Rivian unveils RAP1 AI chip, LiDAR-ready ACM3, Universal Hands Free

Rivian just staked a bold claim in automated driving by unveiling its in-house RAP1 AI chip and the Autonomy Compute Module 3. Often pigeonholed as a Tesla alternative, the American automaker has been quietly building a business around software and services—and now it is signaling that the real contest will be won not only with hardware, but with the brains of the car.

At the core of the new stack is the first-generation Rivian Autonomy Processor, a fully internal design on the Armv9 architecture with fourteen Cortex-A720AE cores, fabricated on a 5 nm process. The chip introduces RivLink, a proprietary high-speed interface that lets Rivian scale performance by linking additional chips. Details remain sparse, but the direction clearly points to a modular future for in-vehicle computing.

RAP1 is the heart of the Autonomy Compute Module 3—an onboard computer tailored for autonomous driving tasks. Rivian cites up to 1,800 TOPS in INT8 and the capacity to process up to five billion camera pixels per second. A notable departure from Tesla’s philosophy is official LiDAR support. Rivian confirms ACM3 is being engineered to work with LiDAR, and says that pairing is already under validation for future versions of the R2 crossover slated for late 2026.

For current second‑generation R1 owners, the company is preparing Universal Hands Free—a driver-assistance system conceptually comparable to Autopilot. It’s designed for extended use, covers about 3.5 million miles of roads across the U.S. and Canada, and operates not only on highways but also on standard roads with clear lane markings. Those seeking more capability will be able to opt for the Autonomy+ package, available as a $2,500 one-time purchase or a $49.99 monthly subscription. Rivian is careful to avoid grand promises of “full self-driving,” emphasizing safety first and presenting this as a measured step toward Level 4 autonomy—a cautious stance that, frankly, feels wise.

Rivian is also leaning on its own Large Driving Model, akin to language models. The company says it uses a Group Relative Policy Optimization approach, similar to techniques seen in DeepSeek, and positions this as a training advantage for its driver-assistance stack. Against rumors that Tesla’s forthcoming HW5 could deliver 2,000–2,500 TOPS without LiDAR, Rivian’s path looks deliberately different. Coupled with a profitable $5.8 billion software-and-architecture joint venture with Volkswagen, the strategy hints at a broader ambition: in time, Rivian could package and sell its autonomy technology beyond its own lineup, potentially shifting its role from pure automaker to platform supplier.