Are 14 nm yields good enough to allow for such huge dies? We've seen how poor the Intel 14 nm yields have been, despite that their process has been live for a long time now. And Intel's chips are much smaller than that. So is TSMC/GF/Samsung 14/16 nm yields that much better, so they can produce 450+ mm2 chips in 2016 at reasonable cost?
Intel isn't comparable to the pure-play foundries because what is profitable for them is much different to a company like AMD. Intel has no margin for error and needs to balance die sizes compared to their own previous nodes.
AMD is literally at the bottom of the barrel (massive, crap bulldozer-derived CPUs and power-hungry GPUs) so node advances like this only have an upside.
In this instance, think back to 2009. The last time AMD had a big lead over Nvidia was when Cypress was released and Fermi was just theoretically a disaster. This time we know that Nvidia needs to get HBM and FinFETs right on a big die on 14nm so the easiest way for AMD to make Pascal another Fermi is to put out their own big GPU first.
AMD can't allow Nvidia free reign to release something "ok" and big just because they have nothing big enough to compete. That's why Fermi was so bad so early - because AMD competed and forced Nvidia to delay and delay.
If AMD doesn't show up with a huge 14nm die then they deserve the losers tag.