Rivian CEO RJ Scaringe: By 2030, Not Having Self-Driving Will Be Like Not Having Airbags

By 2030, buying a car that can’t drive itself will feel as absurd as buying one without airbags. That’s not a hopeful projection from RJ Scaringe; it’s a structural argument built on how the underlying technology has shifted, and why the companies that don’t make the leap will shrink to nothing.

The Episode

Sarah Guo sits down with Rivian’s founder and CEO for a compact 30-minute conversation on No Priors. The discussion moves through Rivian’s autonomy strategy, the decision to build a custom inference chip, what software-defined architecture actually means, the R2 mass-market launch, and why American EV adoption is stuck at 8%. Scaringe is unusually direct about the competitive landscape, naming specific companies and putting hard numbers on who can and can’t survive.

The Great Reset: From Rules to Neural Nets

Rivian launched its R1 vehicles in late 2021 with what Scaringe calls a “1.0 approach” to autonomy: a third-party front-facing camera feeding a rules-based planner. The moment they shipped it, they knew it was wrong.

By early 2022, they made the decision to completely reset the platform. Not iterate. Reset. Not a single line of shared code, not a single piece of common hardware carried over to the second generation. The Gen 2 hardware shipped in mid-2024.

The reason the decision wasn’t hard, Scaringe explains, is that the cost of being wrong was existential. If driving is the core of transportation, and self-driving is the future of driving, then getting this wrong isn’t a product mistake. It’s a company-ending one.

The broader industry faced the same inflection. When transformer-based architectures arrived, the shift from rules-based to neural net-based autonomy wasn’t gradual. It was a complete rethink. Companies that had invested billions in classical systems faced an uncomfortable truth: the vast majority of that work was “pure throwaway.”

The 1-to-5 Club

Scaringe draws a sharp line around who can actually build autonomous driving. Outside China, he estimates more than one, less than five companies have the necessary ingredients. Probably closer to three.

The ingredients:

• Complete control of the perception platform. No intermediary company processing information. Raw sensor signals feeding directly into your system.
• A large enough car park. Thousands of vehicles generating data, not hundreds. Every car on the road becomes part of the training fleet.
• A robust onboard inference capability. Either bought or built, but it needs to be in every car.
• The capital and GPU infrastructure to train models at scale.

When pressed for names, Scaringe confirms: Rivian, Tesla, and Waymo. “And there’s maybe one or two others in the mix.”

The critical point isn’t where these companies are today. It’s whether they have the ingredients to keep improving at a high rate over the next four to five years. Companies stuck in 1.0 frameworks have, in his words, “a truly 0% chance of progressing to be competitive.”

Why Rivian Built Its Own Chip

The perception stack (cameras, radars, LiDAR) is cheap. Radars are extremely cheap. LiDAR is now very cheap. The expensive part is the brain: the onboard inference compute.

“The really expensive part of the system is actually the onboard inference. It’s an order of magnitude more expensive than any of the perception stack.”

People focus on perception because it’s visible, but the compute cost is what determines whether autonomy can ship on every car or only on premium trims. Rivian brought chip design in-house specifically to remove that cost barrier and deploy autonomy across the entire fleet.

The 60-Year Mess: Why Most Cars Can’t Update

Scaringe offers the clearest explanation of why traditional car companies are structurally disadvantaged. It goes back to fuel injection systems in the 1960s.

When cars first got computers, automakers outsourced them to suppliers. Then every new computerized function got its own little ECU (electronic control unit), written by a supplier, often by a supplier to the supplier. A modern car has 100-150 of these independent islands of software, each written by different teams.

The coordination cost is enormous. A simple feature like “car recognizes you approaching and adjusts seats, HVAC, lights, and audio to your preferences” touches maybe 10 different ECUs from 10 different vendors. Updating that sequence is practically impossible.

Rivian (and Tesla) use a zonal architecture instead: a small number of computers running one operating system controlling everything. Rivian pushes monthly OTA updates, adding features and refinements. Customers actually get excited about updates.

This architecture gap is so significant that Volkswagen Group, the world’s second-largest car company, signed a $5.8 billion licensing deal to adopt Rivian’s network architecture across their brands.

Scaringe frames two existential requirements for any car company operating at scale:

1. A software-defined architecture that enables continuous improvement
2. Very high levels of vehicle autonomy

Without both, the company shrinks. The options are build it yourself (hard, because traditional automakers lack the organizational DNA), or find a third party (scarce, and mostly selling 1.0 solutions).

The Autonomy Perception Gap

The levels-of-autonomy framework (L2, L3, L4) is breaking down. The hardware distinction that once separated them has collapsed: camera-heavy L2 systems and LiDAR-heavy L4 systems are converging into the same neural net approach.

For consumers, this creates confusion. A Level 2 system and a Level 4 system feel identical for 99.9999% of driving. The difference lives in the fifth, sixth, or seventh nine: extreme corner cases that are rare but potentially fatal.

Rivian’s strategy sits between Tesla (camera-only) and Waymo (perception-heavy). More cameras with better dynamic range, more megapixels, radar for object detection, and LiDAR on the R2. The LiDAR isn’t just for driving; it turns every car into a ground-truth training platform, identifying what distant specks actually are so the camera-only model can learn.

Driving Personality as Product

An interesting product insight: Rivian already offers three driving modes, mild, medium, and spicy. But Scaringe sees this evolving toward personalized driving models that learn individual preferences, how aggressively you change lanes, whether you prefer the right lane, your following distance.

“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.”

In a world where people rarely drive themselves, these preferences become the primary way car brands differentiate. It’s less about the technology and more about the user interface of autonomy.

The Choice Problem: Why US EV Adoption Is at 8%

Scaringe reframes the EV adoption question entirely. It’s not that Americans don’t want EVs. It’s that there are almost no choices.

The numbers: under $70,000 (the mass market), there are over 300 different ICE vehicle models available. In the EV space, there are “more than one, less than three” great choices, essentially just the Tesla Model 3 and Model Y. Which together capture roughly 50% of the entire US EV market.

Worse, the competitors that do exist are all copying the Model Y. Scaringe describes a design sketch showing the side profiles of every Model Y competitor: they’re nearly identical. “The world doesn’t need another Model Y. The world needs another choice.”

The R2, starting at $45,000, is Rivian’s answer: a mass-market vehicle designed to be the best possible car (not the best possible EV), pulling customers out of ICE vehicles because the product is simply more compelling. On R1, the vast majority of buyers are first-time EV owners, which validates the approach. If they were just shuffling customers between EV brands, it wouldn’t accomplish the mission.

Cars as Inspiration Machines

Scaringe ends with a philosophical observation about why cars occupy a unique emotional space. Unlike a refrigerator, a car enables personal freedom. It’s an expression of self. People self-identify with what they drive.

Rivian’s design philosophy centers on this: vehicles should not just enable experiences but inspire them. A flashlight built into the door panel is an invitation to explore at night. Every design detail is meant to nudge owners toward “the kinds of things you’d want to take photographs of.”

A Few Observations

• The “1 to 5 companies” framing is the most specific competitive assessment I’ve heard from an automaker CEO. Scaringe isn’t hedging; he’s saying most of the industry is structurally incapable of surviving.
• The chip decision reveals something about Rivian’s ambition. Building custom silicon isn’t a cost optimization. It’s a bet that autonomy needs to be standard equipment, not a premium add-on.
• The fuel injection ECU history is the kind of detail that makes structural problems legible. It explains why even well-funded legacy automakers can’t just “add software.”
• Framing slow EV adoption as a supply-side choice problem rather than a demand-side preference problem is counterintuitive but well-argued. 300+ ICE choices versus 2-3 good EV choices is a striking stat.
• The Volkswagen deal ($5.8B) might be the most underappreciated signal here. It suggests at least one major legacy automaker has accepted they can’t build the software stack themselves.