OpenStreetMap (OSM) has evolved as a popular geospatial dataset for global studies, such as monitoring progress towards the Sustainable Development Goals (SDGs). However, many global applications turn a blind eye on its uneven spatial coverage. We utilized a regression model to infer OSM building completeness within 13,189 urban agglomerations home to 50% of the global population. Our results reveal that for 1,510 cities OSM building footprint data exceeds 80% completeness. Humanitarian mapping efforts have significantly improved completeness, especially in low SHDI regions. The digital divide in OSM has receded, but a strong spatial bias associated with subnational human development index, city size and World Bank region remains. In consequence, global studies will provide misleading results if the biases in OSM's coverage are not accounted for. We recommend combining completeness maps with socio-demographic information to guide mapping to ensure that "nobody is left behind" as encouraged by the SDGs.