Intel have had the best fabs for 15, even 20 years and that has _always_ given them an advantage. Semiconductor economics being what they are, it'd take a revolution for AMD to leapfrog.
When Intel were making a 'misstep' with P4 and AMD had a solid K8 and were pushing the technology on with x86-64 while intel were determined not to, was one such time they got ahead. The one time.
And yet, yes they still largely got(get) put into crap boxes by OEMs. I agree exactly what was already said about perception on this.
Intel are surprisingly well run (consider for comparison, practically any other tech manufacturer). They have moved aggressively into low power and they have it locked up on three angles; OEM, architecture and manufacturing. Their blind spot seems to have been ARM, and that, just from looking at their press releases since at least 2011 you can probably figure out what they themselves figured their main competitor has become.
AMD chips are not bad. But tell me, at this point as a company they're not being kept on life support, by intel to preventively fend off antitrust. There's been little pressure to improve IPC, and we, coincidentally, haven't seen much. I'm not saying they're sitting on their laurels but they have the headroom to pursue other performance targets (power consumption). How long were AMD stuck on 28nm?.
(and to be specific as regards the article for the only time in this post, they have no mobile presence either, and that's the only growth area. Qualcomm, Samsung, a dozen others make money here, but AMD don't)
As far as what we used to think of CPU's being for, I hoped HSA (as it's called now) would be a new revolution, which sounded much more progressive in 2006 when they bought ATI. It's 2015 and it's still not a flagship, but just some on-chip chew toy for at best mid range laptops.
How did this take so long to go so short a distance? (I'd guess because 10bn transistors in a 300w package would melt the heatsink, let alone the chip, but let's move on). AMD would need a change of this sort of magnitude, only it would also have to pay off.
And I'm stealing the last thought from I can't remember where, but as intel get slower on process nodes, maybe globalfoundries get somewhere nearer parity (but probably not). I think AMD make solid products (Fiji is near enough nv's 980 I'm going to award them the point), but if they can't fab them they might as well draw circuits with the graphite pencils for all their hope of getting ahead.