Lookup stages load the entire reference dataset into memory for fast matching, requiring significantly more RAM than Merge or Join stages which process data in streams. The primary input link does NOT need to be sorted for Lookup - that's a requirement for Join stages.