The industry and the systems our ICs go in need both -- the advanced process geometries for the Big D-Little A SoCs, and the larger geometry boutique processes for the Big A-Little D analog, power and MEMS functions.
Miniaturization is still critically important, but there are other ways to achieve it besides making one big die in the latest process -- SIP and TSV for example.
There are still plenty of opportunities for fabless startups, but the name of the game is IP, whether digital, analog or electro-mechanical. This has been true for a very long time, ever since the rise of the fabless business model.
Startups are by definition fabless and also don't have deep pockets. Show me a startup with valuable new IP and patents to protect it, and I'll show you a potential ROI. But show me a startup whose business plan is based on taping out the next great 28 nm SoC and I'll show you a bunch of dreamers who will never get funding.
Peter--what you are saying is that success comes from going with analog and MEMS--"we" analog folks knew this all along! Many analog companies have done very well by NOT looking to the next process node for success, but by using older processes to the max--fab is easier to get, less expensive, and plenty of good, used equipment for fab and test is available cheap!