OurPCB is a Chinese PCB fab outfit with an initial cost of $57.00 and a subsequent charge of around 15 to 25 cents per square inch for 2-layer boards and 50 to 75 cents per square inch for 4-layer boards. The per-square-inch costs shrink for bigger orders. So PCB fab and assembly is cheap. What do I do with this? Obviously there is a huge opportunity to do something.
PCBs and assembly (even surface mount assembly) are no longer a significant obstacle to complex electronic projects. The next obstacle is that I'm lazy about learning. To some extent that can be addressed by prioritizing what to learn - if USB does all the communication I need, I can ignore PCI.
We could have an electronics hobbyist renaissance as good as the 1970s, starting with a series of articles in Make magazine. O'Reilly would probably love it.
The economics works better for big boards than for small boards. I can envision scalable VLIW array processors spanning several FPGAs, or maybe tightly networked DSPs or GPUs. I like the look of the Analog Devices ADSP-31362. I could review some of the molecular modeling code out there (Amber, CHARMM, Gamess, Gaussian, Gromacs, NAMD) and build an scalable architecture optimized for molecular modeling, large-scale simulations, and other interesting things.
Building a supercomputer, that's so unimaginative. I don't have any interesting problems to feed it. I suppose I could build it and let other hobbyists figure out what problems to throw at it.