Nvidia Automotive: DRIVE Chips and Self-Driving Cars
What does Nvidia do for cars? Nvidia supplies vehicle computers, software and simulation tools through its DRIVE platform. DRIVE Orin runs many current driver-assistance systems, while Thor is the next central computer for cockpit and automated-driving workloads. Nvidia does not build the self-driving car or own its safety case; each automaker integrates and validates the complete system.
What is Nvidia Automotive?
Nvidia Automotive is the company's vehicle-computing business. It does not build a self-driving car. It sells the computing platform on which automakers and autonomy developers run cockpit, driver-assistance and automated-driving software. That platform spans chips in the vehicle, the DRIVE operating and development software, DRIVE Sim for virtual testing, and data-centre systems for training AI models.
What does Nvidia make for cars?
The stack has four distinct jobs. DRIVE AGX is the in-car computer. DRIVE OS and related software provide the base on which a manufacturer builds its own functions. DRIVE Sim lets teams expose the system to synthetic roads, traffic and edge cases. Nvidia's data-centre hardware trains the perception and planning models before they are deployed to the vehicle.
| Layer | Product role | What the carmaker still owns |
|---|---|---|
| Vehicle compute | Orin or Thor processes cameras, radar, lidar and cockpit workloads | Vehicle integration, thermal design and safety case |
| Base software | Drivers, operating environment and development tools | Driving policy, user experience and validation |
| Simulation | Repeatable virtual scenarios and sensor simulation | Scenario coverage and acceptance criteria |
| AI factory | Training and data processing in the data centre | Data quality, model choice and release control |
Nvidia DRIVE Orin vs DRIVE Thor
Orin is the current workhorse used in many production ADAS and autonomy programmes. Thor is intended to consolidate more work on one central computer, including automated driving, parking, driver monitoring and cockpit AI. A larger TOPS number is not a safety rating. Memory bandwidth, power, sensor interfaces, software efficiency, redundancy and the validated vehicle system matter as much as peak arithmetic throughput.
What happens inside an autonomous-driving computer?
Camera, radar and lidar streams first pass through sensor interfaces and pre-processing. Perception software turns pixels and returns into lanes, free space, vehicles and pedestrians. Fusion reconciles those observations over time. Prediction estimates what other road users may do, planning selects a path, and control converts that path into steering, acceleration and braking requests. A supervisory layer watches timing and health throughout the chain.
Nvidia accelerates many of those workloads, but the chip does not decide whether a gap is safe. The developer chooses the sensor set, neural networks, confidence thresholds, fallback behaviour and operating domain. The carmaker also has to prove that data arrives within its time budget and that a failure is detected before it becomes hazardous. Silicon is an enabler. Driving policy and safety evidence remain vehicle-programme work.
Why TOPS is a poor shortcut for comparing car computers
TOPS measures a theoretical number of arithmetic operations at a stated numeric precision and set of assumptions. It says little about how quickly a real model moves data through memory, whether sensor input can enter without bottlenecks, or how much performance remains inside the vehicle's thermal and power limits. Two computers with similar headline TOPS can behave very differently with the same perception network.
A serious comparison asks about memory bandwidth, supported data types, accelerator utilisation, CPU and GPU balance, video and network interfaces, deterministic latency, safety mechanisms and power consumption. It also asks whether the software toolchain can turn the hardware into a stable production system. A benchmark that ignores the model and workload is advertising, not an engineering decision.
Which carmakers use Nvidia Automotive?
Nvidia's customer and partner list spans passenger cars, trucks and robotaxis. In March 2026 Nvidia said BYD, Geely, Isuzu and Nissan were adopting DRIVE Hyperion for Level 4 development. Mercedes-Benz, Jaguar Land Rover and Volvo have also announced Nvidia-based programmes. An announcement does not mean every model uses the same chip, software or automation level, so each production vehicle must be checked separately.
How big is Nvidia's automotive business?
Automotive revenue reached $2.3 billion in Nvidia's fiscal 2026. Total company revenue was $215.9 billion, so automotive contributed about 1.1%. It is strategically visible but financially small beside data-centre computing. That distinction matters: Nvidia can be important to future vehicle architectures without cars being a material share of today's group revenue.
What Nvidia does not replace
A DRIVE development kit is not a production vehicle computer. A series design needs automotive-qualified components, redundant power and communication paths where required, a cooling concept, electromagnetic compatibility, manufacturing traceability, diagnostics and a service strategy that may last well beyond a consumer-electronics cycle. It also needs an independent path to a safe state if the main compute chain misbehaves.
Nor does one supplier erase the rest of the E/E architecture. Small body actuators can stay on LIN, real-time controllers remain on CAN or Ethernet, and zone controllers handle local I/O and power. Nvidia competes for the high-compute layer. The surrounding network, safety concept and vehicle functions still involve the automaker and many other suppliers.
How Nvidia fits into the automotive engineering stack
The chip is one layer, not the whole vehicle. Start with the automotive E/E architecture that connects it, then the software-defined vehicle model that updates it. For driving responsibility, compare Level 2 with Level 3, the complete SAE L0-L5 framework and Waymo's different Level 4 robotaxi system.
Place those computers inside the complete set of vehicle systems, then follow how hardware and software pass through the automotive development process. The Tesla Robotaxi vs Waymo comparison shows why a compute platform alone does not determine deployment scale or safety evidence.
Primary sources, checked July 24, 2026: Nvidia's fiscal 2026 results report automotive and total revenue. Its DRIVE Hyperion announcement identifies the cited Level 4 adopters. The official DRIVE AGX page documents the current developer platform and its interfaces.
Nvidia Automotive: frequently asked questions
What is Nvidia Automotive?
It is Nvidia's vehicle-computing business: DRIVE chips and computers in the car, base software and development tools, simulation, and data-centre systems for AI training.
What does Nvidia make for cars?
Nvidia makes DRIVE AGX vehicle computers based on chips such as Orin and Thor, plus operating, development and simulation software. Automakers add the vehicle functions and validate safety.
Which carmakers use Nvidia?
Announced customers and partners include Mercedes-Benz, Jaguar Land Rover, Volvo, BYD, Geely, Isuzu and Nissan. The exact hardware, software and automation level varies by programme.
How big is Nvidia's automotive business?
Automotive revenue was $2.3 billion in fiscal 2026, about 1.1% of Nvidia's $215.9 billion total revenue.
What is the difference between Nvidia Orin and Thor?
Orin is the widely deployed current DRIVE computer. Thor is designed to consolidate more automated-driving and cockpit work on one higher-performance central computer.
Is Nvidia building a self-driving car?
No. Nvidia supplies computing and software platforms. The automaker or autonomy developer creates the driving system, integrates it into a vehicle and owns the validation and safety case.
