Might seem that way because of what happened, but the main network is probably more likely to fail than the internal network. In those cases, running monitoring on a separate network is critical. EC2 control plane same story.

The entire value proposition for AWS is "migrate your internal network to us so it's more stable with less management." I buy that 100%, and I think you're wrong to assume their main network is more likely to fail than their internal one. They have every incentive to continuously improve it because it's not made just for one client.

What would you recommend instead? Control plane/data plane is a well known pattern.

It's 2006, you work for an 'online book store' that's experimenting with this cloud thing. Are you going to build a whole new network involving multi-million dollar networking appliances?

Developing EC2 did involve building a whole new network. It was a new service, built from the ground up to be a public product.

Yes, but many services, including S3, SQS, and DynamoDB did not run on the EC2 network for most of their existence.

The netscalers were on prod after all.

Amazon was famously frugal during that era.

No but one would hope 15 years and 1 trillion dollars later you would stop running it on the computer under your desk.

Hehe true! But it's real hard to "move fast and not break things" when you're talking about millions of servers, exabytes of data, and tens of thousands of engineers.