I think the biggest obstacle to true AI is always going to be the "why". Science has matured bast the "because we can" point now and cognitive computing is as useful as actual AI could possibly be, without being true AI. The main drive is to push cognitive computing as we currently have it to greater speed, breadth and depth of knowledge rather than to develop it into a true AI.

The realm of generative self coding and self-improving computing processes is always going to belong to good old fashioned natural biological brains.