The Pulse
NVIDIA Builds Robotaxi Push Around One Full-Stack Platform
NVIDIA says robotaxi operators, automakers and autonomous-driving developers across Asia, Europe, the Middle East and North America are adopting its DRIVE platform. The company’s September 10, 2026 update highlights Uber’s planned 28-city e
AI.info Team ·
“Everything that moves will eventually become autonomous, and DRIVE Hyperion is the backbone that makes that transition possible.”
Ali Kani, vice president of automotive, NVIDIA
NVIDIA is positioning its DRIVE platform as the common computing foundation for a widening group of robotaxi companies, automakers and mobility operators. In a company post published September 10, NVIDIA says its three-part system covers model training, simulation and validation, and in-vehicle computing across commercial autonomous-driving programs.
The announcement names partners and customers in North America, Europe, Asia and the Middle East. It also gives NVIDIA a way to present scattered collaborations as one connected system, running from data-center training infrastructure to sensors, safety software and the vehicle computer.
Uber Gives NVIDIA a Route to 28 Cities
Uber is the most visible operator in NVIDIA’s account. NVIDIA says Uber plans to scale a fleet built around DRIVE Hyperion vehicles to 28 cities by 2028, while the two companies develop a robotaxi data factory using NVIDIA Cosmos to curate driving data from rare and difficult road situations.
The companies are working with Autobrains, Avride, Lucid, May Mobility, Mercedes-Benz, Momenta, Nissan, Nuro, Pony.ai, Stellantis, Waabi, Wayve, WeRide and Zoox on services intended for Uber’s ride-hailing network. NVIDIA’s post does not say that every named company will deploy the same vehicle or software configuration; the relationships span vehicle development, autonomous-driving software, fleet operations and computing.
May Mobility plans to operate autonomous ride-hailing services through Uber while developing its software on DRIVE. Lyft plans to use DRIVE Hyperion as a reference architecture for future autonomous fleets, and May Mobility vehicles are already operating on Lyft’s Atlanta network with NVIDIA DRIVE technology, according to NVIDIA.
DRIVE Hyperion 10 Puts Two Thor Computers in the Vehicle
NVIDIA’s hardware proposition centers on DRIVE Hyperion 10, a modular reference architecture for level 4-ready vehicles. The system pairs two DRIVE AGX Thor systems-on-chip, based on NVIDIA’s Blackwell architecture, with 14 high-definition cameras, nine radars, three lidars and 12 ultrasonic sensors.
NVIDIA says the dual Thor configuration is designed to support perception, reasoning, path planning and driving actions, including workloads from vision-language-action models. The company also describes a redundant compute and sensing design intended to keep the vehicle operating if a sensor or computing component fails.
Earlier NVIDIA material puts the combined system above 2,000 FP4 teraflops, or roughly 1,000 INT8 trillion operations per second, for real-time processing. DRIVE Hyperion is paired with DriveOS and NVIDIA Halos, a safety and certification framework that covers inspection, validation, simulation and testing from the data center to the vehicle.
Alpamayo Extends the Stack Into Driving Decisions
The software layer includes NVIDIA DRIVE AV, simulation tools and the Alpamayo portfolio of open reasoning vision-language-action models. NVIDIA says developers can adapt the models, datasets and training recipes to their own vehicles and software stacks rather than building every component from the ground up.
The company cites one internal evaluation in which adding meta-action and chain-of-thought reasoning data reduced a model’s minimum average displacement error by 43%, from 2.08 to 1.18. That figure describes trajectory prediction in a development evaluation; it is not a measure of deployed robotaxi safety or overall driving performance.
Simulation is supplied through NVIDIA Omniverse, Cosmos and the AlpaSim framework. NVIDIA says those tools support closed-loop testing, synthetic driving scenarios and validation before software reaches a public road. Halos adds independent inspection and certification processes intended to test autonomous systems across cloud infrastructure and vehicle hardware.
Automakers Move From Partnerships to Vehicle Programs
NVIDIA’s list includes several programs that connect the platform to specific production or development vehicles. Mercedes-Benz and NVIDIA are working with Uber on a robotaxi ecosystem based on a new S-Class, using DRIVE Hyperion, full-stack DRIVE AV L4 software and Alpamayo models, simulation tools and datasets.
Stellantis, Wayve and Uber are collaborating on level 4 driverless mobility services. Lucid, Nuro and Uber are developing a global robotaxi service using DRIVE AGX Thor, while Hyundai Motor and Kia are expanding their autonomous-driving work with NVIDIA and exploring an expanded relationship with Motional.
Other programs target different parts of the market. Wayve, Nissan and Uber are developing a prototype robotaxi that combines Nissan vehicle engineering, Wayve’s embodied AI and DRIVE Hyperion. Pony.ai has developed a new autonomous-driving domain controller with Hyperion and Thor, and TIER IV and Isuzu are deploying level 4 autonomous buses on the same hardware platform.
A Platform Strategy With a Clear Trade-Off
NVIDIA’s pitch is straightforward: automakers and autonomous-driving companies can concentrate on their driving software, fleet operations and vehicle designs while adopting a prequalified computing and sensor foundation. That approach may shorten integration work, but it also gives NVIDIA a larger role in the technical decisions that shape future fleets.
The September 10 post is a company account of partner adoption, not an independent survey of deployed robotaxi systems. Several programs remain in development, and the announcement does not establish how many vehicles are operating, how many passengers they carry or whether each collaboration will reach commercial service.
What NVIDIA can document is the breadth of the platform it is offering: DGX systems for training, Omniverse and Cosmos for simulation and data, Alpamayo for driving models, DRIVE Hyperion and Thor for vehicle computing, and Halos for safety work. Uber’s 28-city target by 2028 is the clearest commercial milestone attached to that stack.