Optimization expertise, debuggers etc. are much more advanced for "AAA" games on x86 and powerPC than they are for ARM. That's where the bulk of current developers' experience lies.
First off, there's no such thing as 'x86 optimisation expertise'. You can be an expert with this or that microarchitecture (or a few of them at the same time), but x86 and PPC are both ISAs, not a microarchitecture. Reality in x86 world is that what yields nice IPC on a Haswell might not on a Jaguar, etc. Heck, the differences between IvyBridge and Haswell are not negligible, to the point that one'd have to be careful when targeting the two microarchitectures with the same binaries, if optimal performance on either of them was sought. Devs get proficient with the hw anew with the start of each new generation, and x86 does not magically dispel that. You can ask any ps4/xb1 dev about the cache intricacies of the dual-cluster Jaguar configs in the current consoles - you'd be excited to learn about the 'x86 desktop experiences' they got there.
From there on, Debuggers are 90% as good on today's ARMs as they are on today's x86's (essentially the same debuggers outside of ms windows). Compilers are 100% as good on today's ARM as they are on today's x86 (they are essentially the same if we focus on gcc & clang, which both have excellent ARM backends, and clang has been compiler of choice for SCE this gen). Profilers are 90% as good on ARM as they're on x86 (perf has had aarch64 support since 2012). Of course all those numbers are my personal estimates, but they come from actual experience across multiple architectures.
ARM are not some 'second class' citizen in toolchain-land. And that shouldn't surprise anybody, as way more developers target ARM (v7, etc) these days than x86, in total. So the entire 'x86 expertise' angle is hardly a factor. Seriously, if you're about to play the expertise card, ARM gets the popularity vote, easily, among all modern ISAs.
Not sure what you're on about. I was using Intel as an example of competitive perf/W in the mobile space. My argument has been an AMD design all along.
Well, I guess I fell for the 'premium experience' red herring there. A premium experience in my eyes is something traditionally provided by an 4..8-core desktop Intel and an AMD/NV standalone GPU, so I went on a SoC trajectory from there. Anything else requires very serious efforts on developers' part, as repeatedly demonstrated by pc/console multiplat devs this gen, x86 and all (Ubisoft's recent discussion, etc.)
Do you think any of those are as well position as AMD to provide a performance competitive GPU with wide API compliance and customized CPU core on the same die? Now add to the mix that AMD could do an x86 or ARMv8 custom core and it becomes hard to imagine anyone else with a competitive advantage on them.
Let's see. NV can, easily (sans the x86 part, apparently, the value of which is subject to this very discussion). Apple have proven that they can (with the help of IMG), alas they're not interested. Qualcomm (to whom once upon a time AMD sold ATI's entire mobile division, engineers, architects and all), also can, and are actually interested. Broadcom could, if given the incentive, and the same goes for AMCC. The thing about open ecosystem architectures is that quite a few IP vendors are willing to provide you with all kinds of added-value IP. For instance, three vendors (IMG, ARM, Vivante) offer ARM-tailored GPU IPs, and every ARM vendor could use any of those IPs with their designs, if they don't have a domestic one already. Now, how many of those IP vendors could match AMD at the high end of the spectrum? Perhaps none, but two of the in-house GPU vendors - NV and QCOMM, could, and would, if given the opportunity.
Last but not least, AMD can provide nintendo with all the ARM-centric tech nitendo might need for their next-gen console SoC. So I'm not quite following your AMD argument here.