This is a dataset derived from Microsoft's US Building Footprints dataset. It consists of one shapefile per county, spatially indexed, along with some additional attributes (area in sq. m., centroid, etc.). Buildings that intersect multiple counties appear in multiple shapefiles.
I transformed the original dataset because shapefiles are convenient for spatial indexing and storing computed attributes for the shapes. The max size for a shapefile is 2 GB, and the largest geojson files in the original dataset are larger than this.
The dataset can be found here: s3://glr-ds-us-building-footprints. Individual files are called {fips_code}.{extension}, eg. 06075.shp.
To download and use the files locally, you'll need the files with the following extensions: .dbf, .prj, .qix, .shp, .shx.
NOTE: You need valid AWS credentials to interact with the bucket. Set these up with aws configure from the CLI. CLI docs here: https://docs.aws.amazon.com/cli/latest/userguide/cli-chap-welcome.html
The example script demonstrates loading a shapefile, performing an intersection, and computing some basic information about the intersecting buildings. The example requires a version of fiona with s3 support: see requirements.