Fossil scleractinian corals were collected from the Galápagos platform in the East Equatorial Pacific (0°N, 90°E) on cruises MV1007 and NA064 from water depths between 419 and 650 m. Equatorial Atlantic corals (taxa Caryophyllia, Enallopsammia, Desmophyllum) were collected from a depth range of 749 to 2814 m during Cruise JC094 from Carter Seamount (9.2°N, 21.3°W), Knipovich Seamount (5.6°N, 26.9°W), Vema Fracture Zone (10.7°N, 44.6°W), Vayda Seamount (14.9°N, 48.2°W) and Gramberg Seamount (15.4°N, 51.1°W). Southern Ocean samples were obtained from Burdwood Bank (54.7°S, 62.2°W; taxa Caryophyllia, Balanophyllia, Flabellum, Desmophyllum) and Cape Horn (57.2°S, 67.1°W; taxa Caryophyllia, Balanophyllia, Flabellum) in the Subantarctic Zone and the Sars and Interim Seamounts in the Polar Front Zone (59.7°S, 68.8°W and 60.6°S, 66.0°W; taxa Caryophyllia, Desmophyllum) on cruises NBP0805 and NBP1103 in the Drake Passage. These proximal Sars and Interim sites are grouped as simply "Sars". The shallowest coral samples come from depths of 334 m on Burdwood Bank however the majority are from 700 to 1520 m, at water depths corresponding to modern Antarctic Intermediate Water. Corals recovered from the depth of 1012 m from Cape Horn and further south from Sars Seamount at depths of 695 to 1200 m are currently bathed in Upper Circumpolar Deep Water. Deeper samples at the Sars Seamount site sit within Lower Circumpolar Deep Water (1300 to 1750 m). We use published U-series dates for all samples (Burke and Robinson, 2012; Chen et al., 2020; Chen et al., 2015; Li et al., 2020; Margolin et al., 2014; Stewart et al., 2021). Reported age uncertainties are typically ±1% (2 SD). Whole "S1" septa and attached theca were taken from cup corals while whole calyxes were taken from branching specimens using a rotary cutting tool. This tool was further used to remove surficial oxide coatings and any chalky altered carbonate. Where sufficient sample material allowed, multiple sub-samples were measured to minimize microstructural bias (typically duplicates). Coral fragments were crushed and cleaned using warm 1% H2O2 (buffered in NH4OH) oxidative cleaning and a weak acid polish (0.0005 M HNO3). Samples were dissolved in 0.5 M HNO3 and analysed by ICP-MS to yield Li/Mg ratios. Repeat analysis of NIST RM 8301 (Coral) (n=19) yielded analytical precision of <± 1.5%. Coral Li/Mg was converted to temperature using a calibration applicable to all aragonitic corals (Li/Mg = 5.42 exp(−0.050×T(°C)); (Stewart et al., 2020). The quoted uncertainty on this calibration based on prediction intervals is ± 1.7 °C (1σ). This uncertainty is significantly reduced however at extremely low temperatures close to the freezing point of seawater (~ −2 °C). Corals could not survive in frozen seawater, therefore, where proxy estimated temperature falls below this minimum a value of −2 °C is reported instead. For Li/Mg averages of each coral sample and conversion to bottom water temperature, see the xlsx version of the dataset under Further details.