You can but most of the time, especially with Linked Data, an URL is required. And sometime even a dereferencable one. So you have this weird situation where you never really know if an URL is "semantic" (just a string actually) or can be accessed. Also having a dependency on the DNS system is a bad idea in my opinion. It's really clear when you read 2 years old semantic web paper and every link is broken.
I agree with gambler that GUID + local names is a better solution, and I use that in my research.
It seems rather overengineered to me. IETF and W3C standards do support the general content-addressing, "naming things with hashes" use case via the existing ni:// (Named Information) URI schema (RFC6920).
I agree with gambler that GUID + local names is a better solution, and I use that in my research.