I think the problem is more storage than power. I don't know exactly how Google Translate works but I've heard it uses human translated information as a corpus

I think the problem is more storage than power. I don't know exactly how Google Translate works but I've heard it uses human translated information as a corpus in an attempt to get natural-sounding translations (as opposed to the goofy old babelfish type translations). Eg. Someone is translating something from Russian to English. They'll look back into their history of web pages which have human-translated Russian and English versions to see if any phrases in Russian match what they are actively translating and if they do, they'll attempt to incorporate the human-created English translation of that into their translation. The data allowing for this must be absolutely huge. Having said that I'm sure they could do an offline version which didn't use the fancy human-translated corpus data and just gave you the crappy babelfish-style translation, which in a lot of real-world situations is good enough to get the gist.