The State of Physical AI
The software era was defined by a single elegant principle: marginal cost approaches zero. Copy software once, distribute infinitely. Build a platform, add users for nearly nothing.
The economics of bits — frictionless, weightless, borderless — produced the largest wealth creation in history and the defining ideology of two generations of technologists.
AI breaks this principle at the frontier. Not everywhere, not forever, but right now, at the capability level that matters — the intelligence is not weightless. The machine that produces it is not frictionless. The infrastructure underneath it does not scale like software. It scales like steel.
This is the inversion that has not been fully absorbed. Everyone can see that AI is expensive.
Fewer have understood why the expense is structural, not transitional — that the physics of what we are building imposes constraints that capital alone cannot override, that the digital layer is increasingly a function of the physical layer beneath it, and that the physical layer is more concentrated, more fragile, and more geopolitically exposed than anything computing has depended on before.
The five truths below are not predictions. They are descriptions of what is already true, stated in a way that makes the strategic implications visible.
