72 EU Skylake GPU model (with EDRAM) is already announced, and should be available in a few months. It is quite close to Xbox One in GPU (theoretical) performance already.
Both Apple (with PowerVR) and Intel could be competitors, if they wanted to invest the required money in creating semi-custom next gen console SOCs. But Apple and Intel both enjoy high profit margins in their current market segments, so I doubt they want to invest in lower profit margin business, such as gaming consoles.
On the Intel front, there are signs that its stance on customization and reaching out for outside business has faded a bit. Intel's cores are starting to specialize a bit based on segment, and AVX512 is showing that it is willing to fragment its ISA base further in pursuit of niches.
Apple already has significant pull on specific features for Intel's mobile architectures, perhaps including the 72 EU Skylake.
Other items include the dropping of IVR to get the thickness of its packages down, and rumors that Intel is dedicating porting resources to get its modem used by Apple.
Being allergic to lower-margin business does need to be weighed against the cost of underutilizing incredibly expensive fabs.
Apple could be argued as already being a vertically integrated competitor, in terms of offering products and services that compete with Microsoft and Sony in terms of media and apps. Its the worldwide leader in profitability for the phone and tablet platforms that have been blamed for hurting the console niche in general. How far, given that it is driving Intel to provide high-end integrated solutions, is Apple from just offering the last part of the console niche?
If AMD aren't going to try for reasonable profit margins with the next consoles, perhaps they shouldn't bother?
They're going to be in 100+ million consoles from this generation, but they continue to bleed out in the gutter with minimal revenue coming from providing something that no-one else could.
The console wins on their own seem to be doing reasonably well for themselves. They just can't compensate for the incredible money suck that is AMD's main business being in a 9-figure death spiral. Contracted products with well-defined lifespans, paid-for engineering costs, and a way to avoid AMD's inability to manage inventory are also pluses.
I could see the argument that they still might not be worth it in the end, given how much AMD has contorted itself to service those obligations and how it has distracted from AMD's ability to brace against the aforementioned death-spiral. Consoles have decent overlap with some of AMD's IP needs, but the Jaguar core's hops between foundries and a lack of high-end features that would make server and HPC clients happy mean that semicustom money paid for AMD to be distracted from staunching its bleeding.