Abstract: Gender inequality in academic careers, documented across all disciplines and countries, extends beyond the fraction of women involved in research: compared to their male colleagues, women publish less over the course of their careers and their work acquires fewer citations. Yet, all existing evidence is limited to selected countries or disciplines, restricting our ability assess the roots and implications of the gender disparity. Here we analyzed a large corpus of scientific publications since 1900, identifying the gender and reconstructing the full publishing career of over 1.5 million authors from most scientific disciplines and countries, allowing us to quantify the processes and outcomes for women and men in science. We confirm that men have higher total productivity and total impact than women, a pattern impacting all disciplines and most countries. Surprisingly, we find no systematic difference between the annual productivity of male and female scientists, not only offering a gender-invariant productivity measure, but also suggesting that the observed gender gaps are rooted in gender dependent dropout rates. We find that not only do women leave academia at a higher rate than men, but surprisingly, this gap in dropout rate is greater for the more productive women. We show that when we control for these two gender-specific dropout rates, the career gender gaps in both productivity and impact vanish. Identifying the driving forces of gender gaps can help rephrase the conversation about gender inequality around the sustainability of women’s careers in academia, with important consequences for policy makers and academic institutions.