Rivian showed up in Palo Alto on Thursday with a tidy, ambitious story: build the silicon, train the models, add LiDAR — and sell autonomy as a service. The EV maker unveiled a house‑designed AI chip and a next‑gen autonomy stack that it says will power hands‑free driving on millions of miles of roads and eventually push the company toward Level 4 robotaxis.
The headlines and the hard numbers
At the center of Rivian’s presentation was the Rivian Autonomy Processor (RAP1) and a Gen 3 autonomy computer (ACM3). The company quoted the platform at roughly 1,600 sparse INT8 TOPS and a pixel throughput in the billions per second. RAP1 uses a multi‑chip module approach, connects chips with a low‑latency fabric called RivLink, and — Rivian says — pairs tightly with an in‑house AI compiler and software stack.
Rivian also revealed specifics about its feature rollout and pricing: a new Autonomy+ subscription that will be available early next year for a one‑time price of $2,500 or $49.99 per month. A near‑term software update will expand “Universal Hands‑Free” driving to about 3.5 million miles of roads across North America.
Hardware changes aren’t cosmetic: Rivian plans to add LiDAR to its upcoming R2 models for redundant three‑dimensional sensing, and it will source chips from TSMC. Bloomberg and other reports emphasized that the move reduces Rivian’s dependence on Nvidia for inference hardware — a notable shift in the industry.
Software: a driving model, an assistant, an ecosystem
Rivian described a layered AI architecture. A Large Driving Model (LDM) — trained in ways analogous to large language models — is meant to distill robust driving strategies from massive datasets. That LDM will live alongside Rivian Unified Intelligence (RUI), a multi‑modal AI foundation the company says will touch diagnostics, predictive maintenance, and in‑car reasoning.
The company also teased the Rivian Assistant, an in‑car conversational agent that will be embedded in vehicles and integrate with third‑party services like calendars. The assistant will run on edge models Rivian developed itself, “augmented” by external foundation models for broader language and reasoning tasks.
If in‑car assistants are becoming table stakes, this move sits next to broader industry shifts: automakers and tech companies are embedding conversational and agentic features into maps and devices. For context on how navigation and assistant experiences are evolving, see the recent work on Google Maps’ conversational copilot and Apple’s decision to lean on Gemini for Siri improvements in coming OS releases Apple to Use a Custom Google Gemini Model to Power Next‑Gen Siri.
Level 4 ambitions — and the long road to robotaxis
Rivian explicitly used the term Level 4 when describing its aim: vehicles that can operate without human intervention in defined conditions. That’s the same plateau Waymo and other robotaxi projects target. Rivian says its sensor fusion — cameras, radar, and now LiDAR — plus its RAP1 silicon and LDM will enable a path toward point‑to‑point and eyes‑off capabilities over time.
Practical rollout will be incremental. The company plans to ship expanded hands‑free and Level 2+/3 experiences first, then push toward more ambitious features as validation and regulation permit. CEO RJ Scaringe flagged rideshare and robotaxi opportunities as follow‑on markets once the stack matures.
Why Rivian is making this bet now
A few pressures explain the urgency. EV sales in the U.S. have cooled after the phase‑out of a $7,500 federal credit, and Rivian has felt investor pressure to show new revenue streams and technological differentiation. Offering autonomy as a subscription is an obvious attempt to convert software advances into recurring cash. The $2,500 one‑time or $49.99 monthly price positions Autonomy+ below Tesla’s more expensive FSD option, which could help adoption if features match expectations.
Longer term, owning silicon and software is a vertical play: it reduces reliance on third‑party suppliers (and their margins) and creates IP Rivian could potentially license or reuse in partnerships — a strategy not dissimilar to Tesla’s early silicon work or other automakers’ attempts to internalize critical stacks.
The catch: validation, safety and competition
Specs and demos are one thing; real‑world driving is another. Rivian needs to prove its models generalize to edge cases, its LiDAR integration plays nicely with camera and radar fusion, and the silicon delivers efficiency and safety at automotive grade. Regulators and insurers will also demand evidence before eyes‑off or full Level 4 services can be widely used.
Competition is fierce. Legacy automakers, Tesla, Waymo and others are all jockeying with different technical philosophies — Tesla’s camera‑first approach versus the LiDAR‑backed stacks favored by Waymo and now Rivian. And chip incumbents, notably Nvidia, still dominate much of the industry’s compute roadmap; Rivian’s move away from that ecosystem is bold but raises questions about software compatibility and tooling.
Rivian’s presentation was both a technology reveal and a business pitch: silicon, sensors, and subscriptions wrapped into a narrative that promises to turn in‑car time into something more useful. Whether customers, regulators and fleets buy that future will be decided on roads, not stages. For now, Rivian has placed a big bet on doing more of the stack itself — and put a price tag on the right to not hold the wheel.