Trust Centre · Data residency
Data residency & sovereignty
What runs where, on whose hardware, and the commitment that customer traffic never leaves the EEA.
Mapping
Service to region
| Service | Current European processing |
|---|---|
| Serverless API | Stockholm, Sweden (eu-se-1) |
| Dedicated GPU rental | Málaga, Spain (eu-es-1) |
| Platform/control services | EEA infrastructure, according to current architecture |
Placement
What runs where
| Layer | Where it runs |
|---|---|
| Serverless inference | NVIDIA DGX Spark systems AxForge owns and operates in eu-se-1 · Stockholm. |
| Dedicated GPU rental | NVIDIA DGX Spark (GB10) systems AxForge owns and operates in eu-es-1 · Málaga. One dedicated machine per customer for the rental period; the machine and all data on it are deleted when the rental ends. |
| Facilities | Controlled-access facilities in Sweden and Spain. We publish no further physical detail; detailed physical and technical measures live in a private TOMs annex, available under NDA/DPA. |
| EU edge | Hetzner Online GmbH (Germany): TLS termination, reverse proxy and this website. TLS 1.3. |
| Edge → compute | Encrypted private WireGuard tunnels between the EEA edge and each compute site. |
| Account, billing & log data | Held and processed in the EEA. Card payments are processed by Stripe Payments Europe, Limited (Ireland). See the subprocessor register. Retention limits per the retention schedule. |
Ownership
Owned hardware, not rented compute
AxForge inference and dedicated GPU rentals run on machines AxForge bought, owns and operates, not resold hyperscaler capacity, and no third-party compute for inference. Serving is transparent: named models on named machines (NVIDIA DGX Spark, GB10), so you know exactly what runs where. The model catalogue lists what is served; the subprocessor register lists who else touches data: the edge layer, console sign-in identity and payment processing.
Boundary
No processing outside the EEA
There is no outside-EEA processing and no outside-EEA failover, not as a fallback, not under load, not during an outage. The commitment, in full: If all EEA capacity is unavailable, requests queue or fail. They are never routed outside the EEA.
Data flow
The path of a request
Customer → EEA edge · Hetzner · TLS 1.3 → WireGuard private tunnel → AxForge GPU · Stockholm or Málaga → Response
| Step | What happens |
|---|---|
| Serverless request path | Customer → EEA edge (Hetzner, Germany, TLS 1.3) → encrypted private WireGuard tunnel → AxForge GPU (Stockholm) → response. |
| Dedicated GPU rental | Customer → SSH to the dedicated machine in Málaga. Data on the machine is under the customer's control for the rental period and is deleted, with the machine, when the rental ends. |
| What is recorded | Metadata only (token counts, timestamps, status), kept under the retention table. Prompt and completion bodies stay in memory and are discarded when the request completes. |
Verification
Verify it yourself
Serverless requests are served from eu-se-1 · Stockholm; dedicated GPU rentals run in eu-es-1 · Málaga. The served model identity is returned on every API response, and the region and machine of a dedicated rental are shown in the console. Request and response shapes are in the regions & data handling docs; for a residency review, Talk to an engineer.
Related
Nearby in the Trust Centre
Version 1.0 · Effective 2026-09-04 · Last reviewed 2026-09-04
- 2026-09-04, version 1.0: service-to-region mapping (serverless Stockholm, dedicated GPU Málaga, platform services EEA), dedicated-rental data flow, verification section updated.
- 2026-08-26: first published version of this page.