We migrated the US retail rendering fleet to AWS last year. simonw posted a link about it elsewhere on this thread.

So, about 4 years after it was publicly claimed to be running on AWS.

yes but we all figured this out earlier because EC2 and S3 downtime would never correspond with Amazon.com downtime - which made it obvious that the 'use the infrastructure we use' line was complete bullshit.

Not necessarily. Netflix managed to survive EC2 and S3 downtimes by architecting around the (mostly known) pitfalls - multiple regions, S3 isn't HA, etc.

This presentation contains some fairly detailed information about the amazon.com migration to AWS.

tl;dr The organic migration to AWS started in 2006 and continues to this day.

http://psav.mediasite.com/mediasite/Viewer/?peid=7ab95f6a5d4...

When I left in early 2007 they were working on the transition. More recently I heard that they were substantially if not totally on the AWS infrastructure. So if Werner said it recently it probably is true.

Amazon Silk uses the EC2 infrastructure -- they explicitly state this in their video ( http://www.youtube.com/watch?v=_u7F_56WhHk )

I was with Amazon in the past. I am no authority but I would come to a similar conclusion. Till end of 07, Amazon.com barely used AWS

I'm sure sometime in 2007 or 2008 a directive went down that people had to build new stuff on AWS. EC2 is a reasonably good fit for much of the Amazon.com code, as it's just generic hardware. I'm sure in the past 4 years more and more stuff has been written to use S3 as a data store. The core stuff, Gurupa, etc, would have to be re-architected to work with things like SQS and other services.

I'm not saying that nobody in Amazon wanted to use AWS, just that we didn't know about it until the press release, and so there wasn't any opportunity to use it... and of course, all the code from the 1990s was built in a different way so it would have been nontrivial to migrate it.