I didn't mean that training time is the constraint. It's that when training a more general algorithms (ANN) your hypothesis space has more dimensions than when

I didn't mean that training time is the constraint. It's that when training a more general algorithms (ANN) your hypothesis space has more dimensions than when training a more specialized one (SVM). You therefore need more data to train an ANN than a SVM. The reason for choosing SVM is not that your dataset is too big for ANN, it's that it's too small for ANN.