Right, each pandas column is a numpy array (or optionally pyarrow in pandas 2.0), but numpy is going to be faster for things that needs to be vectorized across columns, such as matrix operations.