Jim Fan at GTC: DreamZero 'World Action Model' Enables Zero-Shot Robot Prompting Across Morphologies

Sequoia Capital·medium signal

NVIDIA's Jim Fan presented DreamZero at GTC 2026, a robot foundation model using video as a universal bridge between hardware morphologies. The 'World Action Model' backbone dreams the correct future in video pixels so robots execute in motors, enabling zero-shot prompting for new verbs, nouns, and environments without retraining. Fan declared world models — not language backbones — are the next-generation foundation of Physical AI.