Bit Energy, Model Weight, and the FRA Cycle: Why AI Information Has Mass

This note links Landauer’s limit with the FRA (Ξ–Φ–ℱ) cycle and treats an AI model as a physical object with non-zero mass–energy.

We introduce three predicates for any model: existence (E_AI), potential (P_AI), and active inference (A_AI), and show how each step of computation corresponds to transitions Φ → ℱ → δ with irreversible heat dissipation (ΔE ≥ kT ln2 per bit operation).

Stored parameters carry “model weight” even when the system is idle; power only creates readiness, and queries trigger real energy loss.

Erasing the model is described as structural relaxation ℱ → Φ → Ξ.

The FRA framework is presented as an engineering abstraction, not a strict theorem, clarifying how information persists, transforms, and vanishes in physical hardware.