Scuttlebutt said GloFos 14nm FinFet was gonna be used for CPUs/APUs only and that GPUs would use TSMC's 16nm FinFet. If they end up using 14nm for GPUs this could be very interesting.
It's also interesting that GloFos process is so flexible that you can use just one flavor of the technology to achieve both super high performance or super low power applications
I don't think initial 16nmFF/FF+ and 14nmLPP designs will differ much. New nodes are always immature in the first couple years. I wouldn't expect node differences, which are misleading and minor, to make a difference until at least the refreshes.
You got that whole Apple Samsung vs TSMC thing to add more weight to that argument of what others have already predicted.
Since Samsung and GF are pretty much the same tech, AMD can go with both and use the one ready first. Which also pitted both companies against each other and forced them to work together for solutions (my bet is here). I think AMD will still want to have work done at TSMC as well but that is different enough to warrant caution. Nvidia seems to be going with TSMC which could be limiting based on past TSMC node changes and could be disastrous.