If the data is that massive, then keeping a copy of it in every single result will very likely exceed the memory/data limits of each stage