Up-to-date geospatial assessment of soil organic carbon (SOC) at national level is critical for developing land management strategies to abate global climate change.  The current national assessment of soil carbon (C) across the U.S. has been based on legacy data (Soil Survey Geographic Data – SSURGO/gSSURGO and State Soil Geographic Data – STATSGO2), which have become outdated due to imposed land use and climate change. In this study, we present new digital SOC maps of 0-100cm and 0-20cm depth at 100 m (3 arc-second) resolution for the conterminous U.S. based on the data from the Rapid Assessment of U.S. Soil Carbon (RaCA) Project (soil sampling time period: 2010-2012). A total of 32,084 soil profile data of the upper 1 m were fused with 700+ environmental covariates (space-time layers) covering the U.S. from various sources (remote sensing,  National Elevation Dataset, climate data from PRISM project, etc.). A data reduction algorithm (Boruta) was applied to identify the controlling factors and forcings of SOC and the optimized covariates were then used for geospatial modeling. We developed bootstrapped regression-kriging models (bootstraps: 100) to predict SOC at 0-100cm and 0-20cm depth at unsampled locations and the local uncertainty of predictions is expressed as the p-probability intervals bounded by the (1-p)/2 and (1+p)/2 quartiles of the empirical distribution of predictions. We randomly selected 70% of the data for model calibration and the other 30% for model validation. We also assessed the model calibration using 10-fold cross validation. The new SOC maps not only provide an up-to-date inventory of SOC in the conterminous U.S., but also contribute to the GlobalSoilMap project.