NVIDIA pitches World Foundation Models as a way to scale physical AI data generation

MonaJalal_ · x · 2026-07-22

At NVIDIA Physical AI Day, the speaker described a shift from painfully slow data collection to scalable, generalizable data generation with World Foundation Models. The slides frame physical AI as the next frontier, tying robot capability to larger compute-and-data volumes across factories, robots, farms, cities, vehicles, and commercial cameras.

The talk emphasizes a progression from teleoperation to classical simulation, then internet-scale data, and finally world foundation models. The core claim is that compute can act as data for physical AI, enabling larger-scale robot training and deployment across many industries.

Original post →

More from Embodied

Embodied channel →