[tomsem]

It is much, much easier to build, test, scale, and improve cloud services than to test an application that does the same thing running on a mobile or embedded system device like Kindle. There are not as many resource constraints, and you can have people work on it without them knowing anything about Scribe internals and development environment.

Amazon is not OCRing static images. Scribe samples at a very high rate (I think 100-200 per second) and notebooks are essentially recording a time series of events, these get 'played back' every time you open a notebook.