Interview

RJ Scaringe on Rivian's in-house 800-TOPS autonomy chip and path to mass market with the $45K R2

Dec 15, 2025 with RJ Scaringe

Key Points

  • Rivian disclosed an in-house autonomy chip with 800 TOPS of compute and 35 billion transistors, manufactured by TSMC, targeting Level 4 self-driving capability in its Gen 3 platform.
  • The $45,000 R2 SUV represents Rivian's bet to reach mass market volume and justify years of heavy R&D spending after the R1 captured 35% share of US premium electric SUVs above $70,000.
  • Rivian's zonal software architecture, shared with Volkswagen Group via a $5.8 billion licensing deal, enables over-the-air updates across the entire vehicle without supplier coordination.
RJ Scaringe on Rivian's in-house 800-TOPS autonomy chip and path to mass market with the $45K R2

Summary

RJ Scaringe is betting Rivian's mass-market future on the R2, a five-passenger SUV starting at $45,000 that he describes as capturing the essence of the R1 platform at a fraction of the price. The R1S currently holds roughly 35% market share among electric vehicles priced over $70,000 in the US, outselling the Tesla Model X by a significant margin, and is the best-selling premium SUV — electric or otherwise — in California. Scaringe frames the R2 launch as the moment Rivian's heavy upfront R&D spending finally finds the volume to justify it.

Autonomy Hardware

Rivian disclosed its in-house autonomy chip in what Scaringe says did not leak despite development beginning in early 2022. The processor delivers 800 TOPS of compute, contains 35 billion transistors, and is capable of processing 5 billion pixels per second. It is manufactured by TSMC. The current Gen 2 fleet runs on an NVIDIA platform at approximately 200 TOPS, which Scaringe says is sufficient to build the data flywheel and deliver point-to-point autonomy. The Gen 3 platform enables larger, less-compressed neural net models to run at inference, which is the core architectural advantage over the current stack.

The new hardware also includes upgraded cameras and a LiDAR arm mounted at the top of the windshield. Scaringe is explicit that the LiDAR is used primarily to train cameras and the radar array rather than as the primary sensing modality, and that radar handles adverse weather conditions where optical systems degrade. The target capability is Level 4 autonomy — vehicle operation without a driver present.

Software Architecture as a Competitive Moat

Scaringe attributes Rivian's technology differentiation to a clean-sheet zonal compute architecture, contrasting it with legacy OEM vehicles that can carry 100 to 150 separate electronic control units from dozens of suppliers, each running independent code bases. In his framing, only Rivian and Tesla in the West, and fewer than 10 Chinese manufacturers, have executed this architecture correctly. The practical consequence is the ability to push software updates across the entire vehicle without supplier coordination — a prerequisite for AI integration.

That architecture underpins Rivian's $5.8 billion software licensing deal with Volkswagen Group, which covers the network architecture and operating system.

Data Infrastructure

Rivian's autonomy training runs on many thousands of GPUs drawing from trigger-event data collected across its Gen 2 fleet, which launched in 2024. Scaringe describes AI and the self-driving platform as now the dominant category of R&D spending, a shift from earlier years when investment was concentrated on high-voltage architecture and ECU-based software fundamentals.

Chinese EV Competition

Scaringe pushes back on the more breathless takes on Chinese EV competitiveness. He argues the cost advantage is structural — lower labor costs and government subsidies or free capital — rather than technological magic. On architecture, he credits a small subset of Chinese manufacturers (his estimate is more than five, fewer than ten) with having built genuinely clean-sheet software-defined vehicles. The majority of China's 100-plus EV brands he does not place in that category. The relevant risk, in his view, is architectural sophistication combined with a structurally lower cost base, not any particular feature or gimmick.

Market Strategy

Scaringe is deliberate about not chasing the very large SUV segment above the R1. He estimates the addressable market for vehicles larger than the R1 at roughly 10,000 units per year, calling it too small to justify investment at this stage. The R2 targets the two-row five-passenger SUV segment, which he describes as having effectively one highly compelling electric option under $50,000 today — the Tesla Model Y — against roughly 300 combustion alternatives. He argues competitors drew the wrong lesson from Model Y's success by building near-identical form factors rather than offering genuine differentiation, and that R2 is designed to occupy distinct territory in the segment.