According to Computerworld, Mike Stonebraker says "RDBMSes, are 'long in the tooth' and 'should be considered legacy technology'". Knowing Stonebraker's work and reputation that seemed like some pretty startling information! It was less of a surprise to find that Computerworld hadn't quite got it right though...
You can read Mike Stonebraker's actual words at www.databasecolumn.com. He makes it clear that he's talking about the technology used for current DBMS implementations, not about the relational model or the SQL model itself. The columnar technology that Stonebraker has championed for years isn't an alternative data model at all; it's a particular mode of storage that in principle could be used by any DBMS, whether relational or not. As far as I know it has been used pretty much exclusively in SQL DBMSs (Sybase IQ for example).
Column store is perfect for BI and analytics. What would be even nicer would be to have the choice of column store or row store (or both) in a single DBMS. In fact I'd like to discard altogether the assumption made by most major DBMSs - that a base table automatically defines the main unit of storage.
It ought to be possible to map any selection, projection or join of tables into their own storage and to create redundant storage structures where necessary to support the best optimisations for different queries. Today, we have to use three different structures to achieve those things: partitions are a means of storing selections, indexes store projections, and "materialised" views are a slightly roundabout means of persisting joins (why do we have to create a view just in order to change the internal type of storage?). Each of those techniques have their own particular advantages and disadvantages.