On most tasks, memorization would be overfitting, but I think one would see that “overfitting” is task/generalization dependent. As long as accurate predictions are being made for new data, it doesn’t matter that it can cough up the old.