Microsoft has made 124,885,597 footprints from all 50 U.S. states available as open data. The building footprints were extracted from Bing imagery using a combination of deep learning to identify building polygons and then a polygonization algorithm to clean up the edges of the buildings.
This process was developed in order to rapidly increase the amount of building footprints available in OpenStreetMap. The Bing Blog calculated that prior to this data delivery, OpenStreetMap had 30,567,953 U.S. building footprints.
An explanation of the process and free data download of the building footprint by U.S. state (in JSON format) is available from GitHub. The toolkit used by Microsoft is also available on GitHub as well.