Since I spent a few years working in EMC's Web Hosting division..
Yeah the company had one of those and it started when some big customers decided they wanted EMC to maintain their web presence back during the boom. I got a job there literally a few months after some big customer said they planned on moving into the EMEA market with an online e-commerce presence and mentioned that they thought EMC's Cork facility would be a good place for it to reside.
Before you knew it they were looking for people to watch the screens at god awful times at night.
So there I was having escaped from manufacturing trying not to trip over all the hastily run network cables while listening to music cranked up way too loud in a small room that used to get so hot that if you didn't open a Window and leave the bitterly cold night air in the Ultra2 box used solely for displaying forwarded X sessions from the monitoring tools would shut down.
Back then it was this weird pirate organisation off on it's own filled with brilliant yet bizarre people from different technical backgrounds running as fast as they could while spending money like water.
And it was amazing.
By contrast the EMC of the time was a painfully dull one trick company (Dull unless you were kicking ass in Sales) which believed in a slow death by pointless multi-hour meeting.
..I think I know what Amazon Web Services cardinal sin is.
They've been treating thousands of Web 2.0 developer customers like dozens of Web 1.0 hosting customers. Scratch that, they've been treating them worse.
This isn't about the fact they had a failure. Even the Internet has failures. (Especially when someone drops a boat anchor sheers the wrong cable and blows the Middle East offline) It's the fact that they don't communicate with their customers at the speed their customers operate at.
Having fixed the problem AWS have or are in the process of writing up a post mortem and filing all the reports which need to be filed. When that's done they'll finally get around to explaining why one of their three locations was inaccessible for a period of time.
But no matter what level of detail they provide it's too little much too late as you've had all these Web Developers (And that's who AWS sell to) who spend their day sitting on Twitter and communicating at the speed of a page refresh kept in the dark for hours. To those people hours are like days.
When you look at AWS what you're looking at is the R&D proving ground for Amazon.com. They've built this massive infrastructure which you can rent space and time on and use their services to run your business, but a lot of those services are in Beta and aren't being used by Amazon's cash cow yet.
You're a fee paying beta testers for technologies Amazon is looking to prove before they'd introduce them into the infrastructure which runs Amazon.com or marksandspencer.com.
You know e-commerce sites where sales revenue is measured in Millions and not Hundreds or Thousands.
So what we have is a symbiotic relationship. Amazon get to test services with a realistic workload of millions of concurrent connections and billions of stored objects (As of October 07 it was 10 Billion objects) in a sandbox while developer customers using those services get cheap storage, cheap compute cycles and a host of bells and whistles they can use to reduce or eliminate the infrastructure advantage of their larger rivals.
I think it works out well for all parties involved, occasional failures and all.