I guess for companies this can be nice? Run some inference in the off-hours on the machines that you already have. Could be useful for analytical tasks that don't require to be instant. On the technical side this doesn't seem that groundbreaking for me though.