Word2Vec is not deep learning. I think the neural network it uses has a single hidden layer. It's much simpler than deep-learning approaches and all the neural network is doing is dimensionality reduction in a way that can be done with SVD for similar results.