumers. There is a combination between standard fit and opt-ins. So the more we'll have information on this as we move forward, we will share this, of course. Thank you. Thanks, Chris. Thank you. Our next question comes from the line of Joe Spak with UBS. Please proceed with your question. Thanks, everyone. Amnon and Nimrod, I'm just curious again on the OEM conversations if price at all comes up and if that's a pushback? And if so, I just wonder how you think about that, given it seems like if you give a little bit there to gain a footing, it would seemingly make you stickier and perhaps more valuable to customers over time? And then I guess also on the conversations with your urgency comment, does that also mean that if and when they do sign, they're also willing to move quicker to implement than they have historically? Price is always a consideration, but I don't think that price right now is any impediment to make decisions. We are very optimized on price, and we work with our customers to find the right solution. Nimrod, anything you want to add? Yeah. I don't think that we have lost a program on prices. I don't think it comes down to this at this stage. And we are, we can say, aggressive and flexible relatively with our prices in negotiation. We don't want to make this a stepping stone. So I don't think this is right now kind of a challenge that we need to overcome. I think that the sense of urgency that you asked about is more about how many vehicles they can deploy such assistance on and so on. That is more a question of the OEM's plans and rollout plans. It's not just a question of they want to do this, but if they can really accelerate the execution and deployment in a larger scale of cars. This is something that we are working with them, and we're not the bottleneck, let's say, for a broader expansion of vehicle integration. We are very, very efficient and have one of our strengths, actually, is the execution side and being able to support multiple vehicle lineups, multiple vehicle architectures. So if and when this will become the chosen strategy, we can indeed support this. Okay. Just as a second question, obviously, a lot of news in AI this week, and you guys have always talked about the efficiency of your development solution, but I'm just wondering, big picture, does any of the developments get you to reevaluate your own approach, or do you have any views on whether the prevalence of open-source foundational models can lower the barrier to entry to others? I think what we have seen with the DeepSeek v3 and DeepSeek R1 really aligns with the approach that Mobileye has been advocating for many years is that you need to build a purpose-built approach, which means efficiency. At the AI Day we had in October, we revealed a transformer architecture that we built, which is 100 times more efficient in terms of runtime and compute than any standard transformer because it's purpose-built for the task of autonomous driving, so everything we do in terms of the silicon design, if you look at the area, the area of our EyeQ6 is one-fourth of the silicon area of our competing chips. We will provide in the next couple of weeks significant benchmarks that we have been doing with the leading high-performance compute chips. And EyeQ6 is exceeding in terms of major KPIs like running both convolutional net and transformers, all the chips that we have benchmarked against. So efficiency is really the hallmark of what Mobileye is doing. And what DeepSeek has shown is that if you innovate in engineering, there was nothing there scientifically new that our community did not know about. But the fact that they created a very tight flow of training in terms of making use of memory bandwidth, making use of efficient reinforcement learning like the GRPO instead of PPO, working with quantized precision, FP8. They really made and they even wrote assembly language to bypass all sorts of CUDA bottlenecks that they had there. So if you are really purpose-built and you want to be efficient, you can gain a lot. And this is what Mobileye has been advocating for many years. This is exactly aligned with our approach. Thank you. Thank you, Joe. Thank you. Our next question comes from the line of Dan Levy with Barclays. Please proceed with your question. Hi. Thank you for taking the question. First, I wanted to ask a question on the guide for '25 and specifically on the IQ shipment guide. Maybe you could just talk about the extent to which launch activity is factored into that guide. To what extent is that based on launches? I think what we saw in the past was that although you're powertrain agnostic, it's typically EVs that are taking on more advanced content, and we saw some slowdown in EV activity. So maybe you could talk about the extent to which launch activity factors into the guide. Okay. I'll start by saying that maybe Nimrod can add, and Moran, that we took a very conservative approach with guidance. If we would have taken the numbers we received from our tier ones and OEMs, the guidance would have been much, much closer to consensus than what we did. We took the conservative approach because we don't want to re-guide during 2025. We don't want to risk that. Our guidance includes new launches as well. Nimrod, do you want to add something? Yeah. I think the mix of new launches versus kind of carryover programs is normal this year. There is no higher or lower percentages of new product launches versus carryovers. So it's pretty consistent with what we have been facing in the past few years. And those new launches are, sorry, go ahead. Oh, I just wanted to say that on the new launches and our market share, there are cuts in some of the global production reduction or volume reduction. This happened also in 2024. So it's not just an expectation for 2025. In terms of market share, we did above market also in 2024. It's just that our customers went down. Yeah, top-down customers just went down like 6 or 7% from 2023 to 2024, but there was definitely some offset to that. Understood. Thank you. Maybe we could then just follow up on China. And if we could just mark where you are now, I think based on the disclosure you're giving, you're assuming something like a million units from the domestic OEM, something like 5 million from the multinationals in China, the volume outlook, and where do you stand now on resource allocation and efforts in China, especially amongst the domestics and taking into account maybe some of the challenges that you went through in China in 2024? Yeah. I'll maybe address this and Amnon and Moran can add. So our volumes in China next year for IQ is growing compared to 2024, and we're close to 2 million units for the Chinese OEMs. What we are finding is that we have some good level of stability with the Chinese OEMs that have significant export volumes, and they are willing to—they're willing and they're very much keen in partnering with Mobileye as a global proven solution that has proven performance. And maybe the risk of potential technology restrictions in some western markets is, of course, lower than adopting a Chinese solution. So that has really helped us in kind of maintaining and solidifying our position with our core customers and expanding our position with our core customers, Chinese customers. We have been optimizing mildly our investment in China. We still have a strong team there that supports all of the local R&D needs, that maintain full compliance with the local restrictions and regulation for data and so on. So we're maintaining kind of a very tight investment compared to the business in China. Okay. Thank you. I wanted to follow up. This is Dan. So just to answer your question about kind of where we stand, last year we did about five million units with the kind of non-Chinese OEMs in China and somewhere around the 1.5 million units with the Chinese OEMs. One of the main reasons why our kind of assumptions are deviating from IHS is kind of the global OEMs in China were down 17%-18% last year. And now the assumption is they're down kind of in the kind of mid-single digits or high single digits. We're taking kind of the view that it could be worse. And if it's worse, we don't want to have to re-guide, like Amnon said. So a significant reduction in kind of non-Chinese OEMs within China is baked into our forecast. In terms of the Chinese OEMs, like Nimrod mentioned, the volumes were encouraging in the second half of 2024. They ran at kind of well above a two million unit run rate in the second half, which was quite a bit above where we thought that they would be back in July, I think, for the reasons that Nimrod went through. We were not seeing any kind of change to that, but we're assuming that we have about 500,000 deterioration versus kind of where we were in the second half, really just to be safe and due to the lack of visibility, so hopefully that helps to kind of size the China business right now. Thank you. That's helpful. Thank you. Thanks. Over to the next question, please. Yes. Our next question comes from the line of Adam Jonas with Morgan Stanley. Please proceed with your question. Thanks, everybody. So Amnon, you said in your comments you do not see significant in-house development from OEMs. You don't see that as a real trend. I'm just curious then, the ZEEKR 009 example of moving in-house, you see that as kind of a one-off limited to China? And I'm curious how you view NVIDIA Cosmos. Even if it's not a direct competitor, are they offering tools that could help encourage and maybe accelerate in-house development from your OEMs that maybe you haven't seen yet? And then I have a follow-up. I think Cosmos is not directly related to empowerment and enablement. There's a big difference between supporting humanoid robotics and supporting autonomous cars. With humanoid robotics, every robot is built differently. Actuators are placed differently. And if you just train on real data, you'll not be able to generalize among robotic platforms. So relying on simulators is crucial. For example, in my other company, Mentee Robotics, we built foundation models just on simulated environments, and we use a lot of NVIDIA tools. In autonomous driving, roadways are structured, and you want to rely on real-world data in order not to create distribution shifts. You do use simulations for edge cases, but you don't build your entire stack on simulators. So I don't think that that would be a major enablement. ZEEKR 009 or whatever is going in China, I think, is separate from what's going in the western economies. So when I mentioned about in-house development, I was referring to the western world. Okay. Thanks for that. Just as a follow-up, and you alluded to it, humanoid robots have been getting a ton of attention lately from Tesla's efforts or other OEMs and tech firms. You're chairman and co-founder of Mentee, as you mentioned. So I'm really interested in how you think about the adjacent market opportunity for Mobileye's computer vision technologies and expertise in other markets. Because you said Mobileye's not the bottleneck. Your industry that you serve, the auto industry, predominantly the legacy auto industry, the vast majority of it, there may be scenarios where they're just not ready. They're the bottleneck in that if you anchor all of your talent and IP to that slow-moving part, that could put your company at risk. So I'm curious, how are you viewing? How do you see the surface area between your computer vision tech and aviation and drones, humanoids? Are these projects that Mobileye are currently exploring right now? And if so, at what stage, at what point could that be material? Thanks. I think Jensen and his CES talk refer to AI in the real world as physical AI, and there are a lot of synergies, and at Mobileye, we are now studying this. We're not in any mature place in which we can make decisions, but we are definitely studying in-depth synergies between our stack and outside of automotive, but it's really early stages. Got you, Amnon. Thanks. Thank you, Adam. Thank you. Our next question comes from the line of Colin Rusch with Oppenheimer & Co. Please proceed with your question. Thanks so much, guys. Can you speak to the cadence of change of incentives within the reinforcement learning platform and any potential strategies for utilizing elements of any of the emerging archetypal models or portions of some of these foundation models that are starting to come to market? Reinforcement learning is becoming a very critical tool in building foundation models. And we use a lot of reinforcement learning in our AI stack as well. The reason you want to use reinforcement learning is when you build a probabilistic engine, right? So the next token prediction you have in pre-training, it's a probabilistic engine. There's no concept of correct and incorrect. With reinforcement learning, through adding a reward function, you're adding the notion of correctness. So you need this when you're building useful foundation models, and you need this also when you're building outputs that are relevant to autonomous driving because correctness is talking about precision, right? You want to be precise and not just be probabilistic. So reinforcement learning is a crucial element. And there's lots of innovation in reinforcement learning. What DeepSeek has done, they use kind of a simple reinforcement learning approach called the GRPO, which allows them not to use a critic, not to use complicated reward functions, and use only outcome rewards instead of process modeling, process rewards. And the fact that it works so well is very, very nice. And we're also looking at it instead of using PPO or DPO to use this GRPO. But it's a small thing. Once you understand that it could be useful, you simply add it to your stack and experiment with it. And we're doing this all the time. Super helpful. Thanks, guys. And then just from a sensing perspective, you talked a lot about the evolution of your imaging radar technology. But can you talk a little bit about what you're seeing in terms of sensor fusion and the ability to integrate some of the sensor data in a more efficient way to streamline the overall system and how that will develop over the next 12-24 months? We see Imaging Radar as a game-changing sensor. And we see now a lot of acceptance from OEMs to include the imaging radar into a Chauffeur platform. And we'll have more to say about it during 2025, but it's really a game-changing sensor. Thanks so much, guys. Thanks, Colin. Thank you. Our next question comes from the line of Mark Delaney with Goldman Sachs. Please proceed with your question. Yes, thank you very much for taking my questions. So if you could share an update on how Mobileye Drive technology is progressing for robotaxis and the timing of when you think there will be AVs on roads for commercial operation, not only in Germany, but I think VW had planned to launch in Texas using Mobileye Drive? Our activity with Mobileye Drive and the lead customer is Volkswagen ID. Buzz. We have also additional activity with Holon and Ruter, which is now starting paid with safety driver paid drives in Oslo. We have an additional with Benteler and additional activity with the Holon platform. We're working with additional opportunities and it's all targeting end of 2026, 2027, so throughout 2025, we're replacing the compute hardware, EyeQ5 based to EyeQ6. That's a Drive 64. The imaging radars are already inserted and working on this platform. We're using the Innoviz LiDARs there. We finished training all the networks for those sensors and by end of 2025, we should be at the right mean time between failure to start experimenting with removing the driver, but SOP is end of 2026, 2027 for removing the driver. Thanks for that, Amnon. You should see milestones occur over the course of 2025. The first milestones are closed user group testing where you're taking actual members of the public and the systems, commercial launch in terms of starting to charge those customers. This is all kind of in the near term during 2025. Expect to hear more about this. Thanks for that, Dan. My second question was just following up on some of the earlier commentary around the RFI and RFQ pipeline. And at the capital markets day in December, there were five OEMs in the negotiation or due diligence phase for supervision, another three at those more advanced stages of evaluation for Surround ADAS. You said today you're continuing to make progress on that pipeline, but could you clarify if any of those opportunities are no longer available to you or have meaningfully pushed out? And especially in Japan with some of the major OEMs, they're announcing M&A or plans to work with some other chip vendors, at least for full AVs. I am hoping to get a bit clearer of an update on where some of those opportunities stand. Thank you. There hasn't been a change negatively in these opportunities so far. So there is no one that is dropped. Thank you. Thanks, Mark. Thank you. Our next question comes from the line of George Gianarikas with Canaccord Genuity. Please proceed with your question. Hi everyone. Thank you for taking my questions. I'd like to ask also about the bake-offs that you're in with the OEMs. You talked about the in-house development, but could you also maybe discuss what you're seeing from other alternatives, whether it's emerging competitors like Wayve? Waymo has made some noise about trying to sell to OEMs, Tesla. How often are you seeing them in your discussions as well? Thank you. We do not see in those discussions real competition outside of in-house development. All those that you mentioned, we don't see them in competition. I think that those are demo vehicles and could be relevant for the end of the decade, maybe production, but it's still a long way to go. If an OEM is interested in a 2027, 2028 timeframe for a SuperVision or Chauffeur, then the best path to get there is Mobileye. Thank you. And maybe as a follow-up, but just curious as to what your collective thoughts are on Tesla's FSD version 13.2 and any progress you think they've made? Thanks a lot. We've been test-driving the Tesla FSD version 13. We have strong conviction that our EyeQ6 platform will greatly exceed whatever we experience with the FSD version 13. I would like to note that really the Holy Grail that we're pursuing is not just a good driving experience. We are pursuing very high precision. High precision meaning that we can perform an eyes-off driving experience like with Chauffeur. So there's way more to address than just the driving experience. It's how do you reach a very high mean time between failure? The mean time between failure that exists today is around 10 hours, even with the best systems like the FSD version 13. We reach tens of thousands of hours. This is really the Holy Grail, and we believe we can reach this with our EyeQ6 platform. Thanks. Thank you, George. Thank you. Our next question comes from the line of Edison Yu with Deutsche Bank. Please proceed with your question. Hi. Thank you for taking our questions. First, I just want to come back to kind of the AI DeepSeek developments. Would there be anything you would call out that you found maybe interesting in terms of their approach and what they did? I know you highlighted a couple of things earlier, but it seems to be this kind of, I think, read-through that some of the companies in the States are going to apply some learnings from that. Do you get any sense that you would do something like that, or you would kind of modify anything you're doing right now differently? I think when we're talking about the stack of autonomous driving, what was interesting there is the use of GRPO in the reinforcement learning stack instead of PPO and DPO. And we're looking into it. It was very interesting. But I think that this is a very small thing. Their achievement is being able to develop a very tight flow, very tight training flow, taking into account every small bit of memory bandwidth and processing of the available chips that they had, and reach a very low-cost training. Whether that is $5.6 million or $50 million, I don't know. But when you read their technical report, you see a lot of engineering innovation. So I think that is real. And what we take out of it, which is relevant to us, is really the use of GRPO and the reinforcement learning stack. Thank you. That's really interesting. Second one, I know there was a Lyft partnership announced back in November. And I think you've mentioned several times there's this urgency from a lot of OEMs. I'm wondering, are you getting any increased urgency from kind of the mobile operators or ride-hail operators, whether it's Lyft or other parts of the world, to try to get the robotaxi deployed faster because of what Tesla is doing, because of what Waymo is doing? If there is more urgency, could we get kind of accelerated discussions along with that? Yes. We do sense urgency to come up with announcements on robotaxis, but we need to align all the partners together. It's not just the operator. We need also the vehicle platform to be aligned. So this is fully aligned with the ID. Buzz and with the other two opportunities we have. And we're building now a fourth opportunity, which also will be meaningful. So I do see kind of a revival of the robotaxi opportunity, mostly due to the success of Waymo. And we see indications from the market in terms of partners, whether it's operators, vehicle platform builders who would like to play a more meaningful, to be a more meaningful actor in this emerging market. Thank you very much. Thank you. Our next question comes from the line of Antoine Chkaiban with New Street Research. Please proceed with your question. Hi, good afternoon. Thank you for taking my questions. I have a question on your current SuperVision design win. So can you maybe tell us about your collaboration efforts with Polestar, Volvo, and Volkswagen? How the relationship is going? Any interesting development over the last 90 days and how to expand the opportunity with those customers? I think our first-generation supervision in terms of development, we are kind of in the low marginal returns. We have been adding automated parking to the stack and recently launched it in China and will continue to develop it further, but in terms of improving that stack, we reached a point in which our focus is on the EyeQ6 platform. We are doing a complete software rewrite on the EyeQ6 platform to make better use of the benefits that we have from the EyeQ6. The 10 times more compute that we have requires some software rewrite, so all our focus is now on the EyeQ6, and throughout 2025, we'll be able, we have already pre-production vehicles with EyeQ6, already doing testing, on-road testing in Germany, and throughout 2025, we'll be able to show significant improvement on supervision performance with EyeQ6. Thanks. And maybe as a follow-up, so you presented at the CMD, the DXP framework to enable OEMs to code and control elements in the system affecting the driving experience. I'm wondering what feedback you're getting on DXP from your current engagements and potential additional design win. So in our activity with SuperVision, DXP is in full use. So it's a very important stack as part of the development stack that we have with OEMs. Thanks, Antoine. Operator, the next question will be our last question. Thank you. Our final question comes from the line of Tom Narayan with RBC Capital Markets. Please proceed with your question. Hi. Thanks for taking the questions. My first one is just a quick follow-up on, I think, Shreyas's question on the famous slides from the investor day. Just wanted to clarify the two OEMs, one for SuperVision, the other for Surround ADAS, that are close to nomination. I heard something like, was it like instead of being two or three months away, that you still believe that'll happen in 2025? Is that correct? Or should we just assume let's not put strict timelines associated with this from now on? Is that kind of the change in the way to think about it? I don't think that there's any change. The Surround ADAS is really imminent, but we don't want to start pinpointing which week or which month of 2025 this is going to be nominated. But all the indications are that we're at the end of the process of nomination. With the SuperVision, the same. All the activity we did in 2024 is bearing fruit, and things should start playing out within the next few months, within 2025. So nothing has changed that would change whatever we presented at the Capital Markets Day. Nimrod, do you have anything to add there? We just don't want to be driven by kind of expectations of an exact time, so nothing's changed on kind of our expectation for where we stand with different OEMs. Yeah. Yeah. Okay. Cool. And my follow-up, I appreciate the comments on when you compare to what the other OEMs are doing, what maybe some other tier ones are doing, that with the Mobileye product, that there's a focus on precision, and especially as it relates to eyes-off. But I guess my question is, as it relates to the Level 2+ and maybe some Level 3. I mean, you have obviously GM this week saying they're going to double their Super Cruise adoption this year. Tesla with their unsupervised FSD launch in Austin in June, Aptiv demo to Level 2+ camera and radar at CES. That's pretty comparable to FSD. And then Mercedes increased their speed to 95 km an hour for their Level 3 from 60. I know these products are probably still not up to snuff, right, where you guys are or where you want to be. But just wondering if there's a risk at the transition to get to eyes-off where these products might just be sufficient for where the consumer demand is. Is that a risk you see, or are you not concerned about that? Thanks. No. It's not a risk because we're doing both, right? We're building SuperVision, which is not an eyes-off system. And we're building Chauffeur at the same time. For example, with Porsche, we're building SuperVision. And with Audi, we're building Chauffeur. So it's all going simultaneously. We believe in the great value of an eyes-off, but regardless of our belief, we're also promoting very, very strongly SuperVision, which is an eyes-on system. So if an eyes-on system is what the market would think is good enough, it's fine. We believe that there's a great value in an eyes-off system. This is why we're pursuing that as well. And we're simply doing both. So there's no risk here. It's not that we are betting on an eyes-off and not doing an L2+ system. Thank you, Tom. Thank you. Ladies and gentlemen, that concludes our Q&A session. I'll turn the floor back to Mr. Galves for any final comments. Thanks a lot for everyone's time, and we will talk to you on our next earnings call in April. Thank you. Thank you. This concludes today's conference call. You may disconnect your lines at this time. Thank you for your participation.