Oh, my toosh, no. I don't agree with everything Matt said, but on this, you're totally off. I try pretty hard to do research that is (a) potentially industry-relevant (key word: potentially); and that (b) industry won't do for various reasons. Thus far, it's seemed to work pretty well. Using my favorite example of the week, take some of our work on cuckoo hashing -- it's contributed quite a bit to the literature of how to implement fast cuckoo hashes, contributed a new and fairly intriguing data structure for approximate set membership, and produced a design that I know to be in use in two Very Large Companies(tm).
The companies wouldn't have done this work, at least outside of their research labs, because the solution's theory is too far from the need of any one problem. But the result -- a more memory-efficient hash table design -- turns out to be broadly useful.
And yes, I do consider this to be "industry-relevant" research. I'm not going to solve their problems for them -- but there can be great synergies between industry and academia for having broad impact through adoption.
(full disclosure: I'm an academic on sabbatical at Google for the year. It's likely I'm a little biased in my belief that both have value. But this isn't a bias unique to me; systems as a general area is close to industry, and most of my colleagues rotate in and out of industry periodically via sabbaticals or startups.)