There has been much speculation following the NYTimes report about Google’s amazing new data center, located near a large river primarily because it needs a lot of power. Why do they want all that storage and processing power?
One interesting idea from Ian Betteridge: To learn to model context. Sounds plausible to me, though I think a reasonable model of how we think has much wider applicability than merely getting the right ads in front of you at the right time.
Applied Abstractions 
You’re completely right to say that it has a much wider application than just ads – but ads are (at present) Google’s bread and butter so they’re the immediate business case for better modelling of context.
Of course, understanding context also leads to better personal search results, which is at the heart of Google’s business. For example, understanding what I’m talking about when I mention “a full house” is a tricky business. Do I mean my house was full because of a party? Do I mean something to do with cards? Or was I visiting a show and couldn’t get a ticket? If Google understands that contexts in which I’ve previously used that phrase – perhaps by scanning my Writely documents – then it can give me better search results, and a lot more besides.