AUC of imbalanced data

I understand if you look at the AUC of a ROC curve of a balanced data set it should be over 0.5 otherwise your classifier is horrible.

Now what happens if you have imbalanced data? Say 90%-10%? Does the AUC have to be over 90%? Over 50%? How does it work, how does it change and how do you calculate the new ratio?