AI, R2 and the Future of Everyday Driving | Rivian CEO RJ Scaringe

Vision for Autonomous Vehicles and EV Adoption00:00

By 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. In order to imagine more expensive than any of the perception stack. 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. The world needs another choice.

Introduction to Rivian CEO RJ Scaringe00:30

Hi listeners, welcome back to No Priors. Today I'm here with RJ Scaringe, 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.

Rivian's Origins as a Transportation and Mobility Company01:01

Rivian's already an incredibly cool company. How did you decide it was going to become an autonomy company? When did that happen? From the beginning, we thought of it as a transportation and mobility company. Even before Rivian became Rivian, when thinking about the first products, it was unclear what kind of car it would be, or even if it was a car, but it was always clear we wanted to be at the front edge of helping to redefine what it means to have access to personal transportation. Autonomy has always been part of the strategy, but it's now fully coming to life with the technology we're building.

Timeline of Autonomy Investments01:32

Rivian launched R1 at the very end of 2021 and used what I'll broadly characterize as a 1.0 approach to autonomy. We had a perception platform using a third-party front-facing camera that plugged into an overall framework we built, but it was all rules-based. The camera fed a rules-based planner. The planner made decisions around the feeds from perception. The moment we launched, we knew it was the wrong approach, but it was what we'd started working on well before launch. At the end of 2021, beginning of 2022, we decided to completely reset the platform.

Decision to Reset Autonomy Platform02:01

Was that hard as a decision? No, because it was so clear. When building something like this, you recognize you're going to spend many billions of dollars creating it. At the core of transportation is driving, and at the core of that is a shift to having the vehicle capable of driving itself. We made the decision to redo it clean sheet, no legacy of what we had built in Gen One. That first launched from a hardware point of view in the middle of 2024 with our Gen 2 vehicles. Not a single line of shared code, not a single piece of common hardware on the perception or compute side. Then we had to build the actual data flywheel, growing the car park to build enough data to train the model. What we showed in our Economy Day late last year, late in 2025, was the beginnings of super exciting steps of how this is going to grow and expand.

Rapid Progress in Autonomy Over Next Three Years03:02

I say this all the time. For Rivian and the auto industry in general, the last three years compared to the next three years are going to look very different. The rate of progress in autonomy between 2020 and 2025, or 2021 and 2025, and what we're going to see between today and 2029-2030 are completely different slopes. That comes back to entirely new architectures now being used to develop self-driving—truly AI architectures—whereas before these were not AI architectures in the true sense. They were using machine vision but really rules-based environments that we defined as humans and codified, which is very different.

Shift from Rules-Based to Neural Net Architectures04:00

You might have perfect timing here. I got to be part of investing in the first wave of independent autonomy bets working with OEMs about 8-10 years ago. There's been several architectural revolutions since then. For companies to shift from separate perception and planning systems to more end-to-end neural networks. I asked because it felt like a hard decision for people in choosing partners and from a technical perspective. You can see it. From the beginning of self-driving, a lot of effort and spend went into building rules-based environments and classic systems. When transformer-based encoding came along a couple years ago, it shifted rapidly. It was clear the future state was going to be neural net based. It was hard because if you're a company that built all these systems, do you keep investing or what do you do with all this prior work? The reality is the vast majority is pure throwaway because it wasn't a gradual shift—it was a complete rethink of how things are architected.

Vertical Integration Philosophy for Core Technologies05:32

How did you decide this was going to be an in-house effort versus partnering, given most car makers said we're going to partner or buy something? On things that are really important, we've taken the approach of vertically integrating them: electronics, software, all high-voltage systems like motors, inverters, power electronics—we develop and build in-house. In a few cases, we started with off-the-shelf or partially off-the-shelf, but today all that's completely in-house. For self-driving, we knew long-term it needed to be developed internally. We started with a Mobileye-centric solution, which a lot of folks did, particularly 2015-2021. But to be successful in a neural net-based approach, there's a core set of ingredients that very few people have, and we uniquely have them.

Key Ingredients for Neural Net Autonomy Success06:31

First, complete control of the perception platform: everything the system observes—cameras, radars, LiDAR, or combination. No intermediary company processing information, so you can feed raw signals into the system. The system needs to trigger unique, interesting, or noteworthy events, capture and save them on the vehicle, then send them up via Wi-Fi when available (LTE possible but expensive). Need robust data architecture on the vehicle and offboard with lots of GPUs to train models. Companies developing independent solutions that aren't car companies don't have access to the mileage we do or the huge data our vehicles generate. If developing from a sensor standpoint, you lack vehicle architecture and car park. We have all these ingredients to do it really well. It's not optional: companies that do this well will exist; those that don't will shrink to zero.

Few Companies Have Necessary Ingredients08:00

You think it can only be delivered vertically integrated? No, more than one, less than five companies outside of China have the necessary ingredients: capital, GPUs, car park with enough vehicles to generate data, control of the training loop. Probably more than one, less than three or four. Unique spot right now is 1.0. Explicitly: you, Tesla, Waymo? I would include all three, and maybe one or two others. The challenge is not just current performance, but having ingredients to continue progress at a high rate over next four or five years. Solutions stuck in 1.0 framework have 0% chance of being competitive with neural net approach.

In-House Inference Chip Development09:00

Neural net approach requires tons of inference—you have to buy or build it. We decided to build an in-house chip. Need enough onboard compute to run models in every car. You could buy from Nvidia, but need to do it at scale. More a cost decision, and we want it on everything so every car operates at high autonomy levels. We design, spec, and build the cameras. Radars extremely cheap, LiDARs very cheap, but onboard inference is really expensive—an order of magnitude more than perception stack. People focus on perception because it's visualizable, but the brain is the most expensive part. Brought it in-house to remove cost so we can deploy on every car.

Evolution of Autonomy Levels10:01

Taking a step-by-step approach to levels of autonomy. How do you think about approaching level four, safety case, how fast your team goes? Just a few years ago, 2019-2021, there were clearly delineated ways: level two camera-heavy with few radars, level four with cameras plus lots of LiDARs. Inconceivable for level two to become level four; level four overbuilt, tens of thousands of dollars of perception, not for consumer vehicles.

Merging of Level 2-4 Systems11:02

Those two worlds have merged: delineation between level 2, 3, 4 in perception and compute has faded. Now it's how capable the system is at addressing corner cases. Hard for consumer to recognize: driving level 2, 3, or 4 for 99.9999% feels identical; difference in fifth/sixth/seventh nine on extreme corner cases. Leads to confusion: level two drives itself under most conditions except unique cases. Safety case: how confident in covering obscure rare events, which if not covered lead to terrible outcomes. Neural net approach changes this: capabilities much stronger, deploy on more vehicles, large car park. From state-of-the-art test fleet of few hundred/high hundreds to thousands—every car on road part of data fleet identifying corner cases, testing models. Now simulating unique cases extensively. By 2030 inconceivable to buy car without self-driving (hope sooner, targeting sooner). Like airbags or AC—once optional, now must-have. Soon hard to imagine car that can't drop you at airport or pick up kids from school.

Market Implications for Car Makers13:30

Most biggest car makers lack ingredients to make this reality. Will autonomy cause big market share shift to those who figure it out? Inconceivable for car company to operate at scale/mass market without software-defined architecture (even before autonomy: OTA updates?). Software-defined architecture: traditional cars have domain-based (function-based) architecture—chassis control, door system, infotainment, AC each have ECUs (100-150 in modern car), each running isolated software by supplier (tier 1 hires tier 2).

Domain vs Zonal Architecture14:30

Impossible to debug, hard to update—100 software islands by 100 teams must coordinate. Feature often combines domains, e.g., walk-up: auto-unlock, HVAC preset, seats adjust, audible noise, lights, audio—all different ECUs, high coordination cost. Unlikely car company changes it. Zonal architecture: small number of computers (1-3) running one OS controlling everything—easy to update sequence in minutes/hour via OTA. Rivian updates monthly, adding features/refinements based on customer feedback; car gets notably better, customers excited for next OTA.

History of Domain Architectures16:30

Domain architectures stem from fuel injection: pre-1960s cars 100% analog, first computers for fuel injection outsourced to suppliers like Bosch. Over 60-70 years, everything computerized got its own ECU, growing into disastrous mess in every car except Tesla and Rivian. Did $5.8 billion software licensing deal with Volkswagen Group (second largest car company) to leverage our network architecture and ECU topology for their brands.

Future of Car Companies18:00

Inconceivable for car company at scale without software-defined architecture allowing features to improve, especially with AI integration. Also inconceivable without high autonomy levels. Choices: shrink, build themselves (hard, lack software/electronics DNA), or source from third party (not great options; most AV 1.0 rules-based good for sensor sales but flawed for data flywheel/constant learning/OTAs—hard with arms-length transaction). Vertically integrated stacks have big advantages.

Differentiation in Autonomy Models19:30

Do you think autonomy models from few companies that succeed will be fundamentally different? In AI ecosystem, language foundation models converging because training on same internet data. Driving has no internet of data—need robust sensor set and car park. Tesla has largest car park. Rivian has higher perception capability: better cameras (incredible low/bright light, dynamic range, more cameras/megapixels), radar for object detection, LiDAR with R2. Not as heavy as Waymo but heavier than Tesla for robust data platform per vehicle, growing with R2 expansion. No common data, but datasets similar; differences in data acquisition decisions. Safety primary; driving feel/UI differences, e.g., three settings: mild, medium, spicy (spicy most aggressive). This will become key decision: how does the vehicle behave.

Personalized Driving Heuristics and Preferences22:35

Vehicle behaves according to heuristics that drive like you. Overall model trained to perform safely but learns driving preferences and creates a model around you. In a world where you never drive, set preferences like aggressively changing lanes or residing in the right-hand lane. These are product/UI decisions. Ability to collect preferences. Different companies like Tesla and Rivian may make different decisions.

R2 as Mass Market Product23:32

R1 is flagship product with average selling price around $90,000. R1S is best-selling premium electric SUV in the country (over $70,000) and best-selling premium SUV (electric or non-electric) in California. Out sells Tesla Model X 2-to-1. Price limits volume. R2 is first truly mass-market product, pricing starts at $45,000, targeting $45,000-$55,000 range (average new car price in US is $50,000). Few great choices in this range; Tesla Model 3 and Model Y dominate with ~50% EV market share. Untapped opportunity to pull customers from ICE vehicles with unique characteristics.

Reasons for Slow EV Adoption in US25:01

EV adoption rate in US is around 8%. Vast majority of buyers purchase under $70,000, average $50,000. Over 300 vehicle model choices under $70,000 (hatchbacks, minivans, SUVs, two-seaters, convertibles). EV space has less than three great choices; Tesla Model 3 and Model Y are the main ones. Extreme lack of choice, especially non-Tesla form factors. Many non-Tesla EVs mimic Model Y side profile. > "If you want a Model Y, buy a Model Y." Companies create versions of Model Y instead of unique options. World doesn't need another Model Y; needs variety for personal expression and self-identification.

EV adoption reflects lack of choice. Partnership with Volkswagen Group enables technology platform across brands, form factors, price points, segments for broader adoption.

Develop vehicles like R2 as best possible (performance, range, dynamics, storage), drawing buyers into EVs. Same with R1; vast majority of customers are first-time EV owners.

Future Relationship with Cars and Vehicles29:30

Cars enable personal freedom, exploration, self-expression. Affinity unlike refrigerators. Will evolve with robo-taxis. Rivian views products through enabling and inspiring memorable experiences (photograph-worthy). Beyond function (fitting pets, gear, friends), inspire action. Design decisions like flashlight in door invite night exploration; treehouse engages imagination.

Closing31:30

Congrats on R2 and autonomy program.