Could AI with infinite memory lead to self-recursive improvement?

I ask because I currently can't see how it wouldn't under certain circumstances but would mostly like to be corrected if I'm wrong or told what else would be required.