Layer map · public-safe

The robot foundation-model layer

Robotics is not only a hardware cost curve. The model/data layer is now worth tracking, but public evidence still lacks the economics needed to call it an S5 value-capture layer.

Draft updated 2026-06-13. This page is a working surface for review before the final publishing flow.

Evidence curves

Physical Intelligence curve

PI's π0, openpi, π0.5 and π0.7 show a public research/open-tooling curve for robot policies, cross-embodiment learning and broader generalization claims.

S3/S4

Skild AI curve

Skild shows an omni-bodied model thesis, US$1.4bn Series C at over US$14bn valuation, ABB/UR/NVIDIA partnerships and Zebra/Fetch robotics assets.

What is proven

The model layer is now credible as a research lens: tooling, capital, data strategy, industrial channels and cross-embodiment ambition are all visible in primary sources.

Signal

What is not proven

No reviewed public record yet proves recurring software revenue, license pricing, robot attach rate, gross margin, retention, deployed fleet economics or customer ROI/payback.

S5 gap

Think deeper

Is the scarce asset model architecture, real-world data, safety validation or deployment workflow integration?
Will model economics be sold as separate ARR, per-robot licensing, outcome pricing or embedded OEM margin?
If open tooling lowers experimentation cost, does it accelerate the category faster than it reduces pricing power?