The race for petaflop supercomputing (equivalent to something like 200,000 PC's) is on. the IBM Blue Gene project has been going on for some time and toppled Japan's Earth Simulator half a year to one year ago - but now Japan is set to build a 10 petaflop machine a remarkable 73 times faster than the currently fastest version of the Blue Gene machine.
It's supposed to go live as early as 2010 which means that top speed of computing clusters will more than double every year for the next five years. Obviously since we're talking clusters, money buys almost any annual speed increase you like. The management hell of running the clusters if you don't innovate will of course kill you in the end in the form of permanent never ending hardware replacements and impossible energy management.