You can't really host your own AWS, You can self-host various amalgamations of services that imitate some of the features of AWS, but you can't really self-host your own AWS by any stretch of the imagination.

And if you're thinking with something like localstack, that's not what it's for, and it has huge gaps that make it unfit for live deployment (It is after all meant for test and local environments)

There's lots of costs to AWS. They'll rope you in with the whole free tier for compute and storage. But AWS has charges for outbound traffic, detailed logging, elastic IPs, etc.

It's a whole job to just do cost analysis for cloud services.

Oh, it's expected costs.
Like, figure out the compute requirements of your code, multiply by the cost per compute unit (or whatever): boom, your cost.
Totally predictable.
Compared to suddenly having to replace a $20k server that dies in your data center.
So much easier.

Except when your code (let's be honest, the most likely thing to have an error in it... At least compared to some 4+ year old production hardware that everyone runs) has a bug in it that requires 20x compute.
But maybe that is a popularity spike (the hug-of-death)! That's why you migrated to the #cloud anyway, right? To handle these spikes! And you've always paid your bills so... Yeh, here's a 20x bill.

Who told you that?

I enabled Cloudtrail to log all DynamoDB read/write data events when trying to troubleshoot an issue. Even though I only left this enabled for a few days, the Cloudtrail line item was $5k more than it should have been. My back of the napkin math with assumptions came out to be 100 times less than that, so I had a really awkward support email asking them to reverse the charges, which they did fortunately.

If you spin it up, fucking own it. When you're done with it, shut it down. I have long lost count of the number of times I've reached out to a team to ask about the coin miner they are running on some random EC2 instance only to find out that some jackass spun it up for a test, gave it a public IP, set the VPC to allow any inbound traffic, installed all kinds of random crap and then never updated it. Nor did it get shutdown when the test ended. So, a year and a half later, when the software was woefully out of date, someone hacked it and spun up a coin miner. Oh, and the jackass who set it up didn't bother to enable logging or security monitoring. But, they sure as hell needed the ability to spin stuff up on their own. Because working with IT to get it done right would be too hard for their fragile little ego.

Can someone explain how you accidentally rack up such a bill?

For example: You can deploy your Python script as a Lambda. Imagine somewhere in the Python script you'd call your own lambda - twice. You basically turned your lambda into a Fork Bomb that will spawn infinite lambdas

You're gonna be fine. Honestly, if your team has given you permissions to do something accidental like this then it's on them. You're not gonna get stuck with the bill. You're not gonna get fired. It wouldn't be your fault.

It's really only scary when you're doing it solo with your own back information lol.

It's actually difficult. You can set a budget so AWS will alert you when you hit it.

They have a price calculator that'll calculate costs before you do anything.

Inbound Internet is free. Only outboard costs you anything.

Network transfers between VPC's using private links are free.

AWS accounts are free.

Lambda functions are ultra cheap as you only pay for the time you use it.

S3 is object storage with 11 9's of uptime and it's cheaper than any enterprise NAS.

Basically you'd have to setup something wrong and ignore it for weeks to incur a large bill.

Just make sure whatever resources you spin up you're spitting down. This stuff though tends to happen when people accidentally let a a script that creates and destroys instances run over the weekend and it didn't appropriately clean up instances for you...

Or you thought you would try your hand at training in llm and then realized you spent way too much money on the infrastructure and resources

AWS has a multitude of different offerings with confusing pricing structures. They have zero incentive to make them understandable.

That said, chances are your new company has people who understand this already and know how to manage it. Hopefully, they'll put up some guardrails that prevent you and others from running up a big bill. I wouldn't expect a junior programmer to know how to do this, but that's ok as long as the company is managed right. Granted, that can be a big if sometimes.

Wow that's.... Pricey. What is your kWh cost?

Years ago, I played with AWS then contacted their support to make sure any AWS billing to my account was disabled.
I thought I'd try it again recently, and couldn't log in.
I still don't think I'm missing anything.

I'd rather have VPS or server providers where I know exactly what I'm getting per month no matter what, tho I've ran near data transfer surcharges.

First thing I learned in a video course was setting a spending budget for your account. I set a $10/month limit and never broke it in 6 years.