Nucleation is the seminal event in crystallization and predicting its progression is a key challenge across diverse scientific fields. Although classical nucleation theory successfully describes some systems, it fails to do so in a growing list of materials due to free energetic and kinetic complexities1. In particular, when systems are driven far from equilibrium, the high chemical potential opens pathways that pass through metastable polymorphs, amorphous states, and even dense liquid phases that can form absent the energy barrier normally opposing nucleation1-5. CaCO3 provides a canonical example6-8; following discovery of the CaCO3 “polymer-induced liquid precursor”9,10, which forms upon introduction of poly-ionic polymers10-12 like polyacrylic acid13,14, numerous studies have concluded that even pure CaCO3 possesses a dense liquid phase. This phase — likely stabilized by poly-ionic proteins15-19 — is proposed to serve as a precursor to biomineral formation15, transforming first to the amorphous phase and then to crystalline products3,20-22. Yet little is known about the mechanisms of CaCO3 dense liquid formation and solidification, its composition is unknown, and its relationship to the polymer-induced liquid-precursor remains unclear. Using in situ methods, we show that this phase forms by spinodal decomposition, is a bicarbonate comprised of 1Ca2+:2HCO3-:7±2H2O, and transforms to hollow particles of hydrated amorphous CaCO3 with the release of CO2 and H2O. Acidic proteins and polymers extend liquid phase lifetimes but leave this chemical pathway unchanged. Molecular simulations suggest the liquid phase forms via direct condensation of solvated Ca2+•(HCO3-)2 complexes that react due to proximity effects in the confinement of the liquid droplets. The findings provide an in-depth picture of CaCO3 nucleation via liquid-liquid phase separation, elucidating an often-proposed pathway by which nature orchestrates the complex process of biomineralization.