This is the data repository accompanying Kruitwagen, L., Story, K., Friedrich, J., Byers, L., Skillman, S., & Hepburn, C. (2021) A global inventory of photovoltaic solar generating units, Nature, forthcoming. This repository contains the training, cross-validation, test, and predicted data set as described in the publication. The contents of this repository are briefly summarised here, see the publication for further details. Repository contents: trn_tiles.geojson: 18,570 rectangular areas-of-interest used for sampling training patch data. trn_polygons.geojson: 36,882 polygons obtained from OSM in 2017 used to label training patches. cv_tiles.geojson: 560 rectangular areas-of-interest used for sampling cross-validation data seeded from WRI GPPDB cv_polygons.geojson: 6,281 polygons corresponding to all PV solar generating units present in cv_tiles.geojson at the end of 2018. test_tiles.geojson: 122 rectangular regions-of-interest used for building the test set. test_polygons.geojson: 7,263 polygons corresponding to all utility-scale (>10kW) solar generating units present in test_tiles.geojson at the end of 2018. predicted_polygons.geojson: 68,661 polygons corresponding to predicted polygons in global deployment, capturing the status of deployed photovoltaic solar energy generating capacity at the end of 2018.