Rivian details in-house autonomy chip and next-generation compute system

Rivian announced a series of hardware and software developments intended to support its long-term autonomy and artificial intelligence strategy during its first Autonomy & AI Day event, held at its Palo Alto offices. The company detailed progress on in-house silicon, next-generation compute systems, software architecture, sensor integration, and future consumer-facing features. According to Rivian, these updates represent a shift toward deeper vertical integration across its vehicle platforms, although rollout timelines remain several years away for most capabilities.

Rivian Founder and CEO RJ Scaringe characterized the advancements as part of the company’s effort to expand its self-driving roadmap and eventually support Level 4 capability. Scaringe said Rivian’s updated hardware platform, which includes an in-house 1600 sparse TOPS inference chip, is intended to enable improvements in autonomous operation. He stated that Rivian’s goal involves offering customers a system that can manage more driving tasks and reduce driver involvement, though Rivian did not announce when Level 4 capability may be commercially available or legally permitted. The company focused primarily on the development of enabling technologies and internal tools rather than consumer deployment dates.

A central component of Rivian’s strategy is the move to in-house silicon. The company introduced the first generation of its Rivian Autonomy Processor, known as RAP1. RAP1 is a 5-nanometer processor designed for what Rivian describes as a “vision-centric physical AI” system. The processor integrates compute and memory into a single multi-chip module, which the company says improves efficiency and supports Automotive Safety Integrity Level requirements. Rivian explained that vertical integration at the chip level provides greater control over architecture choices and optimization, although independent evaluations of RAP1 are not yet available.

RAP1 is used in Rivian’s third-generation Autonomy computer, the Autonomy Compute Module 3 (ACM3). Rivian stated that ACM3 can deliver 1600 sparse INT8 TOPS of performance and process up to five billion pixels per second. The module incorporates RivLink, a low-latency interconnect that enables multiple chips to be linked to scale processing power. The system also operates using Rivian’s in-house AI compiler and platform software, which are intended to support more efficient model deployment and data processing.

The company reiterated that this third-generation hardware is under validation and is expected to reach production on Rivian’s upcoming R2 platform no earlier than the end of 2026. Rivian did not specify whether ACM3 or RAP1 will be retrofittable to existing vehicles, nor did it provide a detailed explanation of how the validation process is being conducted.

Rivian also confirmed that LiDAR will be added to future R2 vehicles. Current Rivian models rely primarily on camera-based vision systems. According to Rivian, LiDAR integration will offer redundancy and improve spatial mapping for scenarios that challenge camera-only systems. The company characterized LiDAR as an enhancement particularly useful for detecting objects and structures in low-visibility conditions or complex environments. LiDAR will form part of a multi-modal sensor suite that Rivian says is necessary for higher levels of autonomy. Specific technical specifications for Rivian’s LiDAR system were not disclosed.

Alongside hardware developments, the company detailed its software autonomy roadmap, which it refers to as the Rivian Autonomy Platform. The platform supports an end-to-end data loop used to train Rivian’s autonomous driving models. Rivian introduced a Large Driving Model (LDM), which it described as analogous in design approach to a Large Language Model, but focused on driving tasks. The company stated that the LDM uses Group-Relative Policy Optimization for training and is intended to convert large-scale driving datasets into operational vehicle behaviors. Rivian did not specify the size of its dataset or how frequently models are updated.

Rivian also said that second-generation R1 vehicles will receive updated autonomy features in the near term. One of the forthcoming features is Universal Hands-Free (UHF), which Rivian describes as a hands-free assisted driving capability designed to operate for extended periods. Rivian reported that UHF will function on more than 3.5 million miles of roads in the United States and Canada. The system will be capable of operating off-highway when lane markings are clearly visible. Rivian did not specify release timing beyond stating that the software would arrive “in the near term.”

The company also announced an autonomy subscription package called Autonomy+. The subscription is scheduled to launch in early 2026 and will be priced either as a $2,500 one-time purchase or a $49.99 monthly plan. Rivian said that the subscription will provide access to expanding autonomy capabilities as software evolves. The company did not specify which features would be included at launch or which would be added later, although it referenced a trajectory toward point-to-point navigation, eyes-off operation, and eventual personal Level 4 capability for Gen 2 R1 and future R2 vehicles. These capabilities will depend on regulatory approvals, validation processes, and software readiness.

Beyond autonomy, Rivian outlined how it plans to use AI throughout its business operations under an internal system it calls Rivian Unified Intelligence (RUI). RUI consists of a shared multi-modal data architecture designed to support a range of applications, including voice interaction, diagnostics, predictive maintenance, and customer support tools. Rivian emphasized that RUI integrates multiple large language models and internal datasets to support its systems.

One consumer-facing element of this platform is the Rivian Assistant, a voice-enabled interface scheduled to launch in early 2026 on first- and second-generation R1 vehicles. The assistant is designed to interpret commands, interact with vehicle systems, and connect with third-party services. Rivian named Google Calendar as its first planned integration. According to the company, the assistant uses Rivian’s edge-based models for contextual understanding, augmented with larger foundation models for reasoning and natural-language interaction. Rivian said it built an “agentic framework” to enable the assistant to coordinate tasks across vehicle systems and external applications. Additional integrations were not announced.

Rivian also said that RUI will support new service and maintenance features. These include diagnostic tools for technicians that analyze telemetry and historical data to identify issues. Rivian stated that the same intelligence will be incorporated into its mobile app to allow customers to perform certain self-diagnostic tasks. The company did not specify how these tools might affect service-center operations, repair timelines, or warranty processes.

Rivian described its autonomy, AI, and software developments as part of a broader effort to unify vehicle hardware, software, and backend tools under a vertically integrated strategy. The company stated that this integration allows it to update core technologies more rapidly and support long-term feature development. Rivian did not provide cost projections, regulatory milestones, or market availability beyond the estimated timelines for R2 hardware and the 2026 release window for subscription features and voice assistance.

The announcements signal Rivian’s intention to build proprietary solutions across its autonomy stack, though commercial deployment remains dependent on validation, regulatory approval, and production readiness over the next several years.