Rivian’s Roadmap to AI Architecture and Autonomy with Founder and CEO RJ Scaringe

0:00By 2030, it'll be inconceivable to buy a car and not expect it to drive itself. Every single one of our cars, we want to have the ability for it to operate at very high levels of autonomy. Radars are extremely cheap. LiDARs are very cheap. But the really expensive part of the system is actually the onboard inference. An order of magnitude more expensive than any of the perception stack. My view is EV adoption in the United States is a reflection of the lack of choice. As consumers, we need lots of choices. We need to have variety. We self-identify with the thing we drive. The world doesn't need another Model Y.

0:29The world needs another choice.

0:36Hi, listeners. Welcome back to No Priors. Today, I'm here with RJ Skirinj, the founder and CEO of Rivian. We're here to talk about their autonomy strategy, proprietary chips, their coming R2 model, whether Americans want EVs, and what our relationship to cars is going to be in the age of AI. Let's get into it. RJ, thanks so much for doing this. Thank you for having me. So Rivian's already an incredibly cool company. How did you decide it was going to become an autonomy company when that happened? I mean, from the beginning, we thought of it as a transportation and mobility company. And in fact, even before Rivian became Rivian, when I was thinking about what's the first products, it was unclear what kind of car would be, or even if it was a car.

1:17But it was always clear we wanted to be at the front edge of helping to redefine what does it mean to have access to personal transportation. And so autonomy has always been part of the strategy, but it's now fully coming to life with the technology that we're building. And you think about the function of Rivian. There's transportation. there's also the experience like when how long did you guys start investing in the autonomy strategy here yes we launched our one in um very end of 2021 and we used what i'll broadly characterize like a one dot o approach to autonomy so we had a perception platform we used a third party of front-facing camera that was essentially a third-party solution that then plugged into an overall framework that we built but it was all rules-based so the camera is fed a rules-based planner the planner would then make a bunch of decisions around the feeds from the perception and it was you know the moment we launched we knew it was the wrong approach but it was the thing we'd started working on well before the launch and so at the end of 2021 beginning of 2022 we made the decision to completely reset the platform and was that hard decision no because it was so clear when we made we made that you know you're building something like this you're you recognize you're going to spend many, many billions of dollars creating it.

2:30So we knew this like at the core of transportation is driving. At the core of that is a shift to having the vehicle be capable of driving itself. And so we made the decision to redo it like clean sheet, no legacy of what we had built in the Gen 1. And that first launched from a hardware point of view in the middle of 2024. So it was with our Gen 2 vehicles. You know, not a single line of shared code, not a single piece of common hardware on the perception or on the compute side. And then we had to build like the actual data flywheel. So we had to grow the car park to build enough of the data flywheel to then start to train the model.

3:08And what we showed in our autonomy day late last year, late in 2025, was the beginnings of a series of really like super exciting steps of how this is going to grow and expand. I say this all the time. I think of not just for Rivian, but I'd say for the auto industry in general, the last three years compared to the next three years are going to look very different. So the rate of progress that we saw in autonomy between, let's say, 2020 and 2025, or 2021 and 2025, and what we're going to see between today and, let's say, 2029 and 2030, are completely different slopes. And that really comes back to entirely new architectures now being used to develop self-driving, actually truly AI architectures, whereas before, These were not AI architectures in the truest sense.

3:53They were using machine vision, but really rules-based environments that we defined as humans. We codified them, which is very different than Apple today. You might actually have perfect timing here in that I got to be part of investing in the first wave of independent autonomy bets that were working with the OEMs at my last investing firm. But this is, let's say, eight, ten years ago. And as you mentioned, there's several architectural revolutions since then. And so for companies to make that shift from, you know, we're going to have these separate perception and planning systems to more end-to-end neural networks.

4:29I asked because I felt it was actually quite a hard decision for people choosing their partners and internally from a technical perspective. Well, I think it, I mean, you can see it. So there's, if you go back to the very beginning of the idea of self-driving, a lot of effort, a lot of spend happened for companies to build these rules-based environments and to build these more classic systems. and when transform-based encoding came along just a couple of years ago and it shifted very rapidly to it was clear that the future state was going to be neural net based it was hard because if you're a company that's built all these systems it's like do i keep investing what i had what do i what do i do with all this work that was was built before and the reality is is a lot of it is the vast majority of it's going to be pure throwaway because it wasn't like a gradual shift There was a complete rethink of how things are architected.

5:20How did you decide that this was going to be an in-house effort versus a partner effort? Most people who made cars said, we're going to go partner or buy something here. I guess the emotional slash philosophical is on things that are really important, we've taken the approach of vertically integrating them. So electronics, our software, all the high voltage systems in the vehicle, so things like motors, inverters, all the power electronics, these are all things we develop and build in-house. And in a few cases, we had to start with something that was either off the shelf or partially off the shelf.

5:55But today, all of that's completely in-house. And in the case of self-driving, we knew that long-term, it needed to be something that was developed internally. we started as i said with a mobilized centric solution which a lot of folks did right particularly in like that 2015 to 2021 time frame but when you really look at what's necessary to to be successful in a neural net based approach there's a core set of ingredients that very few people have and i think we uniquely have them so first and foremost you need to have complete control of perception platforms you have all the everything that the the system is capable of observing whether that's cameras, radars, or LIDARs, or some combination of all three, you need to control that, meaning there's no intermediary company that's like processing some of the information.

6:39And so that's powerful because you can then feed raw signals into your system. The system needs to be capable of triggering unique or interesting or noteworthy events that you can then use to train that triggered, you know, those triggered moments need to then be captured, saved on the vehicle. And then when the time arises where you have Wi-Fi, ideally, send it up. And the reason I say Wi-Fi, this is a lot of data. So you could, of course, do it over LTE, but it's expensive. And so you have to have a really robust data architecture on the vehicle. Then you need to be able to send it off board and use that with a lot of training, so with a lot of GPUs, to train a model.

7:19Companies that are either developing independent solutions that are not a car company, they typically don't have access to the type of mileage that we do. So the huge amount of data that our vehicle is generated. if you're developing this from a sensor set point of view you typically don't have the vehicle architecture and the vehicle car park so we just came to the view that we have all these ingredients to do it really well and it's like not an optional thing it's the companies that do this well will exist the companies that don't do this well like i feel really strongly this they will not exist they will shrink to shrink to nothing the last time they approach zero you think it can only be delivered and really a vertical vertically integrated no i think i think there's more than one less than five companies outside of china that have the necessary ingredients to do this the capital the gpus the the car park with you know enough vehicles generating enough data i say more than one less than five it's probably the control of that whole training loop it's probably like more than one less than three maybe four like there's very small number of companies that can do this i think the unique spot we are in time right now is the 1.0.

8:24Can I ask explicitly then? It's you, it's Tesla, it's Waymo. Is that the three? I would include all three of those, yeah. And there's maybe one or two others in the mix. But I think the challenge is you have to look at not just the moment in time for performance where we are today. Do you have the ingredients to continue making progress at a very high rate over the next four or five years? And so a lot of the solutions that are more 1.0 based and are sort of stuck in that framework, I think have a like a truly a zero percent chance of progressing to be competitive with a neural net based approach.

8:58And then our best approach does take a lot of time. You have to build a ton of inference on it. You have to either buy it or build it. A lot of inference. We decided to build it. So we built an in-house chip to do this. You need to have a car park this large. You just mean enough onboard compute to actually run the models. To run the model. In the car. Yeah. In the vehicle. And so you could you could buy that. Of course, the video makes those. but you need to be able to do that at scale and have it in every car. And so we took the decision to make our chip in-house. Is that more a capability decision or a cost decision?

9:30It's a cost. We want to have it on everything. So every single one of our cars, we want to have the ability for it to operate at very high levels of autonomy. And so we design and spec and build the cameras. Radars are extremely cheap. LiDars are now very, very cheap. But the really expensive part of the system is actually the onboard inference. And so that's like an order of magnitude more expensive than any of the perception stack. I think people focus on the perception because it's the things we can visualize. Right. But the brain is actually the most expensive part. And so we brought that in-house as a way to remove cost from the system so that we can easily deploy this on every car.

10:07You are taking like a sort of step-by-step approach to levels of autonomy. Yeah. How do you think about how quickly you approach level four or the safety case around each of these things? How fast your team goes against this? Yeah. Even this question is unique. It's just a few years ago, 20, 2019, 2021 even, there was very clearly delineated ways to approach autonomy. There was a level two approach, which was camera heavy, maybe with a few radars. and then there was a level four approach which was of course had cameras but had a lot of lidars it was sort of inconceivable to think of the level two system becoming a level four and similarly the level four system was way overbuilt to even like conceivably think about putting that on every consumer vehicle well you didn't want the the big one yeah you didn't want all these parts yes so yeah the tens of thousands of dollars of perception so what's happened is those two worlds just i think have just started to very clearly emerge where the delineation between a level two a level three and level four um in terms of perception and and in terms of compute has started to fade and it's now essentially just removed like how capable the system is at addressing all these corner cases and you know this is what's hard for a consumer to recognize if you're driving a level two system or a level three system or a level four system for 99.9999 like three or four nines identical right the difference is like the fifth or sixth or seventh nine on that is these like extreme corner cases and so i think it's actually led to a lot of confusion where you'll be in a level two system like the car could drive itself and you're like yes it can under most of the roads millions of conditions except these very unique corner cases and so to your point on safety cases the question then becomes is like how confident are we in the system capability in covering these really obscure, unlikely rare events, which, of course, if they're not covered well, it can lead to a really terrible outcome, the vehicle had a bad collision.

12:17And so that's where the neural net-based approach has just changed things a lot. So the capabilities are so much stronger and the ability now, I think, for us to deploy on a lot more vehicles, have a car park that's very large. So we went from, you know, a few years ago, So state of the art was you'd have a test development fleet of maybe a few hundred vehicles, maybe like high hundreds of vehicles to now like thousands and thousands. Every single car on the road is part of your data fleet that's identifying these unique corner cases and then running the model against them to test. And now, of course, we're simulating those unique cases and we can do a lot there.

12:55So just the whole nature of it's changed so dramatically that, I mean, I think by 2030, it'll be inconceivable to buy a car and not expect it to drive itself. You know, maybe that's sooner, maybe like we hope it's sooner, like we're targeting a little sooner than that. But certainly in like a very, very near future, like that will become a must have on a car. Sort of like it's hard to imagine buying a car today without airbags or buying a car today without air conditioning. These things at a moment in time were optional. I think in not too much time, a couple of years, it'll be hard to concede buying a car that can't drop you at the airport or pick up your kids from school.

13:31I would argue that right now, most of the biggest car makers do not have the ingredients that you described to make this a reality. So do you think that that's going to play out in the market where autonomy will be so important as a driving feature, core feature of the car, that there's just going to be a big market share shift to those? those who can figure it out i know you're biased here but i'm like no no no no i think it's it's a hard question answer so i think it's uh i always characterize like this i think it's inconceivable for a car company to continue to operate at scale like mass market i think very niche enthusiast realms sure but like at scale without a software-defined architecture which is even before you get to autonomy just like can you do otas do you have control of a sorry Can you define software, define architecture?

14:23Yeah, that's like before we even get to autonomy, it's like these are like basics. So the way car. The core thesis of. Yeah. Yeah. So the way car electronic systems have been designed and built and have evolved. With the exception of Tesla and Rivian, every car on the road has what is called a domain based architecture. So you could also call it function based architecture. So all the functions across the vehicle, let's say chassis control or door system control or 8-track, your air conditioning system, all have little computers associated with them. Right. What we call ECUs, electronic control units.

14:56And in a modern car, you might have 100 to 150 of these. And each of these run their own little island of software. And that little island of software is written by a supplier, more likely a supplier to the supplier. So you go to a tier one and they hire a tier two who writes the code base to run your HVAC. Is this why it's impossible to debug like a software system? It's also why it's really hard to do an update. So imagine you have a hundred different islands of software written by a hundred different teams that all have to coordinate. And so if you want a feature, you know, something that manifests as a feature often involves combining functions from different domains.

15:31So a simple one to visualize is when you walk up to your car to get into it, you want it to automatically unlock. You want the HVAC to go to your preset. You want your seats to adjust. You want it to make an audible noise on the outside. You want the lights to do something. You probably want the audio system to do something. Those are all different little ECUs in a traditional car. And the coordination cost in it is really high. It's very unlikely that a car company will make a change to that sequence because it involves coordinating amongst maybe 10 different players. In contrast, on an approach where you build a zonal architecture where you have a very small number of computers, one, two, maybe three, depending on the size of the car, that are running one operating system that control everything, it's very easy.

16:12That sequence, you could make updates to in a matter of minutes, maybe an hour. You could change the whole sequence of what happens when you walk up to the car, issue an over-the-year update, and it's very straightforward. How often does Rivian update? We do about one a month. And it's typically, you know, we add a couple of new features. We add refinements to existing features. We're listening to like what customers are seeing and asking for. But, you know, every month the car gets like notably better. And it's created this really amazing dynamic where customers are like excited for the update.

16:45Like when's the next OTA going to drop? The irony of all this is these domain-based architectures goes back to like, how do we arrive at this? It actually goes back to fuel injection systems. So up until early 1960s, every car on the road was completely analog. So there's no computers at all in the car. It was 100 % analog. And the first computers were there to drive the fuel injection systems. And car companies said, this isn't a core competency. Let's push that little computer to run the fuel injection system to a supplier. And the supplier will make that. And this is where you saw things like the Bosch fuel injection systems.

17:19Never planned. It's sort of like a field of weeds. Then over the next 60, 70 years, everything that became computer controlled to any degree is suddenly starting to have a little ECU, a little computer associated with it. And it just grew into this absolute disastrous mess that is today the network architecture that's in truly every car on the road with the exception of two companies. What I just described is what underpins. We did a large software licensing deal, a$5.8 billion deal with Volkswagen Group. the second largest car company in the world, to essentially leverage our network architecture and ECU topology for all their various brands.

18:02And so it's an interesting final point there on your first question, which is what happens to market share? So I think it's inconceivable that if to be at scale, that you don't have a software-defined architecture that allows your features to become better and better. And particularly thinking about how AI starts to integrate into the features, that's number one. Secondly, it's inconceivable to think about a car company existing at scale without the vehicles having very high levels of autonomy. And so car companies have a choice on both of those. They can either accept that they're going to shrink.

18:33That's choice one. Choice two is go build it themselves, which is really hard because they don't typically have these skill sets. They're not software electronics companies in terms of like their organizational DNA, or they can find a third party to source it from. And in both cases, there's not great third parties to go to. And in the case of autonomy, most of the third parties that did emerge over the last 10 to 15 years tend to be very much like classic rules-based, what I call it like AD or Pontius Vehicle 1.0 solutions. And those work pretty well for the business construct of selling like a sensor and a function.

19:11But that structure is really flawed when you want to have like a large data flywheel and it's constantly learning and evolving and you're issuing updates constantly. It's just, it's really hard to imagine that with an arm's length transaction. And so I think the vertically integrated stacks are going to naturally have some big advantages. So this might be an irrelevant question, but I'm curious. Do you think that the autonomy, like the models that maybe the three, maybe the one, maybe the five companies that come up with this develop are fundamentally different over time? Because I spent a lot of time in the AI ecosystem and the, let's say the language oriented foundation models, like feel like they're converging at this moment in time.

19:52I look at a Rivian and I'm like, I don't know, people adventure in that thing. Do you actually want it to do different things, have different styles or capabilities, or is it really just like, as much autonomy as possible safety case? Well, first, this is a great question. I want my car to drive really awesome. Like in the LLM world, a lot of it has converged because it's the training data sets. nearly the same. Yeah. So we're taking the breadth of knowledge that's contained on the internet and we're training models off of that. In the case of driving a vehicle, there is no internet of driving data.

20:25And so you need both a robust sensor set to be able to capture the data and you need a car park that has enough vehicles in it. And so of course, Tesla has the largest car park of vehicles by far. Our approach to this is we have a higher level of capability on our perception stacks. We have better cameras. We have radar. And of course, with R2, we'll have a LIDAR as well. A huge part of that strategy is not only those cover corner cases better, so the cameras have incredible low light and bright light performance. So the dynamic range of the cameras is stronger. We have more cameras, a lot more megapixels.

20:59We have radar, which is great for object detection. And the LIDAR, which is, it's a very powerful tool for training the models. And so imagine 800 feet in front of us, there's a little speck into a camera. It's hard to figure out what that is. And historically, what we would do to train that is you would have a LIDAR sitting on the vehicle on a ground truth fleet to help train your cameras. Putting that in every single one of our cars turns our entire fleet into this amazing training platform, this data acquisition machine. That was a core part of how we thought about our strategies. We're going to go not as heavy as let's say a Waymo on perception but heavier than let's say Tesla to build a really robust data platform on a vehicle-by-vehicle basis and then with a car park that's going to grow significantly with the expansion with R2.

21:51Yeah so I think first and foremost is there is no common internet data so the data sets that we're going to be picking up though are going to be very similar but but you have to go acquire it. But there's still different decisions about what data you care about acquiring yeah well i think this is what to like how does a car feel ultimately it needs to be safe and the differences in the way it drives or feels are going to be more about like what's the ui the user interface of it you know like even we just updated some of our features we have three settings for how the vehicle drives mild medium and spicy spicy is the highest one yeah and so it's just like a little bit more aggressive over time and we've spent time I'm thinking about this.

22:30I think this will start to become part of a key decision is how does the vehicle behave? And there's work we're doing to think about how the vehicle can behave in a way that, against a set of heuristics, drives like you. So the overall model is trained on how to perform in a safe way, but it actually learns some of your driving preferences and creates a model around you. Of course, in a world where you never drive the car because it's always driving for you, there's a way for you to set. I'd like it to aggressively change lanes. I'd like it to reside in the right-hand lane. Like those kinds of decisions.

23:05And those are less around the tech, more around what's the product or the UI feel like. Right. The ability to collect those preferences. Yeah, so it's preference-based. And I think we will see that. And that'll be a decision like a Tesla makes that may be different than how a Rivian makes it. You know, it's hard to say today. Can we talk about what the R2 means for like the company and some of the key design decisions here? I was just talking to Jonathan, one of your lead designers, about the constraints and aiming for more mass market and more volume here. I mean, yeah, you said it. It's an R1.

23:39It's a flagship product. Its average selling price is around$90 ,000. It's the best-selling. The R1S is the best-selling premium electric SUV in the country. So electric SUV is over$70 ,000. And we're the best-selling premium SUV, electric or non-electric, in the state of California. So it sells really well. It outsells everything in its class, like a Tesla Model X, it outsells like two to one. But because of the price, it's just limiting in terms of how much volume we can achieve with that platform. And so R2 is our first truly mass market product with pricing that's, as we've said, going to start at 45 and allows people that are in that, the average price of a new car in the United States is$50 ,000 in that like 45 to$55 ,000 price range.

24:26I think to have a really great choice. And to date, there haven't been a lot of great choices there. I'd say there's like sort of singular set of great choices with the Model 3, Model Y. And of course, that's shown through extreme market share capture, 50 % roughly market share goes up or down. But around that, call it half the EV market is Model 3, Model Y. So there's just such an untapped opportunity to pull customers out of ICE vehicles, out of internal combustion vehicles with a choice that has characteristics that are different and unique relative to a Tesla. These are like two substance types to be rapid fire questions, but they're important for me to ask you.

25:08Do Americans want EVs? Why haven't they adopted them faster? Yeah, I think to the last question, I think causality is always a hard thing to really understand, but let's zoom out here. The overall adoption in the United States of EVs is around 8%. The vast majority of vehicle buyers are buying vehicles that are under$70 ,000 with the average sale price of about 50. And so if you look at the number of vehicle choices you have at a price point that's under$70 ,000, depending on the year, this of course changes year to year. There's well in excess of 300 different vehicle model line choices, putting aside trims and performance packages, but just in terms of like overall vehicle types.

25:47and she can buy hatchbacks, minivans, SUVs, you know, two-seaters, convertibles. I mean, there's a whole array of different things you can buy. And in the EV space, I think, and this is, I think there's more than one, less than three great choices. And I'd say Tesla with the Model 3, Model Y is absolutely one of those. But there's so few choices that if you are looking for a form factor that's not a Tesla. So you think it's just missing product set that people are going to want? I think it's like an extreme lack of choice is how you put it. Like a shocking lack of choice. And this is what gets into interesting like corporate psychology.

26:29But because of the success of the Model Y in particular, the EV choices that do exist that are outside of Tesla are often very similar to a Model Y. Sure. So if you were to like draw like an outline, if you looked at the side view profile of a lot of its alternatives and draw a profile and then put it next to a Model Y. It's almost identical. There's a design sketch over here of basically the Model Y and all its competitors. They're all basically the same. It's like if you want a Model Y, buy a Model Y versus getting... You want something different. Yeah, so you have all these companies that are trying to create their own version of Model Y.

27:02And it's like, it's unfortunate because they didn't say, well, what can we do that's unique and different? And so for us, we think the Model Y is a great car. I've owned one. Many folks on our team have owned one. But the world doesn't need another Model Y. The world needs another choice. And so I think this is a reframing of just how we look at transportation is it's such a big space. It's such an area of personal expression that we need as consumers. We need lots of choices. We need to have variety. We self-identify with the thing we drive. We just haven't had it. So I think my view is the EV adoption in the United States is a reflection of the lack of choice.