Graph databases are ones where I feel there is a lot of potential but now we have graph capabilities in SQL Server. I think we need some really good examples and resulting shift in thinking to get the most out of these. Because these are based on sound mathematical theory (just like RDBMS) these I think will grow.
The question is whether graph DBMSs can do things that at RDBMS cannot do, or if they can do the same things in a more concise and flexible manner?
Functional programming has a sound mathematical basis, but for data management is clearly less flexible than the relational model.
SQL is just once possible language to use for an RDBMS and a language more faithful to the relational model and that removes much of the redundancies and inconsistencies of SQL would be a great step forward (a relational NoSQL DBMS)
Regarding performance and scalability there is no need to abandon the RDBMS to achieve this - with an RDBMS the implementer is completely free to choose whatever physical method of storage they find most appropriate - without changing the interface (SQL in the case of SQL DBMSs).
I am a bit puzzled by Hadoop - programmers write programs that are submitted as jobs - this seems like a return to 1970s style mainframe batch programming - something the we plucky young bucks at the start of the eighties swept away with online, real time systems.