> however inference is very profitable

Is there any actual evidence of that?