NVIDIA’s GR00T N1.6 training data has reached 4.8 million downloads on Hugging Face, with Cosmos world foundation models exceeding 3 million downloads. The “Android of robotics” strategy is taking shape.
NVIDIA Platform Metrics
| Asset | Downloads/Scale |
|---|---|
| GR00T N1.6 Training Data | 4.8M downloads |
| Cosmos World Models | 3M+ downloads |
| Physical AI Dataset | 500K robotics trajectories (open-sourced) |
The Platform Strategy
NVIDIA’s “Android of robotics” approach: open models on proprietary compute infrastructure.
Just as Android enabled smartphone diversity while Google captured search/services value, GR00T + Isaac + Jetson enables robot diversity while NVIDIA captures compute value.
Who’s Using Cosmos
- 1X — Humanoid development
- Agility Robotics — Warehouse automation
- Figure AI — Manufacturing robots
- Skild AI — General-purpose robotics
All are using Cosmos for synthetic data generation—generating “richer training data faster and at scale” for Physical AI post-training.
The Stack Lock-In
NVIDIA’s position:
- Training: Faces competition (TPUs)
- Simulation: Monopoly (Isaac Sim)
- Edge: Unassailable (Jetson Thor has no alternative)
This analysis is part of a comprehensive report. Read the full analysis: Physical AI Is Crossing the Manufacturing Chasm on The Business Engineer.
