Hadoop is a buzzword now. A lot of start-ups use it (or just say, that they use it), a lot of widely known companies use it. But when and what is the border? When person can say: "Better to solve it without Hadoop"?
It's an economic calculation, really. When you have a computing "problem" (in the most general possible sense) that you can't solve with one computer, it makes sense to use a cluster of commodity machines when doing so A. allows you to solve the problem, and B. is cheaper than a forklift upgrade to a bigger computer, or upgrading to specialized hardware.
When those things are true, and you are going the "commodity cluster" route, Hadoop makes a lot of sense, especially if the nature of the problem maps (no pun intended) well to MapReduce. If it doesn't, one shouldn't be scared to consider "older" cluster approaches like a Beowulf cluster using MPI or OpenMP.
That said, the newer YARN based Hadoop does support a form of MPI, so those worlds are starting to move closer together.