Nvidia is positioning itself as the common supplier underneath the driverless-car industry, saying that every major commercial-scale robotaxi program runs on its stack. The company's pitch rests on an open platform that covers AI training, simulation and safety validation, rather than a single chip. It arrives as driverless fleets already carry passengers on busy streets, and as the robotaxi market is projected to reach $400 billion by 2035, with more than 6 million commercial vehicles expected in operation.
The architecture splits into three computers: one for training, one for simulation and validation, and one inside the vehicle. Training runs on Nvidia's DGX systems. For the vehicle itself, the company points to DRIVE Hyperion 10, which it describes as a level-4-ready reference architecture. That design pairs two DRIVE AGX Thor systems-on-chip, which Nvidia says are built on its Blackwell platform, to run vision-language-action models for perception, reasoning and planning. It also specifies 14 high-definition cameras, nine radars, three lidars and 12 ultrasonic sensors, fused for 360-degree coverage.
Read more: Nvidia Launches NemoClaw Security Stack for OpenClaw AI Agents
Nvidia has not disclosed a price for the robotaxi platform, and no launch date has been given for DRIVE Hyperion 10. What it does say is that the ecosystem now spans Asia, Europe, the Middle East and North America.
Uber is scaling a DRIVE Hyperion fleet and plans to reach 28 cities by 2028. Its collaboration with Nvidia pulls in a long list of partners, among them Lucid, Mercedes-Benz, Nissan, Nuro, Stellantis, Wayve and Zoox. May Mobility intends to run ride-hailing on the Uber network and builds its stack on DRIVE, while its vehicles already operate on Lyft's network in Atlanta using the same platform. Bolt is using Nvidia technology to develop and scale autonomous vehicles across Europe, and WeRide plans to bring its GXR to Southeast Asia through a Grab partnership.
Other names show how broadly the stack is being adopted. Waymo is working with Nvidia on an autonomous computing system, and Wayve, Nissan and Uber are developing a global robotaxi program around a prototype vehicle. Zoox relies on DRIVE for in-vehicle computing along with cloud training and simulation, while Momenta builds its software stack on DRIVE AGX running DriveOS. Pony.ai created a new domain controller using DRIVE Hyperion and DRIVE AGX Thor. Tensor is developing a level 4 Robocar with eight DRIVE AGX Thor chips inside, and Lenovo is supplying a DRIVE AGX Thor-based level 4 domain controller for the SWM robotaxi program.
The tooling side matters as much as the hardware. Nvidia's Alpamayo portfolio bundles open reasoning models, simulation frameworks and datasets, with the reasoning models aimed at the long-tail problems that trip up autonomous driving. Omniverse NuRec rebuilds real-world driving scenarios from sensor data, while Cosmos world foundation models generate physically based variations of those scenarios, both running on RTX PRO Servers. The AlpaSim framework trains and evaluates reasoning-based driving models. Nvidia also offers physical AI datasets, reinforcement learning blueprints and distillation recipes. Safety is handled through Halos and its Halos OS, which the company says spans inspection, validation, simulation and continuous testing from cloud to car.
This builds on earlier openings. Alpamayo 2 Super became available for commercial use on Hugging Face, shipping under the OpenMDW-1.1 license that covers fine-tuning, derivatives and commercial redistribution. Nvidia says it tops LingoQA in a field of nearly 40 models, and the Alpamayo family has passed 500,000 Hugging Face downloads. Nvidia's chief executive, Jensen Huang, has predicted that 2026 will be physical AI's ChatGPT moment.
