We propose and test empirically a theory describing the endogenous formation and persistence of mega-states, using China as an example. We suggest that the relative timing of the emergence of agricultural societies, and their distance from each other, set off a race between their autochthonous state-building projects, which determines their extent and persistence. Using a novel dataset describing the historical presence of Chinese states, prehistoric development, the diffusion of agriculture, and migratory distance across 1° × 1° grid cells in eastern Asia, we find that cells that adopted agriculture earlier and were close to Erlitou – the earliest political center in eastern Asia – remained under Chinese control for longer and continue to be a part of China today. By contrast, cells that adopted agriculture early and were located further from Erlitou developed into independent states, as agriculture provided the fertile ground for state-formation, while isolation provided time for them to develop and confront the expanding Chinese empire. Our study sheds important light on why eastern Asia kept reproducing a mega-state in the area that became China and on the determinants of its borders with other states.