Conceptual impedance matching is the biggest problem. To use a personal example, I've got this silly idea, the BitGrid, which might (but probably not) prove to

Conceptual impedance matching is the biggest problem. To use a personal example, I've got this silly idea, the BitGrid, which might (but probably not) prove to be the optimum way to get petaflop computing to the masses. It's based on some obviously bad design choices (when viewed from the FPGA world, for example), so that someone familiar with that world would quickly come to the conclusion that it's just not worth consideration. Programming the thing requires a completely new set of tools, as there's no Von Neuman architecture to be found. It breaks my brain a little bit, every time I work on it. Completely writing new software isn't something we've had to do since the 1960s. We've always been able to bootstrap off of old ideas, and old code. Effectively, none of the tools normally used to develop software help. The "impedance mismatch" of this thing is just too big for anyone to even think about making a startup around it. Something that was a small improvement on FPGAs would be far easier to sell, but there's not much investment money in hardware, let alone ASICs. A similar example would be the Jet Engine... nobody thought it could work, and it required a completely new way of doing things. Thus, nobody wanted to try to develop it. Big changes require stable supplies of long term research and development funding. That's something the market just isn't equipped to handle, especially when driven to meet the quarter's numbers.