I meant, physically, what gives different? why after applying the gate once on 1 there is a 50% the spins flip but when applying to 0 no-chance of that.
The states you get are H(|0>)=|0>+|1> and H(|1>)=|0>-|1>, both of which have 50% probablity to end up as 0 or 1 once they are measured. The difference of the "-" in the second state is only visible if you carry on the computation, not if you measure.