Werner listed a number of principles that worked for Amazon to achieve the current success. He recommends to read his blog to know more about them. One for all: "simplicity is the key, if you make it simple, you can make it scale".
Since few months, the Grid community started to wonder about what is Grid vs. what is Cloud. Yesterday, two sessions were fully dedicated to this (unfortunately I could not attend them). Richardson, Jha et Al and Bégin presented their contribution to the topic.
Werner's viewpoint is that, when they started their work on AWS, they investigated the Grid technologies: "why we didn't do a grid, worried about bind to a framework, we wanted to make sure to provide basic resources where to build upon; today there are lots of Grids that build on cloud, e.g., in pharmaceutical area".
I tend to share his view. Actually, last September together with colleagues at CNAF we made an investigation on how we could expose Amazon S3 using StoRM, an implementation of the Storage Resource Manager interface (SRM). The feasibility study was presented at CHEP 2007 and we wrote also a small paper on this topic. It would be interested to make a similar study in order to investigate the usage of EC2 as a back-end for a Grid-enabled computing service.
BTW, I'm indirectly an AWS user. Since more than one year, I host my pictures using the wonderful services provided by SmugMug. The company, located in Mountain View, started 5 years ago to provide their services building on their own IT resources (i.e., buy, install, configure, deploy, test, manage). With the advent of AWS, they moved their services to them and now they host more than half petabyte of pictures (without counting replication) from around 500K customers on S3. They also make advanced usage of EC2 for the processing of photos and videos (you can read more on this fresh new post and in the older posts).