Skip to content
No PriorsNo Priors

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

Autonomous vehicle technology has moved past human-coded rules and into an era of neural networks and custom computer chips. And to solve the most difficult driving scenarios, electric vehicle company Rivian abandoned its original technology platform to build a vertically integrated data stack. Sarah Guo sits down with Rivian Founder and CEO RJ Scaringe to explore the seismic shift in the automotive industry toward AI-driven, software-defined vehicles . RJ discusses the move away from function or domain-based architecture for vehicle electronic systems to software-defined architecture, which allows for dynamic, monthly updates to features in Rivian’s vehicles. RJ also talks about the upcoming launch of Rivian’s R2 model, which aims to be a distinct, affordable, mass-market alternative to the Tesla Model Y. Plus, RJ shares his vision for a future where vehicles don’t just drive us, but inspire personal freedom and exploration. Sign up for new podcasts every week. Email feedback to show@no-priors.com Follow us on Twitter: @NoPriorsPod | @Saranormous | @EladGil | @RJScaringe | @Rivian Chapters: 00:00 – Cold Open 00:35 – RJ Scaringe Introduction 0:58 – Rivian’s Autonomy Evolution 05:19 – Why Rivian’s Tech is Vertically Integrated 10:06 – Levels of Autonomous Driving Technologies 14:00 – Importance of a Software-Defined Architecture 19:28 – Differentiating Autonomous Vehicle Models 23:20 – R2: The First Mass Market Autonomous Vehicle 25:02 – Do Americans Want EVs? 29:05 – How Our Relationship to Vehicles is Evolving 30:45 – Conclusion

RJ ScaringeguestSarah Guohost
Feb 11, 202631mWatch on YouTube ↗

At a glance

WHAT IT’S REALLY ABOUT

Rivian’s vertically integrated AI stack for autonomy and mass-market EVs

  1. Rivian rebooted its autonomy program after launching R1 with a rules-based, third-party camera approach, moving to a clean-sheet, neural-network-centric architecture starting with Gen 2 hardware in 2024.
  2. Scaringe argues only a handful of companies have the ingredients to win in autonomy: full sensor/control of data collection, a large vehicle fleet for a data flywheel, and massive onboard and training compute—leading Rivian to build an in-house inference chip.
  3. He claims the auto industry cannot compete at scale without a software-defined, zonal architecture that enables frequent OTA updates, and cites Rivian’s $5.8B VW Group deal as evidence legacy OEMs need a new electronics/software foundation.
  4. R2 is positioned as Rivian’s first true mass-market vehicle (~$45k starting) and a key lever for fleet growth and autonomy training, while EV adoption in the U.S. is framed as primarily a “lack of choice,” not lack of consumer interest.

IDEAS WORTH REMEMBERING

5 ideas

Rivian treated early autonomy as a throwaway and restarted fast.

R1 launched with a “1.0” rules-based stack and third-party camera; by early 2022 Rivian chose a clean-sheet rewrite with no shared code/hardware, first appearing in Gen 2 vehicles (mid-2024).

Modern autonomy progress depends on AI architecture, not rules.

Scaringe describes a step-change driven by transformer-era approaches: older systems were “machine vision plus rules,” while current systems are trained neural networks that improve via data and iteration.

Vertical integration is a prerequisite for a serious autonomy flywheel.

Rivian wants raw sensor control (cameras/radar/lidar), onboard event triggering and storage, high-bandwidth uploads (often Wi‑Fi), and internal training pipelines—hard to achieve with arm’s-length suppliers.

Onboard inference is the real cost bottleneck, more than sensors.

He argues cameras/radars/lidars have become cheap, but running large models in-car is “an order of magnitude” more expensive than the perception hardware—motivating Rivian’s custom inference chip to put autonomy-capable compute in every vehicle.

Autonomy “levels” feel identical until rare corner cases.

For “99.9999%” of driving, L2/L3/L4 can appear similar; the differentiator is coverage of extreme edge cases and the safety case around them—best addressed by large fleets, simulation, and continual learning.

WORDS WORTH SAVING

5 quotes

By twenty thirty, it'll be inconceivable to buy a car and not expect it to drive itself.

— RJ Scaringe

The moment we launched, we knew it was the wrong approach.

— RJ Scaringe

Radars are extremely cheap, LiDARs are very cheap, but the really expensive part of the system is actually the onboard inference.

— RJ Scaringe

The companies that do this well will exist. The companies that don't do this well... will not exist.

— RJ Scaringe

The world doesn't need another Model Y; the world needs another choice.

— RJ Scaringe

Reset from rules-based autonomy to neural netsVertical integration: sensors, compute, softwareData flywheel: fleet telemetry, event triggering, uploadsOnboard inference cost and Rivian’s in-house chipLevel 2/3/4 convergence and corner-case safetySoftware-defined (zonal) vehicle architecture and OTAsR2 as mass-market scale driver; EV adoption via choice

High quality AI-generated summary created from speaker-labeled transcript.

Get more out of YouTube videos.

High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.

Add to Chrome