<p>Photothermal (photonic) sintering crystallizes as-deposited amorphous LiCoO2 (LCO) cathodes for solid-state thin-film batteries using millisecond, surface-localized heating. However, process design often relies on 1D models with phase-averaged, temperature-independent properties, which can mispredict peak temperatures and thermal damage margins. Here we develop a multiscale, data-driven framework that provides phase- and grain-size–resolved thermophysical inputs for stoichiometric LCO. We train an Allegro neural network potential with near-ab initio accuracy, enabling Green–Kubo calculations of thermal conductivity for crystalline and amorphous phases. The low, weakly density-dependent conductivity of amorphous LCO motivates its use as an effective intergranular phase in a thin-interface model that reproduces observed grain-size–dependent thermal transport. Combined with measured wavelength-resolved optical properties in 1D multiphysics simulations, we show amorphous LCO absorbs more strongly and reaches higher peak temperatures than crystalline LCO; thus crystalline, constant-property models systematically overestimate safe operating windows.</p>