Chlorophyll is a key indicator of marine phytoplankton biomass and provides essential information on ocean productivity, ecosystem variability, and biogeochemical cycling. However, satellite observations are largely restricted to the near-surface ocean, while in situ measurements remain sparse in space and time. Consequently, the three-dimensional distribution and temporal evolution of chlorophyll in the ocean interior are still poorly constrained. Here, we present an artificial intelligence framework for reconstructing global three-dimensional chlorophyll fields from multiple physical, biogeochemical, and atmospheric variables. By learning the nonlinear relationships connecting ocean circulation, hydrography, nutrient availability, surface forcing, and phytoplankton variability, the framework integrates information across depth, geographical regions, and time. It produces spatially and temporally coherent estimates of subsurface chlorophyll without relying directly on surface chlorophyll observations.