From Token Economics to Action Economics: How Physical AI Changes AI Infrastructure

Jensen Huang’s CES 2026 framing shifted the AI economics conversation: AI factories don’t produce tokens—they produce actions.

Traditional AI Economics (LLMs)

Physical AI Economics

The Infrastructure Cascade

1. Edge chips become critical path: Jetson Thor (1 PFLOP) enables on-device humanoid reasoning. 542,000 robots x continuous inference = massive aggregate edge demand.
2. Connectivity becomes non-negotiable: NVIDIA’s partnership with T-Mobile, Cisco for 6G. Robot fleets need industrial-grade networking.
3. Data center topology shifts: Physical AI needs training + simulation + edge triangulation, not just centralized inference.

The Flywheel

Physical AI isn’t just a market for AI compute—it’s becoming the primary generator of grounded, physics-based training data that LLMs and agents need to move beyond pure language reasoning.

This analysis is part of a comprehensive report. Read the full analysis: Physical AI Is Crossing the Manufacturing Chasm on The Business Engineer.