No, not all features need to be scaled. In general, features that are on a similar scale will work better with machine learning algorithms, but there are some algorithms that are scale-invariant. For example, tree-based algorithms are not affected by feature scaling.
