Worked example · AWS · modelled account
A scale-up on Kubernetes at $25k a month: EKS, Postgres, Kafka, OpenSearch and heavy egress.
A modelled account, not a file: the service mix of a Series-A company on EKS with several Postgres instances, Kafka, OpenSearch and tens of terabytes of egress. The size at which the engineer time stops deciding the answer. Read through the way the calculator reads it: the bill, what it becomes on Hetzner, the verdict, and what would change it.
The bill
$24,950 a month across 14 services after setting aside nothing. The six largest lines:
| Service | Grade | Today | On Hetzner | Change |
|---|---|---|---|---|
| EC2 | Easy | $11,200 | $2,431 | saves $8,769 |
| RDS | Moderate | $4,600 | $1,017 | saves $3,583 |
| S3 | Easy | $2,400 | $606 | saves $1,794 |
| MSK | Moderate | $1,500 | ~$450 | saves $1,050 |
| OpenSearch Service | Moderate | $1,300 | ~$455 | saves $845 |
| CloudFront | Moderate | $1,100 | ~$165 | saves $935 |
~ marks a rule-of-thumb ratio; the rest is priced from usage in the file.
What it becomes on Hetzner
- 34 servers: 11× CX53, 5× CX53, 4× AX42-1, 9× CX33, 2× AX42-1, 3× CX33
- $6,011/mo for servers, storage and traffic
- 75 h/mo of engineer time, $7,125
- $32,280 one-off, paid back in 3 months
Net saving per year
$141,771
Worth moving. 4 of your 14 services move easily. One-off migration of about $32,280 pays back in 3 months.
Net of 75 hours a month of engineer time. 55% of the Hetzner figure is priced from your usage; the rest is ratio-priced services and engineer time, which is where the range comes from.
Likely range $2,594 to $17,201 less a month. Hetzner list prices of 17 Oct 2026, default assumptions.
The same file on nine destinations
Net per year after engineer time, best first. Hetzner is the best of the nine for this account.
| Hetzner | $141,771 kept |
| OVHcloud | $96,814 kept |
| Hetzner (US) | $87,503 kept |
| Scaleway | $80,917 kept |
| Vultr | $69,000 kept |
| Akamai (Linode) | $60,912 kept |
| DigitalOcean | $53,109 kept |
| UpCloud | $43,110 kept |
| Exoscale | $4,749 kept |
What would change the answer
- Engineer time. 75 hours a month at $75/h is $7,125, 54% of the destination figure. Halve the rate or fold the work into an existing team and the verdict moves first.
- Scale. Engineer time grows slowly with the number of servers; hardware grows with the bill. The same mix at three times the size usually flips a marginal answer.
- Moving only part of it. The first wave (foundation: storage, traffic, edge) is S3, Elastic Load Balancing, Route 53: $2,075/mo for about 1 engineer days, with the databases left where they are.
Every one of these is a setting in the report. Open the example and change them.