The Physical AI stack is fragmenting—and consolidating—simultaneously. Understanding where value gets captured determines strategic positioning.
The Value Capture Spectrum
| Layer | Open/Commodity | Proprietary Lock-In |
|---|---|---|
| Foundation Models | OpenVLA, SmolVLA (open weights) | GR00T (NVIDIA), Helix (Figure), Pi0 (Physical Intelligence) |
| Simulation | Open-source robotics sim exists | Isaac Sim (NVIDIA monopoly) |
| Compute Hardware | AMD, accelerators emerging | NVIDIA (90%+ GPU market) |
| Edge Inference | ARM architecture optionality | Jetson Thor (no real alternative) |
| Data | Open X-Embodiment dataset | Operational data from deployed fleets |
Where Value Accrues
Hardware: Rapidly commoditizing
Differentiation shifts to:
- Simulation-to-real pipelines
- Foundation model alignment
- Integration speed
- Data ownership from deployed fleets
The Platform Dynamic
NVIDIA’s “Android of robotics” strategy mirrors mobile:
Open models on proprietary compute → Value capture regardless of OEM winner
Strategic Question for Enterprises
Do you build on open infrastructure (flexibility, avoid lock-in) or proprietary stacks (integration, whole-product benefits)?
The chasm-crossing moment means this decision must be made now.
This analysis is part of a comprehensive report. Read the full analysis: Physical AI Is Crossing the Manufacturing Chasm on The Business Engineer.
More Resources
