Right now the 14nm process is for CPU's most of which have iGPU's. The next die shrink for GPU's 14nm or 16nm for the 2016 lineup, well for AMD, nVidia will probably do the same, along with HBM2.
Reason it's behind is mostly because Intel has so much money for process research, however they were recently beat to a functioning 10nm process, but it will take time for it to mature enough for market introduction. That's is why Broadwell was delayed, the 14nm process yields were problematic.
Only Intel and Samsung have been able to produce 14nm chips for now, and Intel usually reserves it's latest node technology for building only it's own processors (Samsungs node might not be suitable for high power chips like GPUs). It also takes atleast a few months just to move a existing design to a new node. TSMC (which produces most of nVidia and AMD chips) just announced that it has started volume shipments with it's 16nm node. Even if they are immediately starting to produce 16nm chips for both AMD and nVidia it will take a few months to actually have products in the shops. First you need to make sure these chips actually work, you need to make the reference design which you then send to OEMs (like Asus or MSI etc) which do their own evaluation and designs. After all this you still need to wait until enough chips are produced to actually start manufacturing (this depends on process yield, meaning how many of the cips produced are actually working). After manufacaturing starts you need to wait until they build up enough inventory to start selling (there's no point launching a product if there is only a handfull to sell worldwide).
Yep, from April 2012 to April 2016, that's a long time. And it will be even longer for the next one (10 nm).
16nm (and below) is extremely difficult to produce, so availability is limited.
Apple and other Mobile chips generate now more money than GPU, so they can pay higher price and gets priority for foundries like TSMC. Also, their chips don't require the same huge amount of transistors (like 5.2 billion for gtx980).