Reconstructing past ocean [CO32−] allows the paleodepth of the chemical lysocline to be constrained, an important control on past atmospheric CO2. However, the causal mechanisms responsible for observed spatial and temporal variations in [CO32−] are difficult to quantify because of the complicated carbonate chemistry system. Here spatial and temporal variations in [CO32−] are quantitatively and concisely related to variations in ocean carbon storage due to different processes. The spatial variation in [CO32−] is given by Δ[CO32−] = γ(ΔCsoft + ΔCdis + (∂Csat/∂T)ΔT − ΔCcarb), where Csoft and Ccarb are the dissolved inorganic carbon (DIC) from remineralization of marine soft tissue and CaCO3, respectively, T is seawater temperature, (∂Csat/∂T) is the temperature-solubility sensitivity of DIC, Cdis is the DIC from air-sea disequilibrium, and γ is a carbonate chemistry coefficient. A similar quantitative function for temporal variation in global mean ocean [CO32−] is derived in terms of atmospheric CO2, CaCO3 precipitation and dissolution, and carbon exchanges of terrestrial or fossil fuel origin. Comparing published [CO32−] reconstructions at the Last Glacial Maximum (LGM) and the late Holocene, the quantitative relationships reveal how the spatial distribution of ocean carbon storage was altered. Relative to the Intermediate North Atlantic, the rest of the ocean saw Csoft + Cdis + (∂Csat/∂T)T − Ccarb increase by an extra 570–970 Pg C during the LGM. Assuming that the Intermediate North Atlantic Csoft + Cdis + (∂Csat/∂T)T − Ccarb did not decrease during the LGM, this 570–970 Pg C increase in the rest of the ocean is enough to explain 40%–70% of the observed glacial decrease in atmospheric CO2.