This article describes and explains a previously overlooked empirical pattern in state revenue collection. As late as 1913, central governments in the West collected similar levels of per capita revenue as the rest of the world, despite ruling richer societies and experiencing a long history of fiscal innovation. Western revenue levels permanently diverged only in the following half-century. We identify the twentieth-century great revenue divergence by constructing a new panel data set of central government revenue with broad spatial and temporal coverage. To explain the pattern, we argue that sustainably high levels of revenue extraction require societal demand for an activist state, and a supply of effective bureaucratic institutions. Neither factor in isolation is sufficient. We formalize this insight in a game-theoretic model. The government can choose among low-effort, legibility-intensive, and crony-favoring strategies for raising revenues. Empirically, our theory accounts for low revenue intake in periods of low demand (the nineteenth-century West) or low bureaucratic capacity (twentieth-century former colonies), and for eventual revenue spikes in the West.