Despite the rise of Infinera and the PIC, the dream of the all-optical network still lives. The other day I read about the use of 'metamaterials' to slow down light, thereby enabling the next generation of speeds. These are the same materials that scientists are trying to make into invisibility cloaks, they can manipulate light as it passes through them - bending it, slowing it, etc. From a scientists point of view, this stuff is amazing. From an engineering point of view though, it doesn't yet exist.
You can tell that from the article, the author tries really hard to say just what the world of telecommunications might use it for, but he never really gets there. The idea it seems is that one could build a router that manipulates the light directly, rather than convert it to an electrical signal and operate on that level. Since the electronics have frequency limitations, this would in theory allow faster routers. But think of optical computing, something that remains a research project at best. Why would optical routing be any easier? We're good at electronics, we know how to increase speed while cutting costs for those. We know how to run them in parallel to increase throughput, we know how to mass produce the results.
Don't get me wrong, they will find ways to use this stuff - it's just that it's way too early to say just how. But I'll bet that when the engineers finally get their hands on it in 5-10 years, the all-optical paradigm and its challengers will have mutated beyond recognition.