This article lists some of the AWS services that now power Amazon.com since they moved away from Oracle. So, while it may not have been true in the past, it does now appear that AWS powers a significant portion of Amazon.com.
Amazon using AWS today is entirely different from the origins of AWS or GCP or Azure being productizations of internal infrastructure.
AWS services have always been their own thing, they happen to be useful so today Amazon utilizes them just like anyone else. That doesn't mean the services started internally.
I think you're being a bit too literal here. Did AWS merely fork the infrastructure that Amazon.com runs on and start selling it, no. But are you telling me that it wasn't built on the lessons learned of amazon.com and that it doesn't share some aspects of how amazon.com worked a the time (with improvements to the bad parts)? I don't know the answer and maybe it was a clean-room implementation but that seems unlikely to me.
I don't doubt that Amazon.com runs on its own dedicated "installation" of AWS, but are you saying that it doesn't run on AWS at all? That doesn't jive with the infamous "API mandate" email that Bezos wrote that essentially kicked off AWS. Have I been misunderstanding that email all along?
If you think the API mandate necessitated or created AWS then yes, you misunderstand that memo. The API memo simply mandated a service-oriented architecture. That can run on any compute and indeed predates AWS by several years.
The origins of AWS are entirely unrelated, Amazon simply realized it was good at operating infrastructure. This led to the conception of offering infrastructure as a service. The initial AWS service offerings were built based on Amazon's experience and expertise but were not the literal internal services of Amazon.
Amazon (or Google, or Microsoft) is an anomaly, the things Amazon needs to operate Amazon.com are very very different from what most companies need. The internal services and architecture are purpose built although today they do heavily leverage AWS under the hood. Amazon.com is just a customer of AWS, there's no special installation.
Almost nobody needs Amazon.com's infrastructure (except Amazon) or could even afford to staff it. What people do need are generalized, managed services they can combine for their own use cases.
I'm confused because on the one hand you say that their needs are very different and then on the other hand you say that they are using AWS. Another commenter also pointed out that they are using AWS.
I think the only needs that are likely different are scale, so bigger instances, and the fact that they have a massive legacy infrastructure that can't be easily migrated off of. So that's pretty much what I would expect; Amazon.com does use AWS for new stuff but is still operating a lot of legacy as well.
Amazon uses AWS for some new stuff. New AWS services high up in the stack like AI services run mostly on AWS. Most infra-AWS services run on foundational AWS services like EC2 and S3. Foundational AWS services (EC2, S3, DynamoDB, etc.) don't run on AWS for obvious chicken-and-egg reasons. For all of the things that are done by latecomer AWS services you can count on there being an unrelated internal tool, since clearly Amazon was doing those things before they launched those services.
> Guthrie said Microsoft has long had a tradition to "eat our own dog food," which means it runs its own systems on top of the same software it offers to customers. Microsoft IT was an early customer for Office 365, Azure, and Dynamics 365, and has moved the vast majority of its internal servers to Azure, including its SAP implementation, payment systems, and test systems. This has helped Microsoft learn what it takes to run the technology on an enterprise scale.
That's normal progression. Hardly anyone at Microsoft used Azure in the beginning either, it takes a few years for teams to move infra and deprecate old stuff.
Same for Amazon, it took them very long also, including dedicated internal projects with focus on moving. But majority of Amazon and Microsoft workloads now run on their respective cloud offerings.
Google is probably still be at least 3-5 years before that would happen be aligned. GCP is quite immature and young, which this leak highlights also. I assume this would mean that also internal teams at Google won't be moving workloads to GCP.
