Having intimate knowledge with Watson and analytics (this is my profession and you can PM me if you like), you need to understand one critical thing about Watson: it is not a product but instead, a highly customized set of service offerings designed to build an outcome. Don't get me wrong, there is a variety of IP associated with Watson (think of toolkits), but what you are buying in most cases is a services engagement designed to customize these toolkits and associated logic to work to a certain outcome. IBM is working to "productize" Watson such as some of their commercial offerings, but it is definitely not something that you can buy and turn on to get immediate results.
What it did on Jeopardy was impressive, but for different reasons than you may be thinking. The questions and answers used for the Jeopardy tapings came from were part of a corpus of data that was loaded into Watson. Watson was trained on this data. The impressive part was not that Watson could find the answers (since they were already loaded), but that it could use text analytics to interpret how Jeopardy phrased its answers to find its questions. This was the impressive part from a technology perspective.
AI and ML is making impressive leaps but keep in mind, none of this is self learning like you might think from watching movies. We are nowhere near a Skynet scenario. About the most impressive things you can do with ML at this time involves scenarios similar to what I am working with clients on in using ML to analyze security event data to look for patterns that we do not know exist, and thus can not codify. With that being said, it still can only discover patterns within the parameters that you code it to go look in and the constraints that it can work within.