The "Hetzner is 5-10x cheaper" framing is real but the comparison that actually matters is egress, not compute. A $5 DO droplet and a €4 Hetzner CX22 are both fine for a side project. What kills you on AWS/GCP at small scale is the per-GB egress - once I was pulling ~300GB/mo out of an eu-west-1 ELB and the egress line was 6x the EC2 line on the bill. Moving the same workload to Hetzner or even a single Fly.io region dropped it by >80% without touching the app.
One thing people skip: you also lose AWS's managed failure modes. Managed Postgres, VPC peering, IAM roles on instance metadata - you now own all of that. For a solo project or small team that's fine (rsync + pg_dump | pg_restore has worked for 20 years). For anything with on-call rotations it's a meaningful shift.