EU infrastructure · owned hardware
A European sovereign AI cloud in the operational sense: AxForge owns its inference hardware, runs it in EU regions, terminates TLS in the EU, and retains zero prompts. Every claim on this page is checkable.
Definition
"Sovereign" is doing heavy marketing work in AI right now. Stripped of the flag graphics, it reduces to a short list of operational questions: who holds title to the machines, where does inference run, where do TLS keys live, what is written to disk, and who else touches the traffic. Here is how AxForge answers each one.
| Sovereignty property | How AxForge implements it |
|---|---|
| Who owns the hardware | NVIDIA DGX Spark (GB10) systems bought, owned and operated by AxForge — not resold hyperscaler capacity. Running in Sweden (eu-se-1), with more systems on the way. |
| Where inference runs | eu-se-1 · Stockholm and eu-es-1 · Málaga — live. More EU regions — in deployment. |
| Where TLS terminates | TLS 1.3, terminated in the EU. |
| How requests are pinned | x-axforge-region header, pinned on the API key, echoed on the response — you can verify the region on every call. |
| What is retained | Zero prompt retention: prompts and completions are processed in memory in Sweden, not written to disk, not logged, never used to train. Only metadata (token counts, timestamps, status) is kept, for billing and operations. |
| What is disclosed | Transparent serving: named models (e.g. qwen3.8-27b-nvfp4) on named machines (DGX Spark, GB10) — you know exactly what runs where. |
| Who else touches traffic | EU-hosted infrastructure; details in the trust centre. Listed in the privacy policy. |
Substance check
The marketing version of sovereignty is an EU flag on a region page. The operational version is a set of questions any provider should be able to answer in writing: Do you own the machines, or rent someone else's? Who holds the TLS keys? What exactly is written to disk? Name your subprocessors. We publish our answers above — and we encourage you to put the same questions to us and to anyone else you evaluate.
Sovereignty here is a set of checkable properties, not a slogan: EU processing on hardware we own, zero prompt retention, a public subprocessor register, and a trust centre that documents all of it. In short: GDPR compliant — with the depth one click away on the GDPR compliance page.
Dedicated systems
Beyond the shared API, AxForge rents dedicated DGX Spark systems monthly — one machine, one tenant, your traffic only. Run the catalogue models, or bring your own open weights (e.g. DeepSeek, Llama, GLM) on a dedicated machine. Talk to an engineer to scope a deployment.
Data & privacy
Prompts and completions are processed in memory in Sweden — not written to disk, not logged, not retained, and never used to train anything. We keep only request metadata (token counts, timestamps, status) for billing and operations. The full policy is at axforge.ai/privacy.
FAQ
Operationally: who holds title to the machines, where inference runs, where TLS terminates, what is written to disk, and which subprocessors touch traffic. AxForge answers each with a named, verifiable fact — see the table above — rather than a label.
AxForge. Inference runs on NVIDIA DGX Spark (GB10) systems we own and
operate in Sweden (eu-se-1). Serving is transparent: the model
name and the machine class are published.
eu-se-1 (Stockholm) and eu-es-1 (Málaga) are live.
More EU regions are in deployment. Requests are pinned to a region on the key
with the x-axforge-region header, echoed on the response.
EU-hosted infrastructure; details in the trust centre. Inference itself runs on hardware AxForge owns. Details in the privacy policy.
Yes. AxForge is GDPR compliant — EU processing, zero prompt retention, no training on customer data. The full picture (subprocessors, retention, DPA) is in the trust centre.
Yes — bring your own open weights (e.g. DeepSeek, Llama, GLM) on a dedicated machine, rented monthly and serving only your traffic. Talk to an engineer.
Related