So just download parts of the file like in an HTTP Range request.
I’m pretty sure S3, Google Cloud Storage (not google Drive) or Azure Blob storage will cope with the load, but for a price! They are much more expensive than Google Drive, Box etc.
50TB will cost you approx $1500 / month with them.
rclone mount does range requests to read things in the middle of blocks. It could be more efficient though as it does an open ended range request so potentially downloads more data than you need. You’d want to set --buffer 0 to stop that.