The Li-stuffed garnets LixM2M3′O12 are promising Li-ion solid electrolytes with potential use in solid-state batteries. One strategy for optimising ionic conductivities in these materials is to tune lithium stoichiometries through aliovalent doping, which is often assumed to produce proportionate numbers of charge-compensating Li vacancies. The native defect chemistry of the Li-stuffed garnets, and their response to doping, however, are not well understood, and it is unknown to what degree a simple vacancy-compensation model is valid. Here, we report hybrid density-functional–theory calculations of a broad range of native defects in the prototypical Li-garnet Li7La3Zr2O12. We calculate equilibrium defect concentrations as a function of synthesis conditions, and model the response of these defect populations to extrinsic doping. We predict a rich defect chemistry that includes Li and O vacancies and interstitials, and significant numbers of cation-antisite defects. Under reducing conditions, O vacancies act as colour-centres by trapping electrons. We find that supervalent (donor) doping does not produce charge compensating Li vacancies under all synthesis conditions; under Li-rich / Zr-poor conditions the dominant compensating defects are LiZr antisites, and Li stoichiometries strongly deviate from those predicted by simple “vacancy compensation” models.