Rivian CEO Bets on Vertical Integration for Autonomy
Rivian's CEO RJ Scaringe stated that the company is abandoning rule-based systems for an end-to-end neural network approach to autonomy. He argued that only 3-4 companies with vertically integrated data, hardware, and proprietary AI inference chips will survive in the autonomous vehicle market. Scaringe identified the onboard inference computer as the most expensive component of the entire system.
- Rivian's current generation of vehicles, including the R1T and R1S, utilize a hardware suite with 11 in-house-developed high-resolution cameras, five radars, and dual Nvidia Drive Orin processors for their "Rivian Autonomy Platform". The move to a proprietary inference chip marks a significant shift away from relying on third-party processors from companies like Nvidia. - The traditional "rule-based" or modular approach to autonomy involves separate, hand-coded systems for perception, prediction, and planning. An end-to-end neural network, in contrast, processes raw sensor data (like camera images) and directly outputs vehicle controls (steering, acceleration, braking) from a single, large AI model trained on vast amounts of driving data. - This strategy of vertical integration—encompassing custom silicon, in-house software, and data from a customer fleet—directly mirrors the approach taken by Tesla. Tesla developed its own "Full Self-Driving" (FSD) computer with a custom-designed neural network chip to power its Autopilot and FSD features. - To power its next-generation system, Rivian is developing an in-house processor called the "Rivian Autonomy Processor" (RAP1). This custom silicon will be part of the new "Autonomy Compute Module 3," which the company states can process 5 billion pixels of sensor data per second. - Unlike Tesla, which famously relies on a vision-only system, Rivian's future sensor suite will incorporate lidar in addition to cameras and radar. The upcoming R2 vehicle is planned to be the most affordable vehicle in the U.S. to launch with lidar. - The development of custom AI chips is a capital-intensive endeavor, with industry estimates for designing a new, advanced-node chip ranging from hundreds of millions to over $500 million. This high barrier to entry supports the CEO's claim that few companies will be able to compete with this strategy. - Rivian plans to monetize its more advanced autonomous features through a subscription service called Autonomy+. This service, enabled by the new vertically integrated stack, will offer features like "Universal Hands-Free" driving and will be available for a one-time payment or a monthly fee.
