They are making design decisions (that can't be changed without delaying the development process) 2-3 years before the product hits the market so not so easy to predict how things will play out so far ahead. Long term HBM/interposers were and are going to be an important component of their products, not just for their consumer graphics range. They had to get behind it otherwise why would SK-Hynix put sufficient resources into development. At a time when they had to run lean it makes sense that they backed HBM. I wouldn't be at all surprised to see GDDR5X in a future entry level GPU (Amusingly probably not far off Hawaii/Polaris performance level).
Nvidia were effectively waiting for AMD to do the early heavy lifting (sensible). Way back they were originally looking at Intel/Micron's HMC memory. They had a quick test run with HBM in high margin markets prior to their large HPC contracts which commit them to delivering the V100. Having the lions share of the Professional/HPC market already and particularly these contracts allowed them to throw out P100 and now the massive V100 in their existing implementations. If they didn't have those contracts its more than likely that an HBM Volta would be several quarters later and/or smaller in scope.
