I held a 2,000 teraflop robotics computer, talked to the engineers programming robots and self driving cars, and then rode through San Francisco in a Mercedes with nobody touching the wheel. Two people in this video told me about their moms. That's not something I expected from a tech conference. Watch and you'll see what I mean.

NVIDIA gave me access to Jetson Thor, their robotics computer, the GTC floor, and a self driving Mercedes running their DRIVE AV stack at L2++ through downtown San Francisco. I sat down with Spencer Huang, who leads NVIDIA's robotics software (Isaac, GR00T, Cosmos), and he walked me through what actually happens inside a robot brain from the moment it sees something to the moment it moves. Then I met Jan Liphardt, a Stanford professor building OM1, an open source operating system for robots with its own app store. Then NVIDIA put me in the car.

One idea runs through this entire video: the same compute platform powering humanoid robots is also powering autonomous cars.

TIMESTAMPS
00:00 I Rode in a Self Driving Mercedes Through San Francisco
00:35 Inside NVIDIA GTC 2026 Robotics Floor
00:49 NVIDIA Jetson Thor Explained: The Brain Inside Every Robot
01:35 How World Models Let Robots Predict the Future
02:06 Sponsor: Incogni Data Privacy
03:06 NVIDIA's Three Computer Pipeline: Train, Simulate, Deploy
03:20 The Sim to Real Gap: Why Robots Fail in the Real World
05:43 OpenMind OM1: The Open Source Operating System for Robots
09:01 NVIDIA DRIVE AV Demo: Autonomous Mercedes in San Francisco
12:05 One Chip, Many Bodies: Why This Changes Everything

Tiff In Tech

A firsthand look at NVIDIA's robotics ecosystem from GTC 2026, covering the Jetson Thor compute module, the sim-to-real gap in robot training, and the emerging data flywheel concept borrowed from autonomous vehicles. Includes interviews with NVIDIA's robotics software lead Spencer Huang on world models and CI/CD for robots, and Open Mind founder Jan Liphardt on OM1, an open-source robot OS with an app store. Also covers a ride in a self-driving Mercedes running NVIDIA's drive stack at L2++, with plans for Level 4 robotaxi deployment via an Uber partnership. Key insight: the tools (Isaac, Groot, Cosmos, OM1) are now open and accessible enough that domain experts without robotics PhDs can start building.

The Robot Problem Nobody Is Talking About