I would have played things differently than what AMD did except I would have bought ATI at the same time they did.
For CPUs:
K8: The last K8s would be 90nm, on Socket 939 with a DDR controller and two types: Single and Dual Core.
K9: It would be 65nm, the introduction of Socket AM2, a DDR2 controller, and a CPU with two cores or four cores designed kind of like the K8 but different: all cores would have two or three treads (two for Athlon X2, three for Athlon FX) 1MB or 2MB of L2 cache per two cores and 4MB L3 cache and make the quad cores just like the Core 2 Quad, put two Dual-Core K9s on one die. This would have beaten the Conroe.
K10: Use the same architecture but build a true quad core (just like today) but introduce Socket AM3 (which is backwards compatible with AM2) and a DDR3/DDR2 hybrid memory controller. Oh and 45nm. It would be built in the same way as the Phenom II but have many other additions that AMD didn't have time to incorporate like Multi-Treading (three treads or more per core) and I would have a Phenom FX which could mop the floor with Core i7.
For chipsets:
I would have made more 6xx series chipsets, not just the 690G
And today I would have a similar but very different AMD, one that wouldn't have gone the way of Global Foundaries because it would have a lot more money to play with and at this time I would have completely dominated the server market. And K11 is coming with 32nm or even jump over 32nm and go all the way to 22nm beating Intel completely.
Ahhh, one can dream, hehe!