It's about the data, not about the hypothesis. The p-value is a property of the *data given the null*, not a property of the *null given the data*.