Input datasets for annualized earthquake loss (AEL) estimation for California were updated recently by the scientific community, and include the National Seismic Hazard Model (NSHM), site‐response model, and estimates of shear‐wave velocity. Additionally, the Federal Emergency Management Agency’s loss estimation tool, Hazus, was updated to include the most recent census and economic exposure data. These enhancements necessitated a revisit to our previous AEL estimates and a study of the sensitivity of AEL estimates subjected to alternate inputs for site amplification. The NSHM ground motions for a uniform site condition are modified to account for the effect of local near‐surface geology. The site conditions are approximated in three ways: (1) by VS30 (time‐averaged shear‐wave velocity in the upper 30 m) value obtained from a geology‐ and topography‐based map consisting of 15 VS30 groups, (2) by site classes categorized according to National Earthquake Hazards Reduction Program (NEHRP) site classification, and (3) by a uniform NEHRP site class D. In case 1, ground motions are amplified using the Seyhan and Stewart (2014) semiempirical nonlinear amplification model. In cases 2 and 3, ground motions are amplified using the 2014 version of the NEHRP site amplification factors, which are also based on the Seyhan and Stewart model but are approximated to facilitate their use for building code applications. Estimated AELs are presented at multiple resolutions, starting with the state level assessment and followed by detailed assessments for counties, metropolitan statistical areas (MSAs), and cities. AEL estimate at the state level is ∼$3.7  billion, 70% of which is contributed from Los Angeles–Long Beach–Santa Ana, San Francisco–Oakland–Fremont, and Riverside–San Bernardino–Ontario MSAs. The statewide AEL estimate is insensitive to alternate assumptions of site amplification. However, we note significant differences in AEL estimates among the three sensitivity cases for smaller geographic units.