NVIDIA declares a "ChatGPT moment" for self-driving cars and robots at GTC 2026

At GTC 2026, NVIDIA moved the physical AI conversation from demos to a deployment plan. The company said a "ChatGPT moment" has arrived for self-driving cars, and showed how it intends to scale the same stack to robotaxis, industrial robots and digital twins.

A turning point for the auto industry

NVIDIA's central claim is as ambitious as it gets: autonomous driving is ceasing to be an endless R&D project and becoming a commercial platform. At GTC 2026, the company said BYD, Geely, Isuzu and Nissan are building L4-ready programs on NVIDIA DRIVE Hyperion, while Uber is expanding its partnership with NVIDIA to launch a fully AI-driven robotaxi fleet in 28 cities on four continents by 2028. The launch is planned for Los Angeles and the San Francisco Bay Area in the first half of 2027.

The emphasis falls not only on compute but also on safety. NVIDIA is promoting Halos OS as a unified safety architecture for AI-driven cars and ties it to the open model Alpamayo 1.5, which is meant to help vehicles work through rare and complex road scenarios. It is an attempt to move away from the old approach, in which each team solves sensing, planning and validation separately, and to assemble a unified stack for series production instead.

"The autonomous transport revolution has already begun — this is the first multi-trillion-dollar robotics industry," said NVIDIA CEO

Jensen Huang.

Tools for robots

Beyond the auto industry, NVIDIA also expanded the robotics side of its physical AI stack. The company showed Cosmos 3 for generating and modeling worlds, Isaac Lab 3.0 for large-scale robot training in simulation, and GR00T N1.7 as an open model for general-purpose robot skills. On stage, it also announced a future model, GR00T N2, which NVIDIA says handles new tasks in new environments more than twice as successfully as leading vision-language-action systems.

The point of these releases is that NVIDIA is no longer selling just a chip, or even just an SDK, but the full path from training to deployment. Manufacturers such as ABB Robotics, FANUC, KUKA, YASKAWA, Figure, Agility and Boston Dynamics use Omniverse, Isaac and Jetson to first run robots through physically accurate digital twins and then move the models onto real hardware. For the industry, this lowers the cost of mistakes: expensive experiments move from the factory floor into simulation.

A bet on data

A separate group of announcements covers how to train physical AI faster and more cheaply. NVIDIA introduced the Physical AI Data Factory Blueprint, an open reference architecture that combines data collection, synthetic generation, augmentation and model evaluation in a single pipeline. The logic is simple: the real world offers too little data and is too chaotic, and the most dangerous cases for self-driving cars and robots are exactly the ones that occur rarely. So data has to be not only collected but also produced at scale.

This layer includes several components:

• Alpamayo 1.5 for reasoning-based autonomous driving and handling long-tail scenarios
• Omniverse NuRec for reconstructing and refining scenes when training AV systems
• Cosmos 3 for generating synthetic worlds and simulating actions
• Isaac Lab 3.0 and Jetson Thor for moving from training to real-world execution
• Cloud deployments through Microsoft Azure and Nebius to scale the data factory

This strategy explains why NVIDIA increasingly talks about infrastructure rather than GPUs alone. If the approach wins, in which compute becomes data and data becomes ready-made policies for machines, the company can capture value across several layers at once: models, simulation, orchestration, edge computers and partner clouds.

What it means

NVIDIA is trying to establish itself as the foundational supplier of physical AI, just as it did in generative AI. If the company delivers on its plan for Uber, its automotive partners and its robotics stack, the market will get not another demo showcase but working infrastructure for putting autonomous systems into the mass market on roads, in warehouses, in factories and in service robotics.