Werner Vogels gave an interesting talk on the how and why of Amazon Web Services. It was peppered with lively examples of AWS success stories where customers coped with surges in demand or performed large amounts of one off tasks using AWS. One of these was Animoto, a small company "run out of a coffee shop in Seattle" whose customer base went up by a factor of 10 over a couple of days, at one point adding 25K customers per hour when the created a Facebook app. AWS allowed them to seamlessly scale to meet this demand.
Amazons had 1000 different services, each supported by an independent team, the philosophy of AWS reflects this as it tries not to impose a strict model on its users.
One of the reasons for simplicity of the interface is to allow client to move away from AWS should they want or need to. Grids were contrasted with this where their complex API would lock clients in to a solution (and so Amazon did not take the Grid approach when developing AWS)
Everything fails all the time, so your application have to deal with it. Amazon claim that the loss of a data centre should not affect their SLA with a customer.
Incremental and elastic scalability is vital. Should be able to acquire resources on demand and also release then when not required.
Let go of control, there is no way that you will know every input and output. They use probabilistic techniques.
An average experience means that 50% of the customers experience worse. Aim for improvement at the 99.9% level.
Move investment from capital to variable cost.
His view of the relationship between Grids and Clouds is that clouds provide access to the raw resources while Grids provide higher level services.