This paper proposes a novel assessment of the Kuznets curve for an underdeveloped country engaging in rapid late development. We mobilize new long-run data for Brazil, combining surveys, administrative records, and national accounts statistics, to compute macro-consistent income shares and other distributional indicators since the 1920s. Our estimates show a more nuanced picture for the traditional Kuznets hypothesis than what the existing literature has suggested. The major complication for the standard narrative lies in the period roughly between 1950 and 1964, for which existing estimates are either too infrequent, not disaggregated enough, or incomplete to be able to offer a coherent analysis. Given the political and institutional changes that followed, we argue that this period holds the key to what we term the Kuznets curse—the tendency in late-developing countries, according to Kuznets, for endogenous social conflicts linked to rapid structural change to be resolved by authoritarian regimes that ensure adherence to the high-savers accumulation model. We explain the political economy of the Kuznets curse through a narrative approach that combines structuralist development economics with a neorealist approach to institutional change, making use of public discourses and debates among policy-makers and intellectual elites.