Mobileye (MBLY) 2026年第三季度财报电话会议：随着EyeQ需求增长上调业绩指引

核心要点

• Mobileye公布2026财年第三季度营收为4.92亿美元，较上年同期的5.04亿美元下降约2%。EyeQ出货量增长3%至950万套，达到管理层指引区间的高端。
• 调整后营业利润同比增长20%至8900万美元。调整后营业利润率扩大约3个百分点至18%，这在很大程度上得益于来自以色列研发激励政策的5100万美元收益。
• 管理层将全年营收指引上调至19.95亿美元-20.20亿美元，中点约为20.08亿美元。调整后营业利润的中点上调至4.00亿美元。
• 管理层表示，EyeQ的需求继续超过汽车生产趋势。截至第三季度，出货量同比增长11%，而全球汽车产量则同比微降。
• Mobileye预计2026年SuperVision出货量约为65,000套，高于此前近60,000套的预期。管理层仍预计第四季度出货量将出现显著的环比下降。
• 公司强调，Surround ADAS（全景高级驾驶辅助系统）、云增强型ADAS和REM（精细路网图）应用率的提高，是提升单车价值量的潜在驱动力。大众汽车集团的量产项目仍定于2027年和2028年推出。

关键财务数据

今年迄今，Mobileye的营收增长8%，调整后营业利润增长42%，经营现金流超过3亿美元。公司表示，SuperVision出货量增加以及内存成本转嫁导致第三季度公司整体毛利率下降约100个基点。

业务与运营表现

EyeQ出货量增长仍是主要的营收驱动力。Mobileye预计2026全年EyeQ出货量中点将比2025年增长约10%。尽管前十大客户的整车产量合计下降5%，但Mobileye对这前十大客户的出货量预计仍将增长7%。

中国主机厂（OEM）贡献了超过100万套的出货量，为整体出货量增长贡献了约4个百分点。管理层指出，中国汽车制造商将Mobileye系统用于出口到拉丁美洲、南亚、东欧等市场，从而支撑了这一需求。然而，由于这些出货包含的软件内容较少，中国主机厂占比提升导致第三季度平均售价下降了约1美元。

截至第三季度，Mobileye报告SuperVision出货量近60,000套，已略高于其最初的全年预期。管理层将2026年的预测上调至约65,000套，但仍预计第四季度将出现大幅下降，这部分反映了此前为防止零部件短缺而进行的库存储备。

公司提出了提高单车ADAS价值量的两条途径。基于EyeQ6 High的Surround ADAS已被大众汽车、马恒达（Mahindra）和一家美国大型主机厂采用。此外，继此前宣布的Stellantis项目之后，又有两家汽车制造商（一家亚洲前五大汽车制造商和一家日本前十大主机厂）选择了对现有EyeQ6 Light项目进行支持REM的升级。

管理层表示，基础ADAS目前的定价约为47美元至50美元。Surround ADAS的平均售价大约是其两倍，而云增强型ADAS每年每辆车可能增加个位数的美元收入。SuperVision的单车平均售价在1,000美元以上，而Chauffeur的单车平均售价预计约为2,500美元，两者的毛利率均接近40%左右。

涉及基础ADAS、Surround ADAS、保时捷主导的SuperVision、奥迪主导的Chauffeur以及MOIA ID. Buzz的大众汽车集团项目正运行在量产硬件和量产意向车型上。管理层表示，软件加固仍是主要焦点，指定的平台计划于2027年和2028年在多款车型上推出。

在自动驾驶出租车（Robotaxi）方面，Mobileye计划于2027年在一个城市部署100辆车，并掌控车辆平台、自动驾驶系统、车队管理、客户应用程序、远程作业和后台功能。公司正优先考虑奥兰多地区，其次是迈阿密，同时与主机厂合作开发具备L4级自动驾驶能力的车辆平台，以便从2028年开始实现规模化扩展。

Mentee Robotics的3.2版本已投入运行，并计划在CES上展示。4.0版本预计将于2027年第一季度末前后打造完成，同时计划在2028年量产几百台机器人。

管理层指引

管理层预计第四季度EyeQ订单约为880万-890万套，SuperVision出货量将有所减少，营业费用将略高于第三季度的约2.25亿美元。

展望2027年，管理层预计EyeQ的平均售价将比2026年更为稳定，但受软件内容和打包组合的影响，仍有可能出现温和下降。预计Surround ADAS和云增强型ADAS的量产推出将在2028年起扮演更重要的角色。

风险与关注领域

• 中国主机厂及新兴市场出货量占比提高可能会给EyeQ的平均售价带来压力，因为这些项目包含的软件内容通常较少。
• 较高的内存成本正在转嫁给客户，但相关收入的毛利率为零。针对2027年的价格谈判仍在进行中。
• 预计SuperVision出货量将在第四季度显著下降，部分原因在于此前可能进行了提前拉货建库。
• 以色列研发激励的确认可能因符合条件的支出、汇率和会计门槛而有所不同。相关的现金收益预计将滞后于会计确认约两年时间。
• 大众汽车集团及其他主机厂仍在继续评估未来的软件及自动驾驶架构。Mobileye表示，这些评估不会影响其已获定点的量产项目，但未来的新定点仍取决于客户的决定。
• Robotaxi规模化扩展需要车辆制造、系统集成、车队运营、基础设施以及远程支持等多方面的协同配合。Mobileye最初预计每辆车配备一名远程操作员，随着时间的推移再逐步提高该比例。

分析师问答要点

大众汽车合作关系：管理层表示，现有的大众汽车集团量产项目仍按计划推进，预计于2027年和2028年推出。公司表示，CARIAD选择另一家自动驾驶合作伙伴涉及的是未来架构，不会改变Mobileye已获定点的项目。

中国出口机遇：Mobileye将其在出海中国车企中的地位归因于监管合规性和驾驶辅助性能。管理层表示，公司在多个决策周期中屡次获选，且不认为欧洲本地化制造会改变这一格局。

单车ADAS价值量提升：云增强型ADAS被定位为对现有EyeQ6 Light软件比重较高的升级方案。管理层将其描述为通往更广泛的Surround ADAS应用以及实现更高单车价值量的桥梁。

高级驾驶系统：Mobileye表示，相比展示车辆，公司优先开发量产级系统。管理层预计保时捷主导的SuperVision项目将在2027年初达到成熟，为未来L2++级和L3级的讨论树立量产标杆。

Robotaxi传感器成本：Mobileye计划将摄像头作为核心传感器，并使用成像雷达进行冗余。其长期目标是减少甚至可能消除激光雷达，但管理层未提供具体的时间表。

2027年投资：管理层预计将在GPU基础设施、Mentee Robotics和Robotaxi开发等领域进行部分投资，但表示增幅预计不会太大。2027年详细的支出指引将于日后公布。

业绩电话会议完整文字记录

---

完整财报电话会议逐字稿

管理层陈述

Operator

Greetings, and welcome to the Mobileye Third Quarter 2026 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.

Daniel Galves

Thanks, Sam. Hello, everyone, and welcome to Mobileye's third quarter 2026 earnings conference call for the period ending September 26, 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 EVP of Corporate Strategy. Thanks. And now I'll turn the call over to Amnon.

Amnon Shashua

Thank you, everyone, for joining. Mobileye delivered a strong third quarter with results driven by high demand for our EyeQ solutions. Looking at year-to-date trends, revenue was up 8% year-over-year, adjusted operating income up 42% and operating cash flow was over $300 million. We once again raised our full year guidance. The revenue outlook is now 3.5% above original expectations and adjusted operating income is about double the original outlook, supported by good cost discipline and the R&D incentive we discussed in detail last quarter.

The driving force on the revenue side is EyeQ shipment volume, which has exceeded expectations all year and significantly outpacing global production trends. While global automotive production through Q3 has declined slightly on a year-on-year basis, our shipments are up 11%. We expect full year 2026 EyeQ shipments to grow about 10% at the midpoint versus 2025. We're doing very well with our top 10 customers with volume expected to grow 7%, even as those OEMs' aggregate vehicle production declined 5%.

Chinese OEMs are also choosing Mobileye to support the rapid emerging market expansion, adding more than 1 million units and 4 percentage points to our overall growth rate. We expect emerging markets' ADAS penetration to be a key tailwind over the next few years. One relatively new tailwind is that Chinese OEMs' expansion into Latin America and South Asia with vehicles that come standard with ADAS, is driving legacy OEMs to increase ADAS fitment rates in these markets as well. Additionally, India remains a high volume growth market in the near future given new regulations and strong consumer demand for safety, as we have discussed in the past.

Looking towards 2027 through 2030, our growth opportunity within ADAS has two reinforcing drivers: ADAS reaching more vehicles; and Mobileye delivering more value in each vehicle. Rising fitment rates support unit growth, while broader adoption of our software and systems capabilities create a path to higher content per vehicle. That is a multiyear opportunity beyond this year's shipment performance.

In the next generation of high-volume ADAS systems launching over the next few years, technology and product features trends support higher Mobileye ASPs and a greater share of system cost as our core customers grow their reliance on Mobileye for intelligent safety and scale. Our customers appear to be pursuing higher function ADAS along two parallel paths. First, low-cost highway hands-free functionality targeted at high-volume mass market vehicle platforms.

Mobileye technology is part of the most first-generation L2+ systems on the road today. We actually have content on the majority of vehicles produced in North America and Europe in 2026, but often just as a front-camera perception supplier. As OEMs target lower cost and better hands-free availability in the second generation, Mobileye's role is expanding to include components like surround camera vision, crowdsource intelligence through REM, driving functions, driving monitoring and automated parking. Three OEMs have chosen our Surround ADAS solutions based on EyeQ6 High, including Volkswagen, Mahindra and the large U.S. OEMs.

The faster-to-market alternative to Surround ADAS is adding REM to an existing EyeQ Light -- EyeQ6 Light program to support Level 2+ functionality. We announced the first instance of this with Stellantis early in Q3. Since then, two additional OEMs have chosen this solution, including one top 5 automaker from Asia and another top 10 OEM from Japan. And the Stellantis program has expanded to include European platforms.

Taken together, these Surround ADAS and REM selection demonstrate adoption of our platform beyond front-facing camera perception. They show two routes to greater content per vehicle, supplying a broader system or adding software and mapping capabilities to an existing EyeQ program. As these programs enter production, that expanded growth with a path to higher ASP. The second path is the development of higher functionality point-to-point systems similar to Tesla FSD, that go beyond highways and arterial roads and supports hands-free driving on diverse road types, either supervised or unsupervised.

Our main thrust into this market is with the EyeQ6 High-based SuperVision and Chauffeur products expected to be launched with VW Group in 2027 and 2028. While SuperVision is nearer term and an important proof point, Chauffeur is particularly notable as there are few, if any, high-volume consumer Level 3 vehicles on the road today. The customer continues to see this product as an industry first and a critical unique selling point for their launch of a flagship vehicle in 2028.

REM, our road intelligence engines were outsourced from about 8 million vehicles, is worth a specific call out as market traction has really accelerated in the last 2 quarters, even into some of the multi-supplier point-to-point development programs announced recently.

Two points to make here. As OEMs begin to realize the challenge of building a good demo into a vehicle system that can be validated across millions of miles of road types and environment, they're realizing the value of data-driven road intelligence like REM. They're also realizing that REM can substantially improve availability of M3 Systems on various road types.

Secondly, Tesla recently called out their own efforts to use crowdsourced data to serve the same purpose as REM, something previously they claimed not to. Road intelligent databases like this cannot be duplicated without tremendous scale, which is why automakers are coming to Mobileye for that function.

Turning to robotaxi. Our self-driving system integration with various partners made meaningful progress this quarter with several milestones noted in the press release. It's encouraging to see these efforts move to rescaling milestones on our path to driver-out performance.

Beyond our capital-light self-driving system supply efforts, we continue to see value in building the capability to participate across the entire robotaxi value chain. Our rationale for a broader participation through our own vertically integrated solution is to accelerate deployment of our self-driving system and capture more of the value it enables, not to pursue fleet ownership for its own sake.

We have made significant progress in multiple areas. We have integrated movies, rider experience and fleet orchestration stack and initial testing is underway in Tel Aviv. We have multiple options for our vehicle platform and are currently negotiating an option that allows for rapid scaling after initial commercialization. Roles and responsibilities of the vehicle builder, the uplifter, Mobileye and operations are complex and need to be planned in parallel. We expect vehicle build to kick off at the beginning of 2026.

We see advantages in an expanded role in the value chain as it can lead to more rapid validation and software update cycle and enable us to make fast decisions on the hardware stack. For example, our own stack will rely heavily on our imaging radars as a redundancy to cameras, enabling us to reduce Lidar content in anticipation of the next cost reduction wave that can drive scale into the future.

Turning to Mentee Robotics, version 3.2 of the robot is now operational and performing well. The next version V4 is in the final stages of design with a production-ready architecture shipped by continuous feedback between AI development and hardware design. Mentee's differentiated approach combines the simulation-first development pipeline and advanced AI. Synthetic data, imitation learning and reinforcement learning enables the robot to develop agile humanlike movement and sophisticated perception and manipulation capabilities. Our simulation environment continuously generates challenging scenarios, allowing the team to train and improve the robots in situations that are difficult to reproduce using real-world data alone.

At the task level, an Agentic AI architecture combines vision language models with specialized robot policies to break down complex tasks, execute them as sequences of skills and verify successful completion.

Overall, Mobileye's foundation remains solid. Demand for our technology is outpacing production at our top 10 customers, while Chinese OEMs continue to choose our solutions for export market. And over the course of 2026, we have seen a meaningful uptick in traction for advanced Level 2+ systems that indicate higher ADAS content per vehicle in the future.

I now turn the call over to Moran.

Moran Rojansky

Thanks, Amnon. Before I begin, please be aware that all my comments on profitability will refer to non-GAAP measurements. The exclusions in Mobileye's 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, including the partial offsetting impact of the R&D incentive we introduced with our second quarter results in July. For 2026, as discussed previously, we are also excluding the goodwill impairment loss that occurred in the first quarter and transaction costs associated with closing the Mentee acquisition, which was also a Q1 event.

Third quarter revenue of $492 million compared to $504 million in Q3 2025. This decline of approximately 2% was better than the 5% to 2% decline we had guided to. Volume was up 3% year-over-year in Q3, but with a higher mix of Chinese OEM shipments, which carry less software content and therefore, lower pricing. Chinese OEM mix alone drove year-over-year ASP down about $1. Two factors drove these modestly better results. The first was volume. At 9.5 million EyeQ units, we landed at the high end of our 9.2 million to 9.5 million unit outlook, extending a trend we've seen all year of EyeQ demand running a bit ahead of what we expected coming into 2026.

Three things are behind us. One, our technology is being adopted faster than our top 10 customers are growing their production. Two, Chinese automakers are preferring Mobileye to support their push into export markets, especially in non-Western markets like Latin America, South Asia and Eastern Europe; and three, that Chinese OEMs push into emerging markets, with standartisized (sic) [ standardized ] ADAS in encouraging legacy OEMs to increase the penetration of ADAS in their vehicles.

Second, deliveries of EyeQ5 High-based SuperVision systems of more than 15,000 units were above our expectations. Through Q3, 2026 deliveries totaled nearly 60,000 units, already slightly exceeding our original expectation for full year 2026 delivery. We still expect a notable step down in SuperVision deliveries in Q4, but we are modestly raising our full year SuperVision delivery outlook given the strength in Q3. More on that in a few moments.

One additional note is that higher memory costs have increased cost per unit, which we have passed through to the customer on a one-to-one basis. This pass-through added approximately $7 million of revenue in Q3 at 0 gross margin. The higher SuperVision volume and pass-through memory costs together reduced our corporate average gross margin by about 100 basis points.

Adjusted operating income in third quarter was $89 million, up 20% year-over-year. Adjusted operating margin was 18%, up about 3 percentage points versus Q3 2025. The improvement was again largely driven by the Israeli R&D credit loss, which contributed $51 million of contra R&D expense in the quarter.

As a reminder, the law came into effect shortly after the end of Q1 and is retroactive to the beginning of 2026. Q3 is the first quarter that reflects the benefit on a clean run rate basis. I want to reiterate a few points on the R&D incentives that we shared last quarter. We continue to expect the benefit to extend into future years, and there is no end date to this regulation, although it remains subject to future changes in law. Quarterly recognition can be somewhat volatile depending on meeting recognition threshold, the level of qualifying R&D spend in Israel and exchange rate. The cash benefit also lags the accounting recognition with the cash impact of the benefit recognized in 2026 expected to arrive roughly 2 years later beginning in 2028.

Turning to full year guidance. We are raising the midpoint of our revenue outlook by $13 million to approximately $2.008 billion and narrowing the range to $1.995 billion to $2.02 billion. This implies growth of 5% to 7% for the year. The increase primarily reflects the SuperVision upside I mentioned, as we now expect about 65,000 systems for the year, up from almost 60,000 previously, with ASP in Q3 and Q4 meaningfully higher than prior quarters due to the pass-through memory costs noted above. We continue to believe that there has been some intentional inventory building to protect against component shortages. And so we still forecast a step down in Q4 volume.

Our EyeQ outlook is unchanged. We are raising the midpoint of our adjusted operating income outlook to $400 million from $395 million previously and narrowing the range to $380 million to $420 million. The conversion of incremental revenue to adjusted operating income is consistent with our expectations as the increase is SuperVision related and that product carries lower gross margin than our corporate average.

Our outlook continues to include approximately $190 million from the R&D incentives at the midpoint. We still expect full year non-GAAP operating expenses of approximately $910 million at the midpoint, which includes $180 million to $200 million of R&D credits. I'd note that we are keeping a fairly wide range here to incorporate some modest variability from further R&D incentive guidance expected in the next few weeks. At the midpoint, our annual guidance implies Q4 revenue of $450 million, down about 8.5% from Q2 and adjusted operating income of roughly $60 million compared with $89 million in Q3. This is consistent with our expectations as of the July call.

On the last 12-month basis, including Q4, we expect EyeQ shipment volume to grow about 10% year-over-year. That is the more meaningful measure of the volume trend as shipments can vary widely between quarters. The sequential step down in Q4 revenue and adjusted operating income outlook relative to Q3 reflects three factors: EyeQ orders currently pointing out roughly $8.8 million to $8.9 million compared with $9.5 million in Q2; the lower SuperVision shipments I just mentioned; and operating expenses expected to be a bit higher than approximately $225 million in Q3.

To conclude, our third quarter results reflect strong execution and continued financial discipline. We continue to produce sustained cash flows from our core business with operating cash flow of $303 million in the first 9 months. And our balance sheet remained strong with $1.5 billion of cash, cash equivalents, marketable securities and deposits.

Finally, we have repurchased approximately $33 million of shares since announcing the buyback program in April, an average purchase price of $9.36. Our buyback strategy remains subject to market conditions and our capital allocation priorities.

Thank you. Operator, you may now open the line for questions.

Operator

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

分析师问答

Chris McNally

Maybe you could just comment on the status of the VW relationship? I know you made some prepared comments, but it seems sort of top of mind for investors. And I think VW has multiple points -- multiple parts, basically, we see it divided as production car decisions and then their internal or longer term efforts for autonomous. But I think we've all read some of the German articles out there about CARIAD taking on new partners, Wayve in, maybe Bosch out as well as it appears MOIA strategically may not be going as financially planned. So any sort of material update you can give on both the production launches, the three big ones over '27, '28, but really maybe more on the robo relationship would be super helpful.

Amnon Shashua

Okay. Thanks, Chris. It's a big question. Our relationship with the VW Group is defined by -- first by our shared rich history of product launches and then by the nominated production programs we're executing together. So let's go over them, the Base ADAS and Surround ADAS with VW brands, we have SuperVision led by Porsche. We have Chauffeur led by Audi, and we have the ID. Buzz program with MOIA.

Now all programs are now running on production hardware, production intent vehicle platforms. Software hardening is the primary focus and the designated vehicle platforms are intended to launch in the 2027, 2028, multiple vehicle models. So all of that is tracking well.

Now I would say that OEMs continuously evaluate future architecture. That has always been the case. And it's really a separate matter from the programs we're delivering. Now we cannot comment on partners' internal processes or media reports about them, but CARIAD was always there. Even when we announced back in 2022, 2023, our programs with Porsche and Audi, part of the announcement by the Volkswagen Group was that there is CARIAD and CARIAD is part of the plans for future of autonomy. And CARIAD is undergoing a phase of replacing their partner. It used to be Bosch, and now it's going to be someone else. That someone else cannot be mobilized because that's the whole point of CARIAD, to be a second source for Volkswagen Group's ambitions in going into autonomy. So that hasn't changed.

As for the program with MOIA, look, our focus is executing and maturing the technology to drive out on the roughly 100 ID. Buzz prototypes that are deployed in multiple geographical locations and complete the preparations to be driver-out ready. The AI stack and hardware are tracking well. We are -- we continue to see steady progress in data accumulation for validation and key performance metrics. So, all of that is tracking well. Beyond that, we cannot comment on speculations and reports concerning our partners.

Chris McNally

Sure, that's super helpful. And is it fair to just always do the paraphrase that basically, if CARIAD is continuing, right -- it shrank over time. But basically, anything CARIAD works on internal for VW would be for program production awards in coming years, which would be more of your launches in the '30 -- yes, 2030 type time frame. Is that fair? Meaning they would have to win it on their own merits, but obviously, that's something that's a little bit longer dated?

Amnon Shashua

That's correct. It does not reflect on the programs we have. Programs we have are slated for production. We are working on those production programs with production vehicles, hardening software, as I said. And then there is kind of the normal nominations of future architectures and CARIAD was always in the picture. I mean nothing new here. Only they're selecting a new partner.

Chris McNally

[indiscernible]

Daniel Galves

Go ahead, Chris.

Chris McNally

No, yes. Just on the AV side, on the dual track go-it-alone strategy, is it fair to say that you would look for partners to do the physical infrastructure side, the depot, the own the vehicles, and there's a host of players doing that for other AV players. But when you're going it alone on the vehicle side, that you would have someone on the physical infrastructure side?

Unknown Executive

Yes. Chris, yes, the answer is yes, and we are working with partners for, let's say, the operational, the infrastructure, the day-to-day management, warehouses, parking lots, all the low level and mid-level maintenance work that relates to the fleet themselves. Right now, our focus is on the -- as I said it in the script, that basically to put the vehicle platform with our self-driving system as a ready platform in 2027 and then starting in one city and then gradually over the course of the next few years scaling that as the entire ecosystem matures, I would say.

Operator

Our next question is from Dan Levy with Barclays.

Dan Levy

I wanted to go back to -- I know we've talked in the past about the China dynamics. And I think we've been surprised by how well you've been holding share. And I know that some of your competitors in China are making the claims that they do have ways to export their products through various JVs. And so, can you just talk to -- as we continue to see this rise of China exports and competitors like Horizon and Momenta, that are going to make the claims that they can play on those exports as well, why you think you can continue to hold a good share on the exports?

Unknown Executive

Right. So I don't want to speculate too much in the future, but these claims are not new. I think we've been facing these claims by the Chinese competitors at least for 4 years now. And one of the reasons why we are -- we remain a steady and healthy market share with the Chinese OEMs that are exporting, of course, regulation compliance, but that's not the only reason. I think that there is also a standard of performance that the European -- for example, the European consumer is expecting when it comes to driver assist systems. And by and large, most of the driver assist features that are available in Europe by different OEMs are provided by Mobileye.

And I think that for the Chinese OEMs, as they create a market share for themselves in Europe as their next big export market, it is very important for them to come with the most credible, high-performing driver assist offering to the European consumer. So that's basically what we've seen. And that's not true just for Europe. I think it's also affecting other markets, although the U.S. market maybe for the Chinese OEMs is not the primary focus today. But even Asian markets outside of China and Asia Pacific markets, South America, which is also coming up with new regulation, and India, and also in Europe.

I think -- again, to summarize it, I think that we are confident because of the results so far, and we've had multiple decision-making cycles by now by the big Chinese OEMs, and we have been selected repeatedly in those. So I think we have a lot of merits.

Dan Levy

Great. Second question is, perhaps you can talk to the two Cloud-Enhanced wins, which I know you said those follow the Stellantis wins. And I know your commentary gave -- it implies that this should have a [indiscernible] ASP accretion. But maybe you can just talk to how wide an adoption we see of those? And are those potentially gateways to getting further advanced content wins with these automakers?

Unknown Executive

Yes. So the way I think to look at this is this is an upgrade for existing EyeQ6 like products with these OEMs. So basically, they've launched EyeQ6 like based ADAS products. And then once those became mature and kind of proven in production, there was an incremental upgrade with Cloud-Enhanced ADAS, which is a more, I would say, straightforward execution project as it basically uses the same chip, it uses a lot of the same hardware and it's mostly software validation work. And they can offer better features, better performance on these features at a very affordable cost.

So I think for those OEMs, it's a pretty sensible decision. And definitely, it improves our long-term relationship with these OEMs, and it helps us to add -- bring new technologies to these OEMs and kind of show what else we can do with these technologies and the next logical step is the Surround ADAS with EyeQ6 High and so on. So it is a kind of building a bridge.

Daniel Galves

Next question, please.

Operator

Our next question is from Edison Yu with Deutsche Bank.

Xin Yu

First I want to follow up on the internal robotaxi efforts. I think the idea here is obviously you have a lot more control over what's going on. So do you have a sense or can you provide a sense on kind of the big milestones for next year? And also, what do you envision as the form factor for that robotaxi because we've obviously seen lots of different vehicles now on the road, even in places like L.A.?

Amnon Shashua

Our plan is to get better control of the life cycle of the product, first, better control definition of the product, better control of updating the software, selecting the hardware, by owning the full stack gives us control to move faster and make decisions faster. I'll give you an example. One of the things that we are looking -- kind of future-facing is considerable cost reduction of the robotaxi platform.

Today, there is kind of a standard of how a robotaxi looks like in terms of the number of Lidars and radars and cameras. It's quite expensive. But when we're looking into the future, say, 2, 3 years from now, we believe that there is room for a considerable cost reduction, and we want to use our robotaxi activity to experiment there. We invested a lot in imaging radars. We think we have the best imaging radars today on the planet in terms of the resolution, the dynamic range, the site log detection and thinking of ways to reduce the sensor set and the compute cost build to create really cost-optimized robotaxis. So this is something that if we don't control the entire product cycle, it will be very difficult to have those experiments. So that was kind of the background.

Now, plan for 2027 is to deploy 100 vehicles, and have -- to be able to control the full cycle, not only the self-driving system, but also the vehicle platform, the uplifting of the system onto the vehicles, the operations, the fleet management system, the customer-facing application, the teleoperation, the back office activity, all of that on 100 vehicles in one city in 2027. We are thinking of the Orlando area, Miami next. So this is what is planned for 2027.

In parallel, we are working with a number of OEMs to define the at the Level 4 ready vehicle platform that we can buy at scale for expansion from 2028 and beyond.

Xin Yu

And just a follow-up on Mentee. I know you mentioned you're in the final stages of the design. I guess, could we expect this to be ready by CES? And kind of somewhat related to that, I'm curious that there was an acquisition made by AMD of World Labs. Any just like -- any thoughts on kind of the importance of something like that?

Amnon Shashua

Again, what did you mention about World Labs?

Xin Yu

Just any thoughts on just the need to have something like that in-house or the importance of this world model or spatial model?

Amnon Shashua

No, no. We are developing world models. That's part of our stack. We plan to bring the 3.2 on stage for the CES and it will perform some interesting end-to-end tasks, not just locomotion, but actual useful tasks. So we plan to bring that to the CES. But the version 4 should be built around the end of Q1 2027, more or less. And the plan for kind of building a few hundreds of the robots for production is planned for 2028.

Daniel Galves

Next question please.

Operator

Our next question is from Joseph Spak with UBS.

Joseph Spak

I wanted to circle a little bit back to Chris's question, but maybe broaden it out a little bit from the VW Group because, look, I know you've always sort of talked about the view that customers may seek multiple or internal solutions. And as you sort of indicated, I don't think this really sort of changes anything about your midterm opportunity. But it does feel, at least from the outside, like there's been a little bit of a sea change in how OEMs are thinking about solutions for beyond and sort of the next generation.

So maybe you could just sort of give us a little bit of the lay of the land and some of the conversations you're having with customers for the next generations and how that could impact Mobileye positively or negatively beyond the current awards?

Amnon Shashua

Thank you for your question. I would say that our strategy over this period has been very deliberate. One can argue whether it was the right strategy or not, but this is our strategy. We put all our resources and focus into bringing four new platforms to production simultaneously with our leading customer, that's Volkswagen Group, and that instead of building a demoware and to use it for business development like some of our competitors are doing. So when you have demoware, you can move fast, you can show things that are very impressive. But then there is the uncertainty of going from demo to production, that people usually don't put sufficient weight or importance to do it.

So we believe that scaling a proven production-grade product is a stronger position than productizing across many customers let's say at once. And this trade-off is a fewer new nominations in the near term and more durable, sustainable growth in the long-term. And we are reaching the point in which at least our first production, the SuperVision with Porsche, is coming to maturity somewhere early 2027, that would put us in a better position of demonstrating already production worthy system rather than something that is half baked.

Now we could have focused on demoware, and I'm sure we would have done very impressive demos. But we put our focus on production, production programs. And when we are working on production, it's not that easy to go and demonstrate because in a production, not everything is under your control. Maturity of other sensors are not in your control. The application layer is sometimes not in your control. There are all sorts of requirements from the carmakers that some of them are coming from other suppliers. So being able to demonstrate the best performance of our system when we are in the midst of production, is very problematic, and we didn't put resources on demoware. But I believe that period is coming to an end.

And during 2027, we can be back on to doing the business development on the advanced programs by showing real production systems, and that would put us in a good position on -- in the conversations of the next advanced Level 2++ programs and Level 3 programs that OEMs are discussing.

Joseph Spak

Thank you for that answer. As a second question, maybe just something on the full year guidance. I know it was raised essentially by what you beat expectations for in the third quarter. But there was also -- you did have the 7 million memory pass-through. I'm assuming it's probably at least a similar number in the fourth quarter. So I guess, does that mean there's something underlying has changed? I didn't -- I don't -- maybe I missed it, I didn't hear sort of the full year unit number. Or was memory sort of always considered in your prior outlook? And can you just clarify that?

Moran Rojansky

Yes, I'll answer that. So memory was considered in our prior outlook. So out of -- the $7 million is just the total memory cost that we have had in Q3, but probably I assume half of it or 60% of it was already baked into the forecast. So that's one thing.

Second, on the fourth quarter, we cannot expect to have the same memory cost into our forecast because the SuperVision volume is much lower in Q4. So we said we sold more than 15,000 units in Q3, and we're expected to be at -- so currently, we are almost 60,000. We expect to be 65,000. So that is not a significant upside.

And on EyeQ, we are on track with volume. So in Q3, we were on track with volumes and also Q4, we didn't change our guidance for EyeQ.

Daniel Galves

Next question, please.

Operator

Our next question is from Shreyas Patil with Wolfe Research.

Shreyas Patil

Maybe picking up on the last question. I'm just curious how we think about memory price and the pass-through as we go into next year. And memory prices industry-wide are expected to increase by even more in '27, and you are expecting a bigger step-up in SuperVision shipments as well. So just curious how we should think about that at a high level?

Amnon Shashua

Yes. So we are in the process of working with our customers on the prices for 2027. That process is still in the midst of it. I think we'll be prepared to give the full picture at the next earnings call. But in general, as I said, this is an industry-wide topic and it's not Mobileye specific. And we, in the past, have had successful occasions of working with our customers on these macro-driven cost increases, and we take a similar approach this time.

Shreyas Patil

Okay. And then maybe just to put a finer point on an earlier question. Your exports with the Chinese OEs are primarily going to rest of world markets, I believe, across Asia, Middle East, Latin America, not so much Europe. And so -- if so, I mean, how do we think about that as European OEMs are ceding market share in Europe? And those Chinese automakers that are exporting or eventually will localize in region, are they deploying their own locally developed ADAS and autonomy systems? Or are there regulatory challenges to doing so?

Amnon Shashua

Yes. I think as we commented before, I think that as of today, Indian market, that the -- I think this is where the Mobileye solution is being adopted. And specifically, we -- in the European market, the regulation is very strict and comprehensive and Mobileye's products have been always known to achieve the highest scoring in these regulatory tests. I think that's part of the reason why the Chinese OEMs are adopting Mobileye products, and we do not expect them to change that just because of local manufacturing. If anything, it will help us to improve our competitiveness further in the European market.

Shreyas Patil

Okay. And maybe just a quick last one.

Daniel Galves

Yes, go ahead.

Shreyas Patil

I was just going to ask on Q4, the EyeQ shipments, if there's a way to break out kind of the split between how you're seeing Western -- like your international shipments to international OEMs versus the Chinese OEs just given the seasonal uptick in Chinese volume in Q4?

Moran Rojansky

Yes. So we don't give specific volume, but we can say that the China OEM demand has pretty -- has been sustained through the year. So Q2 and Q3 were a bit higher. But overall, we are delivering more than 4.5 million chips in China, and that is roughly equivalent between the quarters Q2 and Q3 were a bit higher, but not a significant drop from China OEMs in Q4. We're seeing sustained demand.

Operator

Our next question is from Mark Delaney with Goldman Sachs.

Mark Delaney

I was hoping to better contextualize the company's growing opportunity around an ASP uplift for L2+ vehicles. And with Mobileye's outlook for -- to be about a 20% CAGR for L2+ vehicles through the end of the decade and that volume outlook, coupled with the awards you've had in Surround ADAS and the growing momentum you're speaking about now in REM, can you help us understand what that means for where Mobileye's ASP per EyeQ chip may trend over the next 2 to 3 years?

Moran Rojansky

Yes. So I can say for this year, we've obviously experienced some headwinds in ASP coming from China OEMs and emerging markets second chip programs that we already mentioned. But we really -- there are two factors that impact ASP. So one is the geographical area, emerging market, China OEMs with lower software content and lower ASP. And on the other hand, you have the Cloud-Enhanced ADAS and also advanced programs in general. So programs with bundles with higher software. We are even seeing, for example in Q4 some expected increase in EyeQ ASP due to more demand for higher bundles or higher software. And with Cloud-Enhanced ADAS plus REM, that obviously can add a few tens of dollars there per EyeQ depends -- of course, on the volume depends on the take rate, but it might give a [indiscernible] to ASP that will offset some of the -- at least offset -- more than offset some of other impacts that we have.

Unknown Executive

Yes. To also maybe add on the longer term, Mark, on the longer term question. So today, ASP is primarily a Base ADAS business, and it mostly depends on the mix of bundles and features. And it's roughly around the $47 to $50 range, as it's been in recent quarters. In the next, let's say, 2 to 3 years when we launch our products and gradually, you'll see a different mix of products across the Mobileye revenue stream. We have to run ADAS with roughly twice or 2x the ASP with similar gross margin. Cloud-Enhanced ADAS programs contribute roughly single digits in dollars per car per year. In some cases, it can be kind of a lump sum of -- for multiple years in day 1. In some cases, it can be kind of a yearly license. SuperVision is in the above $1,000 per car, but with a different gross margin, maybe closer to the 40s. And Chauffeur is in the $2,500 ASP, again, with a closer to the 40s gross margin.

Now the overall average -- compounded average ASP is a question of how many volumes we'll have in a given year of these products and all of them are expected to be launched in production programs with meaningful volumes. But for us now to predict exactly how many volumes per year, it's a little bit early. But we do expect overall to see a very -- kind of a positive tailwind for ASP as we launch more and more of these products because again, just for example, the Surround ADAS program with VW is expected to be adopted on a double-digit number of vehicle models that sell in millions per year. So when it saturated into kind of peak volumes, it's a very significant revenue increase and ASP increase just from that program alone. And as Amnon said, we have three of those with a similar meaningfulness, I would say. Hope for -- I hope I helped you kind of model.

Daniel Galves

Yes. And this is Dan. I just wanted to make a couple of final points on -- because it's an important question. So the L2+ segment is the fastest-growing segment in ADAS in North America and Europe over the next several years based on Mobility Global projections, and these projections have really been rising. So it's a very kind of strong segment that today, we have a good presence, but with essentially normal Base ADAS products as part of a system. These wins that we've had over the last year indicate that our content and our positioning in the next generation of L2+ is growing.

And I think we wanted to make the point that if you look specifically at single-chip type of ASPs, which have been kind of slowly eroding the last couple of years for reasons that Moran mentioned, we have a good opportunity to turn that around with Surround ADAS at 2 to 3x the ASP of Base ADAS and Cloud-Enhanced ADAS that has a gross profit per unit that is roughly similar to Surround ADAS. Next question please.

Operator

Our next question is from George Gianarikas from Canaccord Genuity.

George Gianarikas

You mentioned you decided not to focus on demoware, but I'm really curious as to whether or not with Tesla's recent FSD milestone announcements, 75% take rates in North America, have you observed any acceleration in customer inbound and interest in adopting advanced autonomous solutions?

Amnon Shashua

Well, I'll tell you what, when we do internal testing and driving expedition with the current program that we have with SuperVision, which is SSD like, the performance is super, super good. I believe that when we come out in production, it will not be that far from a Tesla FSD performance, without the parking, we're not doing the parking there.

So I believe that when the time will come that we can really demo that vehicle, when we finish hardening the software, doing all the -- burning down all the tickets and maturing the sensor maturity and everything. I think we'll be in a very, very good position. And Tesla's very high take rate of 75% is a very, very good tailwind that will enable us when the time comes to get more attention from OEMs who see this very, very big take rate of Tesla.

George Gianarikas

And maybe as a follow-up, as you start to launch your robotaxi fleet next year, how are you modeling and thinking about network operating center headcount and vehicle-to-operator ratios? I'm just sort of curious as to how that should start and how you expect that potentially to come down over in the next several years?

Amnon Shashua

It always starts with one operator for one vehicle at the beginning. Then later, it starts reducing. We talked about lasting certain technologies of VLMs to enable to kind of reduce the attention of retail operator to allow more vehicles for the retail operator. And when the time comes, we'll integrate it into our stack.

Daniel Galves

Next question, please.

Operator

Our next question is from Colin Rusch with Oppenheimer & Co.

Colin Rusch

Can you talk a little bit about the thought process in making the investment in Beep and how you see the public transport market evolving as a demand creator for some of your self-owned vehicles?

Amnon Shashua

Yes. So the Beep investment is basically around establishing a partner for the robotaxi operations and the day-to-day management and maintenance. It goes back to George's question about the headcount for remote operators. So basically, we've been working with Beep for multiple years now, and we've developed a common approach and a collaboration on our technologies with their platforms that are very effective to manage robotaxi fleets. And they will be our partner for the next few years when we launch our robotaxi fleet in the U.S.

At the same time, there is also a demand creation part of it that have a very rich activity around public transit and micro transit, which is a very interesting angle for autonomous mobility use cases. That is something that we're evaluating in parallel with them.

Colin Rusch

Great. And then in terms of the evolution of the sensor suite and the cost reduction, as you look at the introduction of colored Lidar into the market, is there a real opportunity for you guys to evolve the sensor suite and simplify it with that sort of technology out in the market?

Amnon Shashua

So our bet is on -- first, the basic is always cameras. But on top of cameras, our bet is on imaging radars, and we invested for the past 7 years on developing very capable imaging radars that can take the heavy weight of redundancy. So our belief that long-term, you would not need the Lidars. It's only cameras and imaging radars. That's our bet. How -- the time horizon for that is still unknown, but that is our bet.

Operator

Our last question will be from Andrew Percoco with Morgan Stanley.

Andrew Percoco

Just two quick follow-up questions, and I'll just ask them at the same time. So first on the ASP commentary, super helpful. But if I could just kind of double-click around kind of the cadence of that -- it sounds like a lot of the Surround and Cloud-Enhanced ADAS is more of a 2028 story in terms of when production actually starts to kind of kick in and drive meaningful volumes. So should we generally expect 2027 ASPs to kind of follow the trend that we've seen in 2026 before you get that benefit, obviously, in 2028 and 2029 on some of the higher-end Surround and Cloud-Enhanced solutions?

And then my second question is just around investment needs for 2027. Obviously, moving ahead with Mentee and your own robotaxi business, do you expect a notable uptick in R&D or more broadly OpEx as we get into 2027 to support some of those development goals?

Unknown Executive

So on the ASP side, I'll talk about the product launches and maybe Moran can talk more specifically about next year. So you're correct that the Surround ADAS and Cloud-Enhanced product launches are 2028 events, and on '27, Moran?

Moran Rojansky

Yes. So 2027, you need to bear in mind that this year, we had, as I mentioned, headwind for ASP for China OEMs and also the second chip that we didn't sell in 2025 and we're selling in 2026. So we had roughly, I think, almost $2 ASP decrease in 2026.

So 2027, we expect to be more flat. There might be some moderate decrease that relates to, again, software content and demand for higher bundles versus lower bundles, but we are not expecting to see the same headwinds as we experienced in 2026.

Daniel Galves

And I think the second question was -- I think there was a question around investments into 2027.

Amnon Shashua

So, there are going to be some investments in 2027, but we're not ready to talk about them at this point, but it's not something major. Planning to buy GPU farms, all sorts of investments, but we will talk about them when the time comes to talk about 2027.

Operator

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

Daniel Galves

Thank you, Sam. Thanks, everyone, for joining our earnings call. We will have another one in January. Thank you so much. Talk to you soon.

Operator

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