Hetzner
Dedicated GPU servers from Hetzner. Preview: adapter built, not wired into apply.
- Status
- Preview
- Kind
- European cloud
- Provider id
hetzner- Regions and residency
- Germany, Finland, plus US and APAC
Preview: not wired into apply
The provisioning adapter exists and is tested against mocks. It has not been tested against the live service, and amicu apply cannot deploy to it yet.
Use this provider
resource "hardware_profile" "hetzner_gpu" {
provider = "hetzner"
region = "fsn1"
accelerator {
kind = "gpu"
model = "rtx-4000-ada"
count = 1
memory = "20GiB"
}
memory = "64GiB"
}Running amicu apply against this profile fails before anything starts. To check it today, run amicu validate and amicu plan.
Overview
Hetzner is a German provider. Its GPU hardware is dedicated bare metal (the GEX line), not a GPU cloud. The adapter creates servers through the hcloud REST API, drives them over SSH, and prices them monthly in EUR with no spot market.
Hetzner GPU capacity is sold in the EU data centers only. By default the adapter admits only EU data centers and rejects any other region before a backend call. Widening to other jurisdictions is an explicit choice in code; it is not exposed in HCL.
Setup and credentials
The adapter authenticates to the hcloud API with an API token sent as a bearer token. apply does not run this provider yet, so there is nothing to configure today. Credentials will be secret references only; never write a token into HCL (see Secrets).
Configuration reference
| Argument | Type | Default | Required | Notes |
|---|---|---|---|---|
provider | string | none | yes | "hetzner". |
region | string | none | no | One of the region codes below. Used for residency checks and cost estimates. |
accelerator.kind | string | none | yes (with accelerator) | gpu. |
accelerator.model | string | none | yes (with accelerator) | A GPU model from the table below. |
accelerator.count | number | none | yes (with accelerator) | Number of accelerators. |
accelerator.memory | size | none | yes (with accelerator) | VRAM, for the fit check. |
memory | size | none | yes | Host RAM. |
price.per_hour | money block | none | no | Your own price, for cost planning. |
These are the generic hardware_profile fields. There are no hetzner-specific arguments.
GPUs and regions
| Region code | Location | Jurisdiction |
|---|---|---|
nbg1 | Nuremberg, DE | eu |
fsn1 | Falkenstein, DE | eu |
hel1 | Helsinki, FI | eu |
ash | Ashburn, Virginia | us |
hil | Hillsboro, Oregon | us |
sin | Singapore | apac |
model | VRAM |
|---|---|
rtx-4000-ada | 20 GiB |
rtx-6000-ada | 48 GiB |
Limits
- Not wired into
apply. - The catalog is a representative slice of the GEX line, one GPU per server.
- Dedicated servers bill monthly, with no spot or hourly option.
- Prices in the catalog are illustrative values, not live quotes. amicu does not sync live capacity or pricing yet, so cost estimates are indicative.
Troubleshooting
Source
Adapter: crates/amicu-providers/src/hetzner.rs. Status of every provider: Providers.