"Squeaks of excitement from academics"
I'm sorry, but where? I don't know of any academics squeaking with excitement, and speaking as an ersatz academic myself, I'm neither impressed nor squeaking.
Wolfram alpha is a thin and rather brittle NLP (natural lang. proc.) layer around some structured data sources and mathematica tricks. It is not innovative; this sort of approach has been studied before. Nor is it particularly well done -- try MIT's START question answering system ( http://start.csail.mit.edu/ ) for a taste of a well-engineered natural language question-answering system. True, it doesn't graph things for you (or infuriatingly render its text responses as images); but that's hardly a very useful or robust feature of Alpha.
(It's pretty symptomatic of the hybris of Alpha that the thing it does best -- invoke Mathematica and graph the result -- is let down by the interface. It turns out that natural English is not a good way to describe an arbitrary formula, and WA doesn't even start to solve it. Try graphing sin of x squared. Now try graphing the *other* sin of x squared. Etc.)
And for the most part, the conclusion of academia was that this approach -- representing knowledge with logical relations, and mangling English into some sort of relational query -- wasn't the silver bullet it was hoped to be. In fact, it was highly fraught task -- difficult and time consuming to do the "knowledge engineering"; and prone to dissolve under the weight of its own complexities.
Wolfram's claim seems to be that he has done something new, and that it will be revolutionary. Sadly, neither seems to be even defensible, let alone true. Another example of how "industrial research" seems frequently to discover more about engineering hype than engineering computation.