really? Kepler is coming out?
Everyone wants to speculate on future product, but frankly no one knows when it is coming out, nor do they know if there's going to be implementation issues with the 28nm process. Do remember NVIDIA gambled (and lost big) with Fermi on 40nm.
NVIDIA's strategy has always been to go big and scale down accordingly. Great if you have a mature process, not so great if you run into yield and manufacturing issues. AMD's strategy (midrange product, then populate stack up or down) is less risky but at the expense of outright performance (out of the gate).