I will try. The situation in manufacturing as i see it is roughly this:
Going into 45nm currently only offers performance-advantages if you use technologies like MG/Hi-K stacks - which come at a price: It significantly increases variation. What you get out of such a DFM are couple of parts with very nice characteristics on the upper end - but inevitably a lot of silicon at the low end as well which you need to find markets to sell it to. If you get back your silicon costs in these markets and building another fab is no headake, this is an option. However, getting your production cost back in handset-space is tough. 10USD for a CPU in handsets is expensive. If you lose only a little bit there it is no problem - just a price to pay for impressive performance-parts in serverspace, so Intel still might be ok with it.
AMD can't follow this model currently, they have neither capacities to do it nor the abilities to exploit the markets they would need - so a DFM minimizing variation is more suitable, which gets them probably nothing to brag about at the top-end, but much more within the envelope of characteristics suitbable for PC-space (not all - current variation does not allow to avoid silicon you can only sell in embedded space.)
Now variation increases from node to node as well, so AMD might be better off to stay one node behind or even more going forward under these considerations. I believe they can make similar margins with such DFMs as Intel - just because there are fewer and fewer segments of the markets where performance matters all that much. Even in serverspace this is a decicive criterion in some subsegments.
K.
Facts are irrelevant to the emotional brain
