US automotive company Rivian is advancing its vertically integrated vehicle technology through a proprietary, purpose-built chip.
The company made the announcement, which also included an updated autonomy roadmap and an artificial intelligence-led software architecture, at its first Autonomy & AI Day in Palo Alto.
Rivian founder and CEO RJ Scaringe said: “Our updated hardware platform, which includes our in-house 1600 sparse TOPS inference chip, will enable us to achieve dramatic progress in self-driving to ultimately deliver on our goal of delivering L4.”
Central to the plan is the move to in-house silicon built for a vision-centric physical AI approach.
The company has introduced its first-generation Rivian Autonomy Processor (RAP1), a custom 5nm processor combining processing and memory within a single multi-chip module.
Rivian said the design is intended to improve efficiency and performance while meeting Automotive Safety Integrity Level requirements.
According to Bloomberg, Taiwan Semiconductor Manufacturing Company will produce the chips.
RAP1 is used in Rivian’s third-generation autonomy computer, the Autonomy Compute Module 3 (ACM3).
ACM3 is designed to deliver 1,600 sparse INT8 TOPS and process five billion pixels per second.
The company also highlighted RivLink, a low-latency interconnect intended to link chips and scale processing capacity, and said RAP1 is supported by an in-house AI compiler and platform software.
Rivian also plans to add LiDAR to future R2 models as part of a multi-modal sensor approach, with the sensor intended to provide three-dimensional spatial information and redundancy to improve detection in edge-case driving scenarios.
Gen 3 autonomy hardware, including ACM3 and LiDAR, is currently in validation, with shipments for R2 models expected to begin at the end of 2026.
On the software side, the company described a “software-first” approach supported by the Rivian Autonomy Platform and an end-to-end data loop for training.
Rivian introduced a large driving model (LDM), described as a foundational autonomy model trained in a similar way to a large language model (LLM).
The company said it is using Group-Relative Policy Optimisation (GRPO) to distil driving strategies from large datasets for vehicles.
Near-term updates are planned for second-generation R1 vehicles, including Universal Hands-Free (UHF), which Rivian said will expand hands-free assisted driving for longer periods to more locations.
The company also announced an autonomy subscription offering, Autonomy+, due to launch in early 2026.
Pricing is set at $2,500 as a one-time payment or $49.99 per month.
The company also set out an autonomy trajectory for Gen 2 R1 and future R2 vehicles, encompassing point-to-point capability, “eyes off” driving and personal Level 4.
Beyond autonomy functions, Rivian said it is applying AI across operations via Rivian Unified Intelligence (RUI), described as a shared data foundation using multiple modalities and LLMs.
Rivian said the architecture will support new features, improvements to service infrastructure and predictive maintenance.
A key component is the Rivian Assistant, a voice interface scheduled for early 2026 on Gen 1 and Gen 2 R1 vehicles.
Rivian also said RUI will be used in service diagnostics, acting as an assistant for technicians by scanning telemetry and vehicle history to help identify complex issues.
