Free credits are the ingenious marketing ploy of all big clouds to delay price comparisons until some amount of lock-in is achieved. Sure, you could likely host it for 1/10th to 1/5th of the cost elsewhere. But then you would have to pay from the get-go, and on AWS it's free™ (until it isn't)
I have a hard time understanding where the cost is coming from. I can only assume it's because of the DB as opposed to the instance or S3.
I wouldn't be surprised if it was over 10x cheaper if you didn't use cloud hosting.
You can get it down to $12/month by going with a t4g.micro (2 vCPU, 1GB RAM), single-az, no proxy, no backups, 5GB storage; with price after that scaling roughly linearly with RAM. But the interface will scream at you the entire time that this isn't reasonable for a production setup.
It is really no different than managing VPS's, in all honesty (just less expensive and tangible). I spend less than one hour a week doing updates and routine maintenance. Drive failures happen rarely, and can usually be swapped out within a few hours (this is what RAID is for).
More time is spent dealing with k8s and it's shenanigans than dealing with the actual servers.
Many people don't do that. My company uses EKS and we just send dockerfiles and yaml files at it. We don't ever actually interface with an ssh prompt on our servers
I think Proxy _is_ overkill and perhaps expensive for what it is. For a pg install, running pgBouncer isn’t difficult. But I guess that depends on whether you’re authenticating with IAMs, which doesn’t sound trivial with bouncer (it might be).
I am not sure I know the reason behind it honestly.
Either you go with something like Dynamodb or planetscale or the least you pay is 40€/m using AWS Aurora serverless.
As soon as databases start to charge for computing time the cost balloon out of reach of hobby projects.