Geopolitics & Market Structure
China vs US in Physical AI: Who Is Winning the Robot Race?
China produces roughly 80% of all humanoid units. The US produces roughly 100% of the published industrial deployment KPIs. Both statements are currently true.
Last updated: July 2026
The core split
- Volume — ~80% of global humanoid units
- Price floor — Unitree G1 at $16,000
- Manufacturing integration — robots in Chinese auto plants
- Government support — 900M+ yuan sector investment, early 2025
- Speed to 10,000 units — AgiBot shipped its 10,000th in March 2026
- AI software stack — GR00T, Helix, Gemini Robotics
- Enterprise deployments with real SLAs — GXO, BMW
- Regulatory milestones — Agility Digit: first OSHA-certified humanoid (Nov 2025)
- Capital raised — Figure AI $39B valuation; Apptronik $935M
- Chip access — NVIDIA Jetson Thor designed and produced in US ecosystem
China: the volume play
China's advantage is structural. The government has set explicit robot adoption targets across manufacturing sectors. Domestic OEMs — automotive, electronics, logistics — are required or incentivized to trial humanoids. This creates a guaranteed early market that Western companies do not have.
US: the stack and the capital
The US lead is in the AI that makes robots useful, not the robots themselves. Foundation models for robot policy — NVIDIA GR00T, Google Gemini Robotics, Physical Intelligence's π model — come from US labs. The enterprise deployments with published performance data also come from US companies.
EU: regulation as differentiator
The EU has no significant domestic humanoid producers. Boston Dynamics operates globally but is US-owned (Hyundai-controlled). Fourier GR-2 runs at ETH Zurich as a research platform, not a commercial deployment.
The EU's potential moat is regulatory. The AI Act and Machinery Regulation (enforcement from January 2027) impose compliance requirements that may favor robots from companies with full documentation, Notified Body assessments and EU database registration. A robot certified in the EU could command a premium in regulated industries globally.
Whether EU regulation becomes a competitive moat or a barrier to adoption depends on enforcement pace. See the EU AI Act analysis for detail.
The key tension: shared chip dependency
S. 3275 — Humanoid ROBOT Act of 2025: bars US federal agencies from acquiring AI humanoid robots made by companies in China, Iran, North Korea, or Russia. Applies to federal procurement only.
The political tension is real. But the technology dependency runs both ways. NVIDIA Jetson Thor — designed in the US, critical for robot real-time inference — is widely adopted by Unitree, AgiBot, and other Chinese humanoid companies. NVIDIA sells commercially and does not restrict by buyer nationality within current export control law.
This means US chip technology is currently enabling the Chinese volume ramp that the Humanoid ROBOT Act seeks to contain at the federal level. A future chip export restriction targeting humanoid robotics would be the highest-impact single policy lever — and would also constrain the cost curve that BofA's $17K projection depends on.