Probably delays. We first heard rumblings that this GPU would be arriving Q4 2014 or Q1 2015. Anyway, when you spend R&D money on technology, it makes sense to actually get it out there at some point so you can make a return of some kind. These technologies have a long development pipeline.
Also, there is an awful lot that the layman does not understand about how much VRAM is needed for any particular scenario. With GDDR5, to increase bandwidth needed to feed these GPUs, designers had to add more memory modules. This meant that capacity increased "for free". It was not always the case that more VRAM would actually be useful for said GPU.
As AMD have stated, there was essentially zero VRAM management going on with their GPUs with GDDR5 and a lot of data sitting in memory was never touched by the application.
AMD have some great engineers and I'm sure that Fiji is going to do very well in bandwidth limited scenarios, such as high resolutions combined with anti-aliasing.