Well it might, I mean Nvidia have an extra year to make sure it works compared to the competition. Plus to be fair if they release something even 10% faster they win because they already won.
AMD have been on a parallelisation / independent shader units / throughput minded architecture for years now (GCN). They also designed HBM with SK Hynix, and have a product out well ahead. They've also been working with interposers far longer. NVIDIA didn't even intend to use HBM until Volta got hugely delayed.
AMD's big thing for Arctic Islands is shrink, move to FF from bulk planar and almost certainly moving foundry from TSMC to GF/Samsung (I trust Samsung's FF process far more than TSMC's).
NVIDIA's big thing for Pascal is absolutely everything is new.