Fast inference = 6x markup? Don't be giving us any ideas 😼