Your linked article doesn't show that, at all. AWS as we know it today essentially had its first public release in 2006 with EC2 and S3, which means it existed for 7 years before your article even breaks out profit for AWS in 2013.

But is there anything to support the huge losses angle?

It's hard to compare apples to apples here because AWS leverages infrastructure Amazon already had to keep around for seasonal surges to its core business. I do not believe that Google has excess capacity for seasonal demand.

This was the story given to public. The reality was that all of the infrastructure for AWS was brand new purpose built for AWS. There was no sharing between the stacks. Source worked there in the early days. The story sold was that Amazon was utilizing spare capacity from retail side, completely wrong.

Thanks for your post. That origin myth never made sense to me in the first place: "OK, so during the holidays when retail needs that spike in capacity again, where do the AWS customers go? Oh yeah, and a lot of those customers also have the same seasonality patterns as Amazon retail."

The hardware was? Or the software? Both?

From their historical prices, I find it difficult to believe they're not dogfooding.

Everything was purpose built for AWS. Retail was just coming onto AWS side around 2011/2.