We have a fully automated stack based on autoscaling groups in AWS. New nodes takes 2 minutes to be fully operational AND synced with full DB.
We started using t3a instances too but I noticed that we would chew the cpu credit and it would require cpu credit buy back (which is automatic on t3 btw).
So we migrated to more standard instances as mentioned @BugBounty3 , more m5a instances but same spirit.
In the meantime we worked on compiling cleanly cardano-node and cardano-cli on aarch64 architecture (ARM). Got a nice success so we moved to the arm based instances.
AWS is great if you use it smartly with all the tools (automation, autoscaling, mutliple AZ/Region, etc …). You can then leverage all the benefits and keep the price under control for a great result and awesome resilience. If you are just login in the console to click and get an EC2 instance, you can get cheaper elsewhere, but probably not as reliable.
We are linux/cloud profesionnals so this is our bread and butter, I understand it might not be easy for everyone. I’ll start writting article about all this and distributing the ARM compilation we had.
@Alexd1985 Is 1.26.1 lastest stable yet ? I though it was not flagged as such.