I think the main constraint is the lack of creativity and the variety of tasks one has to perform in a big company. In a startup there are so many challenges and so many things to learn its always changing,
Interestingly, Google research is not very good in this respect due to their focus on products. Restricts creativity; if you really want to do products and be creative, get into a startup instead.
Btw, compensation wise, MSR is one of the best as well. The only (and quite serious) downside to MSR is their stupid insistence on using Windows everywhere. Researchers absolutely hate that.