Latest / Autonomy Insiders / Autonomous Driving Chips: Tesla, Xpeng, Rivian and the In-House Revolution

Transcript

Autonomous Driving Chips: Tesla, Xpeng, Rivian and the In-House Revolution · Autonomy Insiders

• Daniel Abreu Marques: Welcome to Autonomy Insiders. I'm your host Daniel, and my guest today is Augustin Friedel Associated Partner at MHP. And Augustin is an expert in software-defined vehicles, advanced driver assistance system, and AI-enabled mobility. He works closely with OEMs and suppliers, guiding them through the complex transformation processes towards the mobility of tomorrow. Today, we're diving into a hot and highly relevant topic. It's all about chips today. Today with Augustin Friedel Welcome to Autonomy Insiders, the show where global industry leaders in autonomous driving unpack their real-world insights. I'm really happy to have you on the show. Welcome. Augustin Friedel: Thanks, Daniel. Glad to be here and let's make it an exciting podcast. Daniel Abreu Marques: Absolutely, looking forward to it. So maybe to kick things off, can you explain briefly what an SOC is, in simple terms, and why SOCs are important for modern vehicles like autonomous driving vehicles? Augustin Friedel: Yeah, sure. So first, SOC stands for system on a chip. And so that's already what it is. So ⁓ it's a whole computer on one chip. So it's combining ⁓ CPU, it's combining the GPU, so the parallel processor that's needed then for like heavy workloads that we know from autonomous driving systems and also for the AI. And we have to know also AI accelerators. often also memory interfaces and so on, like everything packed into ⁓ ⁓ or like one package instead of having there like all the different modules separate, which was basically the past and automotive or also in your computer. So you might remember that you had your separate CPU and you had your separate memory ⁓ and you had like all the little cards ⁓ in computer that you could be able to put in and put out. ⁓ And basically with the SOC, we put everything on one chip and pack it in different domain controllers and they're empowering the ADAS systems, autonomous driving systems, also smart cockpit solutions and modern vehicles. Daniel Abreu Marques: Thank you for the explanation. Why are these high-performance chips so critical for self-driving cars? Augustin Friedel: I mean, have like the modern cars and especially if you talk about autonomous or also ADAS system, we have a huge data or a huge amount of data that need to be processed ⁓ more or in real time to make the like prediction than the planning the vehicle, the control of the vehicle. ⁓ And there basically have, we are talking about ⁓ up to 10 even more ⁓ high ⁓ definition We talk about LiDAR signals, radar signals, and then other sensors that send data to this high-performance compute system or SOC. And then basically we need this compute capacity to have the different steps that are needed to allow then safe autonomous driving or automated driving if you talk about ADAS levels like level 2 +, level 2++, or level 3. Daniel Abreu Marques: And then if you're different chip options for autonomous driving tasks, are the most important criteria ⁓ to choose the right chip? So is it just the raw performance or are any other considerations like energy efficiency or something else? Augustin Friedel: Yeah, it's like there are multiple factors as you already teased Daniel. So it's quite interesting when we have the discussions at MHP with our customers around the world. It's often, okay, I need to chip with the highest tops and okay, how could I increase like the tops in my vehicles? And sometimes you also had it that basically Chinese OEMs added more tops to just like pace out their competitors. If you talk about an Nvidia Orin chip, for example, I think they have around 250 tops. And then we had like the phase in the industry where then OEM had maybe like three of them, so 750 tops. And then like the other one said, okay, I need four of them just to be better than my competitor. But if are then ⁓ like really smart decisions, we need to take more decision factors or like KPIs into account. ⁓ If you talk about vehicles, often have a very and small system. So need energy efficiency and need think about ⁓ thermal as well to avoid overheating. then, in the worst this could destroy the chip or then throttle the chip or lose capacity. So that's definitely one important KPI to look after, how efficient the chip is. What are the thermal factors of the chip? What is the thermal performance to avoid and also that there's a huge battery drain. we talked about ADAS solutions as well, I think also super important to ⁓ consider bandwidth and input and output signals and processing because as mentioned, the sensors that we have that we for autonomous driving or also automated driving on level three or level two plus, plus, plus, they're like very data hungry. So they need a huge bandwidth of like memory and so on ⁓ yeah, to have like then like a smooth system. Then of course you also need to consider security aspects. ⁓ So you might have heard of like ASIL certificates or certifications. Yeah, so basically then a tool to see how safe and secure one chip is. And that's also something that needs to be taken into account. So that's more if you talk about the chip and the SOC. And then most important then also, and that's often ⁓ ⁓ the discussions we see in the industry is like the whole tool chain development environment and then also like ecosystem ⁓ So they're basically all like linked to compilers to ⁓ model optimization, and then like the whole software layer on top of the chips and then also like off board basically for then the development of the solution. So that's often not considered when making like chip decisions or not like on highest priority, but that's basically the make or break. Daniel Abreu Marques: Well, thank you for this explanation. And you also briefly touched on Nvidia already and who are the major players providing these computing platforms for autonomous driving besides Nvidia. Augustin Friedel: Yeah. Yeah, so there we have, I would say, of different players and ⁓ need to differentiate there between the different regions of the world ⁓ then also ⁓ the different automation or ⁓ levels ⁓ linked the SAE's like Scala. So ⁓ there, ⁓ I ⁓ we there are differentiate between ⁓ level level three, then between level two plus, level two plus plus and... and level four, yeah, and for each of the different, let's say trim levels, you have different SOC or like chip providers with like different performance criteria, pricing criteria, local presence, et cetera. So if we talk about there about different names, of course it's Nvidia. It's then like a Mobileye from Israel. I think they belong now to Intel and also like public traded company. They have like their full stack including and also the sensors and the software. We have then also from the US Qualcomm. But then also if you look the direction to China, that's getting super interesting because there we have then also a mix of ⁓ Western players. So Nvidia was also often the go-to source for autonomous and also like ⁓ ADAS companies China ⁓ to kickstart have a fast time to market. But what we then there is ⁓ like let's say a fork in two directions. One is that OEMs or then also like tech players build their own silicons or at least design it then tailored to their needs. And then the second one, second direction is that we see like local players like Black Sesame or Horizon Robotics and others basically that build like the local ⁓ chip SOC industry for ADAS solutions for autonomous driving. but then also for then like IVIs or infotainment. ⁓ one of the biggest topics now is to know, okay, how can you reach more volume with like robotics solutions? Daniel Abreu Marques: when look at the different tiers, so to speak, like the difference from an ⁓ L2 chip to an L4 ⁓ chip, it then mainly the ⁓ performance or tops, or what is ⁓ the difference those two categories? Augustin Friedel: Yeah, so it's a mix of then like the tops. Yeah. So you see then often, if you talk about level four solutions, you see ⁓ tops the range of like 1000 to 2000 tops. So that's definitely drawn. But then ⁓ like the and the bandwidth or memory bandwidth because then okay, with like higher automation levels, like level four, you often have like more sensors, like more cameras, LIDARs more radars ⁓ or then like a more sophisticated sensor suite. So you also need a higher memory bandwidth to then be able to manage all the different data signals. And then of course, ⁓ then like the tool chain and software layers that are needed to make it like a whole package ⁓ for the development part, but then also like for the application softwares in the vehicles, in the devices. Daniel Abreu Marques: And if we're staying in the level 4 category, do you have like rough price indications how much costs such chips, maybe also between the different suppliers? Do you have some rough estimates? Augustin Friedel: think it's, ⁓ don't have like numbers. ⁓ can't share numbers. That also depends on like the, I would say the negotiation power of like the customer of the OEM or the tech company. This also depends on like then the regions we are in. So there could be that in China for like from the local companies we have than like lower prices than like with than NVIDIA or like Western companies. maybe also driven by some trade politics. So that's one. then also we need to differentiate between like the chip generations. So there, if you ⁓ talk Nvidia, they have the newest generation, it's called Thor which might be then more expensive than like the previous one, the Orin ⁓ chip, Orin solution. there, yeah, I would say it's a bandwidth of pricing. ⁓ that is offered by then like the different chip vendors. And then it also depends on like the volume on the negotiation power and then like also the region. Daniel Abreu Marques: And as you already noted, we see several OEMs ⁓ also thinking about in-house chips programs. So we see players like Xpeng or Rivian that create their own chips. ⁓ So what's driving this vertical push? several OEMs ⁓ also thinking about in-house chips programs. Augustin Friedel: Mm-hmm. Mm-hmm. Daniel Abreu Marques: So what's driving this vertical integration push? Augustin Friedel: If you look back, think one of the first OEMs that really pushed for own chips was Tesla. I think they have now the fourth of fifth generation in development. think the fourth is now on the market. I think they released 4.5 couple of weeks ago, at least was spoted in vehicles on the street. think version five is coming soon and they're already working then on version six. version six And the main reason is that, of course, they could optimize the software and silicon play. So to have the most efficient package, most optimized package of then software and then hardware. And of course, one of the aspects is also then like the costs. So if you think about big silicon providers or silicon companies, they often have like a very high margin on that. main reason ⁓ is ⁓ of they could optimize the software and silicon play. And of course, one of the aspects is also like the costs. And there then also ⁓ OEMs the opportunity that they could basically reduce then ⁓ bill of material by ⁓ long-term by going with own chips. But the most important decision factors to go for own solutions is linked to ⁓ the and then also the control of having the best chip and software combination. then also... be able to ⁓ and manage the roadmap of new features, of new functions. Daniel Abreu Marques: Besides like giants ⁓ Tesla, we also see newcomers like Rivian venturing into chip development. ⁓ recently announced their new chip, RAP1, 800 tops or something like that. What do you of relative small automakers taking such a, I think, costly and also complex project in-house? Augustin Friedel: Yeah. Yeah. so there, mentioned, like Rivian ⁓ is this direction. Also like in China, we have Xiaopeng and then Li Auto. I think they also announced last week that they with their own chip. So I mean, at the end, it's about scale, which they would need to reach. I mean, what we see in there, it's quite interesting. Okay, they tried to scale, of course, this automotive and with like EV models or software defined vehicles or AI defined models. So if you talked about Rivian, they have like the next vehicle coming like this year onto the market. And I think from the end of the year, the system also equipped with like the new chip. So they have them or they hope that this will generate enough volume then also to make the business case work. But if you look to China, what we also see is okay, they build them basically their portfolio around chip. And if you take Xiaopeng, for example, they have them, like not only the vehicles ⁓ are powered by the different or like by their own SOCs. Often we have them like more than one SOC per vehicle. So that also drives them volume, but they are also like robotics. So humanoid robots that Xiaopeng and also before Christmas and they should come on the market soon. then based on ⁓ the in-house chips as well and then also like at Xiaopeng they have like air taxis. So they have like their own subsidiary for vertical takeoff and landing like eVTOLs and these are then also powered by the own chips. basically see what could be equipped with the own silicons to get them to the volume they need to make the business case work. Daniel Abreu Marques: makes sense. basically automotive alone is enough. So therefore they try to feed other verticals as well with their own chips. ⁓ do you think the strategy Augustin Friedel: Yeah. And it could be also the case with Rivian and others. So when you follow the CES for example, then Mobileye said go into robotics as well. They acquired a company, I think for close, ⁓ for than 900 million in the humanoid robot space. Rivian, I'm not sure if they teased already anything ⁓ related to but most of the different OEMs, especially out of China, they see different growth paths or horizons and different business verticals. And of course, the ⁓ most hyped one at the moment is robots for production facilities, but also for household uses. Daniel Abreu Marques: do you see then like the competition for Nvidia and the mobile eyes of the world ⁓ are they still a go-to solution or do they face now really pressure because they all built their chips basically in-house then? Augustin Friedel: Yeah, think it's, there will be more competition for sure, but also the market is growing. I mean, if we now look at like SOCs or high performance computers deployed in vehicles, it's at a very low percentage of like vehicles sold. Yeah. So there, the market is also like growing and it's huge. And of course, like we also have like the political tensions. Yeah. So We don't know if Western companies will be able to sell chips in China, for example, in the future or versa. If ⁓ maybe the US will be blocked for like any Chinese companies. I think that's also like part of the game or the development that we see more like regional value chains or companies for the SoCs and then the chips. then... your question, will Nvidia be replaced or also Qualcomm and the others? I think they have also a very strong moat or lock-in is then the software layer. So if you go with the semiconductor companies, often also include in the products ⁓ the different software layers on middleware, maybe also on applications and across your vehicle programs and then also you do the same. in the development environment where you have to make the solutions provided by the different semiconductor companies implemented in your enterprise engineering IT. And that's not then too easy to replace. So basically then they try to generate this lock-in effect and we are like software solutions. And if they have to reach that or if the new OEMs or partners, tech partners, then implement the software solutions, it's then very like difficult for them to replace. Also for the next chip generation, they might go into the same chip vendor they had in the past because of their systems are optimized for that. Daniel Abreu Marques: Makes sense, makes sense. And are there, besides software, ⁓ or buying criteria which make like the NVIDIA, Mobileye, or any other supplier solution more attractive than building it in-house? Augustin Friedel: I mean, also if you talk to it, or if you have a look at Mobileye versus then also NVIDIA and others, talk about different, let's say, systems, designs. So with you get one monolithic package ⁓ out of, or or based on the SOC and based on, get the sensors from Mobileye, the software, and so on. So it's not very modular. ⁓ these types of systems are more than optimized for OEMs that want to buy the whole package. Like NVIDIA, and also like Qualcomm and others, they have a more modular solution. So they're basically, you could then combine the ⁓ SOC from one of the semiconductor provider suppliers, with sensors from another ⁓ supplier source. And you could then also run... like multiple or different application layers on top of the SoCs or on top of the solutions in the vehicles. So that adds way more flexibility than also for future development and so on. And there it really depends on ⁓ what an OEM wants. So do they want to have like one package that's like all in and they don't want to change anything? Do they want to ⁓ a more flexible solution where they could change then some of the hardware and software components? Or do they really want to develop everything in-house, including ⁓ the And on this scale, ⁓ think every different OEM should make their thoughts around, what do they want? What financial bandwidth do they have to invest in the solutions? Which resources do they have to build in parts of the system? And then also, ⁓ what is ⁓ time to ⁓ they would to reach? ⁓ And there are several other factors probably that are influencing them, like make versus buy decisions. And this kind of assessment is needed to decide then, okay, what is then the best solution to go for? it more than buy? Is it buy? Then like the whole package, just part of the solution. Is it then partner, co-development, or is it then like the make decision? Daniel Abreu Marques: before you mentioned that the market is growing, and do you have numbers, the percentage of vehicles equipped with such chips, or can you them with us? Rough numbers? Augustin Friedel: It's in a low percentage, don't have it in front of me and I don't have ⁓ correct numbers in my head. But ⁓ we see ⁓ you talk about the different SAE levels, like ⁓ three, for example, is the penetration is super ⁓ basically that would be a solution where we need high-performance chips. So if you move then downwards, we talked about level two plus plus, level two plus, level two. So there was like level two, there's like basically the kind of like turning point if you need like on high performance chip, like Nvidia or like the other ones we talked about and there so basically you have the highest volume share and like level two plus, level two plus plus. I think it's, if you talk about China, we are roughly at like 10, between 10 and 20%, ⁓ but could a bit off there. In Europe and the US, you're a bit lower, yeah. ⁓ Basically there you can get a direction of ⁓ shares that are equipped with high performance chips. So it's not super high. So maybe it's in ⁓ low 10 or between 10 and 20%, I would assume. If you talk about level four vehicles, mean, there we have, talk about a couple of thousand vehicles in total globally. So there probably we have less ⁓ 10,000 vehicles that are... equipped then with high performance compute for like level four. So that's a very small market as of today. But you covered very well in your newsletter Daniel. So that's one very exciting market and a growing market. ⁓ So yeah, I think there we see then, then high growth and like link to level four, but also like level two plus level two plus plus will grow as well. The industry is about unsure about level three. Yeah, so will this be like a solution going forward or where we have more like level two plus, level two plus plus solutions and then jumping from there to like private or shared use of level four vehicles. Daniel Abreu Marques: What is your personal opinion on level 3? Augustin Friedel: It depends on the market again. if I mean here, you look in, so both of us are based in Germany and Europe. Here the use case is on the Autobahn and there I think it's very tricky to get like a full ODD coverage of all the different like highways that would be needed. So I think there to get this to work is quite challenging and also like the validation, homologation costs are. relatively high in comparison to like level two plus solutions. So I think that's a high, would be a high price product with limited demand. So I personally have my questions about that. If you talk about China, there you have a completely different use case and ODD. So there we talk about inner city use cases. So urban use cases, often with like lower speed. with also maybe a like, let's say dense urban environment and traffic. But then this should then also reduce the complexity of the homologation validation. And so there could be a market for like level three in China ⁓ because of different customer expectations and the different use cases that would need to be covered. So there it's less about, okay. Let's drive 200 km per hour on the Autobahn ⁓ and have level 3 solutions. It's more like, okay, let's stay in the city or be in the city like Shanghai. You might have crazy traffic situations, ⁓ low speed. Daniel Abreu Marques: Yeah, makes sense. Maybe staying in China, also talked about that there's this developing in-house trend also observable. ⁓ And do think this is also to reduce the reliance on Western technology and to gain an edge there? Augustin Friedel: Mm-hmm. This could be a case, yes. I think also like one of the reasons that I see as the Chinese companies are also very ambitious. So they want to prove that they could get things done, especially to high tech. And their semiconductors and systems on a chip are definitely one of the differentiating factors then going forward for like software defined vehicles, for AI defined vehicles. autonomous mobility. So I think that's of the drivers and motivators is their ambition to basically show the world we can do it, we can do it better. And then of course, the political tensions that they might not be able to source chips from Western world or they might be able to source it, but there could be then super high tariffs on different products. ⁓ want to be prepared for these scenarios as well. Daniel Abreu Marques: sense. And then we also have automakers that don't design their own chips. And you anticipate that there will be ⁓ deeper collaboration to get still some of the benefits of the customization? So I think about moves like Volkswagen investing in joint venture with Horizon Robotics ⁓ develop the chips. Will see more car companies forming such alliances or maybe even making acquisitions to secure the chips and also the talent needed? Augustin Friedel: Mm-hmm. I think the chip design or semiconductor design activities of OEMs are broader than most of us would imagine. A couple of weeks or months ago, I shared something on my LinkedIn profile about the different approaches and programs of OEMs to get more control on the semiconductor supply chain. And we see there are different, let's say strategies or trends, but basically, like most of the big OEMs are involved somehow in ⁓ own chip designs or getting closer to the control points of chip designs, of semiconductor designs. Also, have to like Stellantis that have, they have formed like a joint venture ⁓ Foxconn to develop their own ⁓ So they start with ⁓ so with, I would say less performance chips, but ⁓ done also the ambition to. go in the direction of high-performance chips. you look to Japan, ⁓ quite interesting from my personal point of view, because you have the different consortiums. So they are basically the different OEMs, ⁓ forces, ⁓ with the local and ⁓ regulators build in on the local tech ⁓ companies and conglomerates there to build in there ⁓ a supply chain. ⁓ And that's then also the case what we see that in Korea, for example. Yeah. So basically you see it in there a lot of activities in the different regions, you see different activities. And my question would be, okay, do you need to go alone as OEM or might there be also like a solution like what we see in China, sorry, in Japan, for example, where we have the whole industry, the different OEMs, join forces to get to the best. semiconductor solutions, but also then at the same time benefit from like scaling and from volumes, because that's definitely one of the, as we discussed earlier, one of the main reasons for success or like for super high, then semiconductor prices. If there's like no volume, then basically there will be no like, let's say cost efficiency or like cost benefit. Yeah. So this could be something I could imagine for. something I could imagine for. the Western world as well, especially here in Germany or like in Europe ⁓ So that we see OEMs joining forces to not everyone is building their own ⁓ semiconductor so that we have them like an industry wide semiconductor ⁓ solution, which is then on the side securing the supply on the other one then, so it is then also leading to lower costs and more control. the Western world as well, especially here in Germany or like in Europe or maybe also in US. So that we see then OEMs joining forces to build like, not everyone is building their own semiconductor so that we have them more like an industry wide semiconductor solution, which is then the one side securing the supply chain, on the other one then, so it is then also leading to lower costs and more control. Daniel Abreu Marques: Yeah. And you mentioned also that they want to build their own supply chain. Do you think the recent semiconductor shortage was then a wake up call for the auto industry that showing that, okay, if we don't have chip supply, the production holds? Augustin Friedel: mean the Nixperia crisis we had at the of last year then? Yeah, ⁓ think that was something no one had on the radar. I mean, ⁓ you look there at the chips, I think they were like 44 nanometer, 40 nanometer ⁓ chips. So very, let's say, commodity semiconductors that basically were breaking down in the ⁓ production and so on. And there, think it was more forced by some, let's ⁓ stakeholders the ⁓ politics in the But yeah, ⁓ I think was more like a wake-up call for the different OEMs to have more control about their whole supply chain. So we were all talking about high-performance semiconductors and chips and basically how we are at risk if you don't own it. that we will be dependent from NVIDIA and from others, but at the end, one very, let's say, tiny commodity part was slowing us down. I think that was more the wake-up call, okay, let's not focus only on the hype topics ⁓ like ⁓ It's okay, we need overall supply chain resilience to be independent ⁓ from then the shutdowns production lines and... and high losses that we have seen and probably have seen last year or at end of 2025. Daniel Abreu Marques: if ⁓ are now working maybe directly with semiconductor companies or making their own chips, ⁓ where this leave the Bosch and the Continentals of the world who used to integrate and supply a lot of electronics? Augustin Friedel: think there ⁓ we see then also they depends on the different suppliers, but some of them invest also in like own semiconductor IP. Of course, they don't produce it then, like in-house, but they use them like global factories, for example, ⁓ Fabless ⁓ partners ⁓ the chips for like some of the modules. think Bosch is doing that for for radars, where they develop their own radar chip, for example. So you see then also the activities at tier ones, that they develop their own semiconductors and solutions for chips, maybe as of now, not like the ⁓ HPCs, like high-performance chips, more like for different modules. But if are going back then to the high-performance chips. What are the opportunities then there? One is they still need an integrator and ⁓ who is then building the ⁓ domain controllers based on the different SOCs. So if you look at then all the different suppliers, normally partner with all the different semiconductor companies to have then production ready domain controllers either autonomous driving, than ADAS systems or for IVI solutions. So that's definitely one value stream the tier ones are tapping into. And then second one is, as you mentioned, this integration part is still there. So might directly to the semiconductor suppliers and source then the semiconductor. So what we have seen with Volkswagen and Qualcomm. But then as they need someone who is then building their domain controller. controller based on the semiconductors and there's still the need for then the integrator system integrator at scale. Yeah. So that you have them, they also like the safety case that you have the life cycle operations that you have to domain integration. You have the You have like the whole fleet rollout you support. And then like the homologation. Yeah. So these are the tasks that still, we're like still third parties are needed. And that's like tier one can win. And as we discussed earlier, so there we have like also the two types of solutions on the market. So one is the turnkey. So that Mobileye for example is providing this, like ⁓ lot of ⁓ integration and their own chips, their own software, their own like mapping solutions, their own sensors. But if ⁓ then talk like alternatives, where it's more modular. the OEMs or the customers are still interested in getting more like turnkey-like solutions. Then like the tier ones or the ⁓ suppliers, the partners could be then also the ones that are basically combining the different modules, also combining the chips and domain controllers, the software integration, ⁓ and also diagnostics and everything that's needed to that. So they are then becoming more like the glue in the system ⁓ then ⁓ I would also say they won't disappear. Just because we have the new players now on the market. But they definitely would need to replace or need to change and reposition their value proposition. Daniel Abreu Marques: Makes sense. Absolutely. And maybe staying on the vertical integration theme. all of us are ⁓ talking about Nvidia chips. ⁓ in the end, Nvidia is the chip designer, not ⁓ chip producer. Can you give us here a short explanation how this whole ecosystem or supply chain looks like? Augustin Friedel: Mm-hmm. Mm-hmm. Yeah, normally you have, as you mentioned, like the different chip design companies. Nvidia, have to know. So Synopsys, which is more like for custom chips, ⁓ have like Qualcomm, who's also more on chip design. Intel, ⁓ like basically the big names you were talking about, the same then from in China, it's like Black Sesame and then ⁓ I think also Horizon Robotics. They also don't have their own production. And then normally they go to a so-called Fabless partner, which is then building the, or producing the ⁓ chips you Of course, the biggest supplier in Taiwan then was producing all the chips, you have to know other solutions that could be used in either global, I think it's global factory it's called. and sorry, Global Foundry, not factory Global Foundry. They could be then used as chip producer. Of course, you have to mention like TSMC from Taiwan, ⁓ Intel, I think they also want to ⁓ offer them like their FAB-less infrastructure as like a producer for different chips. You have to know, so if you look in China, different players like Huawei, they have like the whole supply chain then ⁓ set up then not only design chips, also built in the chips, have to know so consortiums again of Western companies with local companies like ARM, with Geely, for example, they joined forces. Yeah, so it's there then different stakeholders that could be used to produce the chips and often ⁓ they selected and depending on of course, like the regions ⁓ where they are required to produce, ⁓ but then on like the node size. Yeah, so if you talk about TSMC out of Taiwan. So they are seen in the market as the player that could handle like very small node sizes. So I think it's they're like at four nanometers, three nanometer chips at the moment. And that's like they're cutting edge or like their value proposition. So that's whether ⁓ at so high demand or see so high demand for their production facilities. you look at China, we talked more about like five, six or seven nanometer. node sizes, so basically some of the previous generations that TSMC mastered and they are like a couple of years behind of getting to the latest or the smallest node sizes and these are then some decision factors, which let's say value of each company in the value chain is then producing the different solutions, the different chips for the different regions at different cost levels. Daniel Abreu Marques: And if we have like then the OEMs saying that they're producing their chips in-house, do they still rely on foundries like TSMC or do they really do the whole process in-house? Augustin Friedel: Yeah, no, think they're like, it's like very capital intensive to build up the chip production. And so then also, as, as mentioned earlier, like someone like Bosch, one of the suppliers, they joined in forces with then like Global Foundry or with others to manage the end of production of the ⁓ of the Because this this type of investment would be super high and ⁓ I mean, they're also, you would need to basically then also prepare it for future chip generations. also you need to like super high volume ⁓ make the business case work. And they hand off the supply risk ⁓ to ⁓ players ⁓ TSMC, Intel, Foundry and others. Daniel Abreu Marques: Makes sense. And you also mentioned that TSMC is able to produce the really small node sizes as maybe the only one or one of the few players that are able to do that. Can you also explain why that matters? What is the difference between like four nanometers or six nanometers? Augustin Friedel: like what we discussed at the beginning. So it's then about ⁓ influences, the performance of the chips and then also the ⁓ energy ⁓ of the chips and then ⁓ also size of the chips and then ⁓ the then also at the end. So it's then like the reducing the node sizes, of course, is then reducing the complexity. ⁓ So when the new chip generations are launched, it could then also lead to higher costs and like low margins. You see this also in if you look at the different capital reportings of TSMC and others. if we ⁓ there from like a systems perspective, it's most like the most relevant sort of like performance, then also like size, thermal aspects, efficiency, ⁓ energy drain, and so on here. So that could be then reduced. by or could be optimized by reducing the node size. But this would mean that you would often need new production lines or production technologies. So ⁓ there, ⁓ think that's and that's that leads into like high investment ⁓ of ⁓ for production and new production lines and production machineries, etc. And the different OEMs or suppliers, they're nowadays like cash constraints, so they have like, like budget restrictions ⁓ that I can need to write off like a lot of budget on EV programs and so on. Yeah. So they don't have basically then the budget available for setting up their own facilities and own production. So they go then to the ones that have already built up the capacity. Daniel Abreu Marques: And if you're looking at level four vehicles, do you have the one go-to chip that would, in your perspective, have the best balance of cost and performance? Augustin Friedel: It's hard to say because you need to look at the whole package of the chip, of the sensors that are being used, often the software, the application software, the middleware, how efficient basic you then design the whole system. So I think there's no answer to this question if there's one single best chip. For me personally, it's more interesting to see the different strategies. If you look at Waymo, They, at least what I know, what I see, ⁓ they going for like optimization on all the different layers. Yeah. So they optimized and probably their own like chips or like own semiconductors. They then optimize their ⁓ They're then also heavily in like sensor design and then like central configuration. So they are like, ⁓ like all in, would say the ADE stack or Self driving stack. So that's. ⁓ That's an interesting direction or progress to look at this will evolve over time. Now, when they have the 16 of new funding, which they could use for further scaling and then also system improvements. And then the other direction is that you have players like WeRide or Pony and others that rely more on NVIDIA chips or Qualcomm chips and others, or Chinese chips, and then try to optimize based on that. Yeah, think that's the second direction I'm looking at. I don't know or ⁓ I believe that there is still no answer which one will be more successful, which will be more cheaper. I think that the time will tell. We are so early in the ⁓ driving ⁓ It's hard to predict there what would be the best solution. Daniel Abreu Marques: Absolutely. Then maybe one final question, If you were advising ⁓ an executive on the semiconductor strategy, so what key questions should they be asking and what factor should guide their decision to build chips in-house versus partnering or buying off-the-shelf chips? Augustin Friedel: Mhm. Yeah, so I think what was clear in our conversation, it's not like a binary decision if you if you go left or right if you buy or, or design, yeah, so you need to consider different decision vectors. Yeah, so I think one could be okay, how do you really want to differentiate? Yeah, so do you want to be then also like the ones that sending the message to the market that you own your autonomy brain, I like as Rivian in this doing for example, yeah, so there might be In addition to what we discussed about performance, etc. there might be also, let's say, investor narrative attached to that. So how do you want to be seen in the market? How do you want to make the difference? And there could range from, we see this autonomy brain? Could be then also, do you want to be like the cost leader? Do you want the ones that is basically the OEM with the shortest update cycles and time to markets? Or do you want to be like the OEM, for example, that has like the most supply chain? So this is then, I think, ⁓ of the ⁓ OEMs should think about, or also like autonomous driving companies should think about. ⁓ do they want to differentiate? How do they want to be seen in the market? And then the one is then more linked to internal operations. So do you have ⁓ the operating model to have like a the most performance CI, CD simulation at scale. Do you have like all the data ops, safety cases for automation, et cetera, that is needed to succeed or are you not ready for that and you need partners? And then you can continue with which control points you want to own. So you want to own like just the data, do you want to own the tool chain or do you own like as mentioned earlier, the whole supply chain? So I think that's an also important discussion. what of the control points must be owned by the different OEMs or tech players. And then based on that, you can make the decision, are you going more for buying, more for partnering, or then are you open to invest in your own semiconductors? And I think also what could be always considered is, you like just following one strategy, like one long-term strategy, or what we then also see in China is there may be like a short-term strategy where you buy or where you partner? And then in the meantime, you build up all the competencies, the control points, the budgets, then for the long-term in-house semiconductor chip supply chain. So this could be then also something to be considered and then link it to your own time horizon. I think these are some of the aspects ⁓ OEMs and players need to think about before. They make ⁓ them like a decision if they invest in their own semiconductor supply chain and then also link it to like their own philosophy and history of the company. Does it fit to their risk appetite to then also the skills they have and so on and so on. Yeah. So that's then it's getting really complex, but all like discussing and deciding on all the different factors is super important to have to like a sustainable ⁓ like semiconductor strategy. That's how we normally work with the different stakeholders in the industry. Daniel Abreu Marques: That's super insightful. Thank you, Augustin. So this was a really, really great episode with you, Augustin, really insightful. And as we mentioned, it's a market where there's a lot of movement, a lot of developments happening. So I'm happy to have you on next time to talk about how the market evolved. But thank you very much for your time. It was very insightful. And see you soon. Augustin Friedel: Thank you Daniel, pleasure to be on the show here and I hope all the listeners will learn something and more than happy to be back and to talk about a different topic in the future.