Will manipulation and unstructured environments keep embodied AI narrow, or will general physical competence eventually arrive?
Skeptics argue touch-rich manipulation lacks both the data and the forgiving physics that made language tractable; market forecasts and engineering assessments currently point in different directions. (Labor-market consequences are treated under Economy 3.2.)
View on the map → · Open in Browse →
The supply side got measured: Epoch finds precision reducers cap humanoid production near 500,000 a year on current capacity, with plausible paths to millions by 2030; Unitree's IPO filing shows a profitable Chinese hardware business shipping 37x more humanoids than its US rivals — while only 9% of Chinese humanoid sales go to real industrial use, and Brooks argues deployment at economy-reshaping scale takes decades regardless.
While software can be copied and deployed nearly instantly, each robot must be manufactured from real components in real factories by real workers.
First rigorous supply-side analysis of the physical-AI buildout: ~16,000 humanoids made in 2025 with 6-month doubling times, precision reducers as the binding bottleneck, and plausible 2030 paths to 1.5–10 million humanoids per year.
We don't see any significant competitors outside of China
Deep reporting on China's robotics push: Unitree shipped 5,500+ humanoids in 2025 while slashing prices, over 140 Chinese firms build humanoids, and cities have poured $26B+ into development funds — but only 9% of sales go to actual industrial use.
A typical high-tech company heading toward IPO is money-losing, carries a high R&D cost ratio, and runs on investor patience. Unitree doesn't fit that profile.
IPO-filing analysis: Unitree is profitable with 60% gross margins, 335% revenue growth, and a 37:1 humanoid shipment advantage over Figure or Agility — China's hardware cost curve, not Western software, currently drives embodied-AI economics.
It really does take decades, essentially a human lifetime, to deploy a technology at large enough scale that it reshapes the world's economy.
Framework essay arguing hype cycles, deployment, and economic reshaping run on different clocks: even successful embodied AI implies decades before physical AI is economically transformative.
You're literally sitting there watching a robot do a job, and your only role is to unstick that thing if it stops working.
Ground-level measurement of warehouse automation, the structured-environment case that is actually working: ~1 million Amazon robots versus 1.5 million workers, and average facility headcount at a 16-year low.