You're partially right, but only in a very broad sense. One of the really ugly problems with web search is identifying real content from meaningless crap, and wikipedia rather nicely gets rid of nearly all the extraneous stuff, or boxes it in.

The actual semantics stuff is pretty robust, but like any piece of software can be tuned to get incrementally better results on any corpus.

Really, one of the major reasons we started with Wikipedia is that we wanted to build out the core search infrastructure and website infrastructure. These are non-trivial things to do while also setting up a search interface thats suitable for public consumption. I'm sure you can find a longtime powerlabs user to tell you how far we went in the interface and speed department.

Even just with the Powerset dataset, Microsoft now owns a significant number of concepts and the semantics and context that bind then. Without much effort, they can easily use this to improve the quality of their rankings simply by giving them a yard stick to compare to.

Is this page that talks about "flags" relevant to programmers or to heraldic buffs? Just check which Wikipedia page it compares better with.

With a lot of work by a lot of smart people this could become the first step in the long road to the much fangled "Semantic web"... or at least, some type of semantic search. Frankly, I'm surprised Google didn't buy them first. Unless they're already working on something similar.