CouchOne, the commercial entity behind CouchDB even announced that it's moving away from associating the company with NoSQL as focuses on enabling offline data and applications.
The current orthodoxy would have you believe that if you are trying to get your head around "big data" or "Web scale" (see video), NoSQL is the answer. If you are dealing with preset data definitions being accessed by all the divisions of your global 100 company, SQL is better.
Here's the reality--relational databases have been around forever and Oracle, Microsoft SQL Server, MySQL, and IBM DB2 won't disappear any time soon. Too many vendors rely on RDBMS for their applications and the ecosystem around relational databases is extremely rich.
What's important to note is that using a database in a cloud-like manner requires system architects and developers recognize the principles associated with building a massively distributed data store.
Traditional SQL-based databases such as Oracle, Microsoft SQL Server, and IBM DB2 were designed to run on a single physical node/cluster in a single location, typically hooked to unified storage with full control over all software/hardware elements.
Running the databases in a virtualized environment with multiple nodes and very limited control imposes obstacles that are difficult to overcome. The wave of NoSQL databases seen recently is a reaction to these limitations.