I think we will see something novel regarding parallel processing, like a combination of a "traditional" CPU (probably a few of them on chip) with more than enough horse power for most traditional applications, connected to a massively paralleled, high speed GPU that will join in on "general computing". I think the key here would be the removal of any bottle necks in between them, to get *massive*, super fast, direct, on-chip bandwidth in between them. That would bring a whole new meaning to CUDA, and this together would bring *a whole new class* of processors.
Could be wrong of course! But there are interesting things ahead...

That's an interesting approach and it allows you to retain the performance for when you need it.
Basically, I decided that most of the time I just didn't need it so my current primary machines are two Intel Atom powered netbooks (I'm typing on a 1.8GHz one right now).
They were cheaper then the price of a Chromebook, they have real hard drives (not cloud or flash storage), and they probably perform as well or slightly better then a similar ARM based system.
Plus they run X86 Windows apps (no small selling point).
And, of course, when I still want my Amiga flavored fix I can used my PPC based MorphOS system.