Large companies like IBM and Amazon must spend that kind of money on patent research, mainly to maintain a defensive portfolio of patents.
Not only will it help protect them in court from would-be infringers or other folks looking to sue one of the 'big boys' for just doing business, but a large portfolio of patents is handy for trading and cross-licensing agreements with partners/competitors in the industry, so everyone can get on with business (and not so incidentally, crowd out the smaller players).
Unfortunately, the real losers in this high-stakes game are the smaller industry players, who can't afford the R&D budget for such endeavors.
In particular, the recent abnoxious practice of patenting software algorithms, so popular in the U.S., has repeatedly threatened to cripple the free (as in freedom, not as in free beer) software industry, of which the GPL-licensed Linux is usually held up as the poster-boy.
SCO, in particular, is notorious for their onging legal battle to have IBM called on the carpet for releasing what SCO claims to be patentable algorithms from proprietary software that was somehow in IBM's posession into the open source Linux code base. The real kicker is that SCO has so far refused to reveal just exactly which lines of code are in question, because if they become identified, the open source community can just simply re-code around the algorithms in question, and thereby, eliminate SCO's case. What SCO wants to do, is no less than get court injunctions against all the major Linux providers, like Novell and RedHat, to force them to cease shipping Linux-based product.
So far, SCO has been held up to ridicule by most of the remainder of the software industry, but this kind of legal manuevering has been contemplated by Microsoft, as evidenced by their infamous "Halloween Papers". In those leaked papers, Microsoft was toying with the idea of using patent litigation as a means of gaining the upper hand over the increasingly popular open source movement.