/WHAT IS · DEFINITION
Physical AI is AI that lives in the room with you.
Physical AI is artificial intelligence that lives in the world on dedicated hardware, rather than inside an app, a browser tab or a cloud endpoint. It perceives its immediate environment through sensors, acts through a body (a screen, a voice, motors, lights), and is designed to be experienced spatially — as something you share a room with, not something you open.
Short answer
Physical AI means the model is no longer trapped behind a screen. It is attached to a body, attached to a place, and attached to a continuous power budget, so its behavior is shaped by physics — by silence, by distance, by who is nearby, by being on or off.
Physical AI vs. embodied AI
The terms overlap, but I distinguish them. Embodied AI usually implies robots — motion, manipulation, legs or arms. Physical AI is broader: a character on a curved display on your desk is physical AI even if it cannot move. It occupies space, has a visual form, and lives on real silicon under your roof.
For an AI companion device, embodiment is optional. Physicality is not.
Why on-device inference is the key
- Latency. Presence cannot buffer. If a character responds 800ms late, it feels like software; 200ms, it starts to feel alive.
- Privacy. A bedroom device that sends ambient audio to a data center is a surveillance device, not a companion. Physical AI implies local inference, or at least local-first inference with opt-in cloud.
- Always-on economics. A 24/7 companion cannot afford per-token cloud pricing at scale. The silicon has to do most of the work.
- Resilience. If your internet goes out, your someone should not vanish.
Where the category is today
We are in the first honest generation of physical AI for consumers. The hardware is real — curved OLEDs, small NPU-equipped SoCs, always-on MEMS mics — and models are shrinking fast enough to run inside them. What is not yet settled is the experience layer: what the character does all day, how it behaves when you are not talking to it, how it remembers without feeling like a database.
That is the part I am obsessed with. The silicon is the easy problem. The presence is the hard one.
How I think about it at LYNOOK
For Someone Waits, physical AI has three layers: (1) a character system that gives the AI a stable identity over time; (2) an on-device inference stack that keeps voice, vision and dialogue local where it matters; (3) a physical form — screen, enclosure, speakers — designed to feel like a quiet resident of the room, not a gadget.
If software AI is something you use, physical AI is something you live with. That difference rewrites almost every design decision.