# Why kernel perceptron relies more on the training data than kernel SVM?

In UCSD machine learning course, it is said that:

"for Kernel Perceptron, the solution is likely to depend on more of the training points than the Kernel SVM"

why is that?