The thing is, AMD has been promoting system modularity; Fusion, as such, is an interesting concept: no more need to program a bus, you just have to ask for a calculation and it will be made by the most apt component in the ystem; whether it's a CPU's floating point unit or a bad-ass video card...
The socket game does, in fact, put something forward: when you get rid of design limitations, you can basicaly build a compatible system out of odds and ends. I mean, while AMD's 4x4 isn't interesting performance-wise, it is fascinating concept:wide: you can have a computer that acts as a single machine, while equipped with one to eight CPU cores, one to 4 graphics cards, one to two chipsets of a common family... Basically, it's mix n match.
Something Intel hasn't been able to do yet: multiple socket platforms need chipsets designed specifically for them and can't scale as much. Now I know that it's not that important right now, since Intel can still squeeze a few MHz out of FSB, but the relative ease with which AMD and Nvidia can glue together their components to make something new, is pretty interesting.
Now, if memory modules could be thrown wherever one may want in a system, we'd get high performance graphics and sizeable CPUs or astounding CPU perf (with only barebone graphics) with the same basic elements; and you'd just juggle what dedicated or all-purpose processor you'd require for a specific task.
We're not there yet, but AMD is getting close - while Intel isn't even in the race yet.
Which doesn't mean that a C2D doesn't kills a K8. But most investors don't care about actual CPU power: what they care about is deals, flashy presentations and bold moves. Otherwise, Intel stocks would have plummeted ever since they started the P4+Rambus.