We study the determinants of lifetime earnings (LE) inequality in the U.S. using administrative balanced panel data between the ages of 25 and 55 by focusing on the roles of job ladder dynamics and on-the-job learning. We first document that lower LE workers change jobs more often, mainly driven by higher nonemployment risk. Second, average annual earnings growth for job stayers is surprisingly similar at around 2% in the bottom two thirds of the LE distribution, whereas for job switchers it rises almost linearly from zero at the bottom to around 4% at the 90th percentile. Third, top LE workers enjoy high earnings growth regardless of job switching. To interpret these facts, we estimate a life-cycle job ladder model with on-the-job learning featuring ex-ante heterogeneity in learning ability as well as in job ladder risk—job loss, job finding, and contact rates. We find that learning ability is Pareto distributed and explains almost all earnings growth heterogeneity above the median LE—the main driver of LE inequality. As for below the median LE, we find that 80% of lifetime earnings growth differences would vanish if workers had the same ex-ante job ladder risk. We validate this large ex-ante heterogeneity in job-ladder risk with direct evidence from survey data.