Physical AI Takes the Wheel: How Global Robotaxi Leaders Are Building with NVIDIA Technology

The global robotaxi market is projected to grow to $400 billion by 2035, with more than 6 million commercial vehicles expected to be in operation. NVIDIA provides a 3-computer architecture to support these large-scale fleet operations, covering AI training, simulation, and real-time in-vehicle processing.

NVIDIA's 3-Computer Architecture

NVIDIA's solution consists of three core computing domains that cover the entire development lifecycle.

• Training Computer (NVIDIA DGX): Through the NVIDIA Alpamayo portfolio, it provides reasoning-based Vision-Language-Action (VLA) models and physical AI datasets. By applying reasoning models that decompose complex driving scenarios step-by-step, the minimum average displacement error was reduced by 43%.
• Simulation and Validation Computer (NVIDIA Omniverse/Cosmos): It reconstructs real-world scenarios from sensor data and generates physics-based variations using the NVIDIA Cosmos world foundation models. This allows real corner cases to be expanded into millions of combinations for validation.
• In-Vehicle Computer (NVIDIA DRIVE Hyperion): The DRIVE Hyperion 10, equipped with two DRIVE AGX Thor SoCs, fuses data from 14 HD cameras, 3 LiDARs, and other sensors in real time. It features a fail-operational design that ensures driving capability even in the event of component failure.

Expanding the Global Ecosystem

Major companies such as Uber, Mercedes-Benz, Hyundai Motor, and Kia are adopting NVIDIA's architecture. Uber plans to operate a fleet based on NVIDIA DRIVE Hyperion in 28 cities by 2028. Hyundai Motor and Kia are expanding their collaboration on developing data-driven autonomous driving systems based on NVIDIA DRIVE Hyperion, while also considering enhancements to Level 4 robotaxi services through their joint venture, Motional.