I'm not saying hedge funds don't need big iron, they just don't need a Cray. Cray designs computers for a certain class of numerical problems that are difficult to parallelize, typically mesh optimizations, finite element analysis, and other very large linear algebra based codes. Cray supercomputers have very poor performance per dollar except in these cases where bisection bandwidth is more important than raw compute power. I am not familar with any problem in finance that has these characteristics. I guess they could exist, but for the most part financial computations are highly amenable to cheap parallelism. Martin