>> Of course, you need to sell millions of chips to recover the massive investment of 15nm, but if you can, it still ends up being cheaper per chip.
Theoretical, that is the reason. Smaller feature size, small chips and that means more chips per wafer. However, other issues come up. It has worked for years, nothing says it cannot work. The point is that people must not be fixated on that size. Not every product needs small feature size of transistors.
THis will be quite a help for students in Engineering college or recent passouts. They can get lots of hands on experience without spending much. Companies can also make use of it for internal research and development or for trainees.
My impression is that the move to 15nm is not driven by the need for speed, but lower cost per chip. Of course, you need to sell millions of chips to recover the massive investment of 15nm, but if you can, it still ends up being cheaper per chip. And as for this being a "post Intel world", I note that TSMC is leading the way to 15nm, every bit as much as Intel.
It is still something that data and market successs must prove that firms can make money by this endless race to the bottom. What defines good products is not just speed. You need intelligence, power management etc on the chip. In the post-Intel era at least in the mobile world, speed is not the defining factor. You look for other factors. When you go below 22nm, I do not think you get lots of value when you benchmark the new problems created.