Go to a CS PhD program if you really want to innovate, forget all the crap that is spread on HN if you do get into top 5 or top 10 school, you will have equal amount of resources and at University even though they do have their own interests, they still want to see you succeed.
This is main difference between working and PhD, at a job you are a replaceable number at university depending on your advisor they are more concerned about your success and theirs as well.
First, innovation is not limited to what is of interest to a CS PhD program.
Second, I'm not interested in "innovating" per se, I'm interesting in creating things that add value to peoples lives and not being pigeonholed as either an engineer or a product person. I don't care if the ideas I'm using are new or not, only if what I'm building is enriching someones life and that my product has users.
Third, I've never been a replaceable number at a company, even large companies. If you work on hard problems, and are good at it, you aren't easily replaceable. People realize this, even when you're working for the man.
Finally, at university you often don't have access to real world data to do research. I know some big universities like epfl do partnerships where they work with researchers and engineers at companies in exchange for access to that companies data for research and publication.
However, there are just as many counter examples. Elon Musk left Stanford's grad school after one day!
The best "man" is your free will. Its not the "perfect man" but the best we have.
I happen to work for the man and here are some issues. I can't release code as open source, I get paid the same regardless of whether I work my ass off or just do enough to be ahead of the curve, I have virtually no say in our products because I'm just an engineer (we have product people for product design), I have to get an act of congress passed to setup a server in a non company standard configuration, etc, etc, etc.
There are some up sides, one is that I get to focus and think deeply about my specific problem space which at one time was search relevance, and is now data mining. I don't have to do sys admin work, even though sometimes I'd rather do it, I have access to a very large hadoop grid that non startup would have built prior to success, there are a lot of really smart people with diverse backgrounds to bounce ideas off of, etc, etc.
Oh, but this man I work on is focused on ad clicks :)