Watson is made up of ten racks of IBM POWER 750 servers running Linux, and has 15Terabytes of RAM; 2,880 3.55GHz POWER7 processor cores and operates at 80 Teraflops. You're not going to find one of these at your local Best Buy.
Watson isn't just an ordinary supercomputer though crunching linear equations, the Linpack Benchmark, at ever faster speeds. By comparison, that's easy. No, IBM has been working on Watson for almost four years on solving the problem of 'understanding' natural language questions.
Sure if you ask a question Google just the right way, it will give you the right answer. But, as Stephen Chapman explains in his story on how to become a Google search ninja, you need to ask questions in a way that Google understands, which is far more complex and precise than the way you'd ask a person a question.
Expert systems, which have been around for decades, can answer natural language questions, but only within a narrowly-defined field. No one thought a machine could do well at "Jeopardy!" because there's just too much trivia-just like human contestants Watson isn't allowed to hunt for answers on the Internet-and Jeopardy's clues are mini-puzzles in themselves that require "understanding" before you can come up with the right question to their answers.