NativeKloud · Sovereign EU AI, Compute and Storage

01 · PLATFORM

Three layers. One jurisdiction.

Generative AI needs somewhere to run and somewhere to keep its data. Today both come from American providers, which puts every regulated European workload one CLOUD Act request away from a problem. We are building the European alternative, layer by layer.

/ AI

MVP build

Inference

NativeKloud Inference

A drop-in, OpenAI-compatible inference API running only on European bare metal. Change one line — the base URL — and your prompts stop leaving the EU.

Open-weight models: Mistral, Llama, Qwen
Zero prompt and completion retention
Transparent per-million-token pricing
Works with the official OpenAI SDKs, LangChain, LlamaIndex

Request early access

/ COMPUTE

In build

Compute

Dedicated GPU & bare metal

Dedicated GPU capacity in European data centres, operated by us. No hyperscaler underneath, so no foreign jurisdiction reaching through the floor.

Bare-metal GPU nodes in Germany and Finland
Poland zone via Tier III/IV partner data centres
Dedicated instances for your own model weights
Per-request residency pinning, down to the country

Discuss capacity

/ STORAGE

Roadmap

Storage

Data that stays yours

The layer everyone forgets. Sovereign inference is worthless if the vectors, documents and logs behind it sit in an American bucket.

S3-compatible object storage on EU soil
Open formats end to end: Parquet, Iceberg, Postgres
Vector and RAG storage next to the GPUs that read it
Exportable by design — your data leaves when you do

Become a design partner

Migration, in full

  const client = new OpenAI({
-   baseURL: "https://api.openai.com/v1",
+   baseURL: "https://api.nativekloud.eu/v1",
    apiKey: process.env.NATIVEKLOUD_API_KEY,
  });

02 · SOVEREIGNTY

Sovereign by construction, not by contract.

Most "EU AI" offerings are an American platform with a European region selected and a data processing agreement stapled on. Ownership beats geography: if the provider is US-controlled, the CLOUD Act reaches the data wherever the disks physically spin. So we removed the American layer entirely.

/ 001

No US vendor, anywhere

Not just in the request path. No US cloud, CDN, DNS, analytics, error tracking, fonts, identity provider or payment processor. Every dependency is checked for who owns it, not only where it is hosted.

/ 002

Zero retention, stated precisely

Prompts and completions are never written to disk, logs or third parties. We keep token counts, timestamps and error codes so we can bill you and stop abuse. "We keep nothing" is not credible; this is.

/ 003

Open weights, open formats

Mistral, Llama, Qwen. Parquet, Iceberg, Postgres. No proprietary model API behind the curtain, no format that holds your data hostage on the way out.

/ 004

Audit-ready from day one

Signed DPA, published subprocessor list, documented technical measures. The artefacts a CISO needs to sign off, ready before you ask for them.

/ 005

Compatible, so switching is cheap

The API is wire-compatible with the OpenAI SDKs. Migration is one line, and so is migrating away. We would rather earn the renewal than trap it.

/ 006

Priced in public

Per-million-token pricing for every model, on a page, in euros. No "contact us" tier, no negotiated opacity.

03 · COMPANY

What is NativeKloud.

NativeKloud is a Sovereign EU AI & IT company, registered and operating in Poland. Not an agency, not a reseller. We design, build and run the infrastructure we sell.

The thesis: Central and Eastern Europe has the largest untapped AI market on the continent and one of the strongest engineering talent pools in the world — and it runs almost entirely on American infrastructure. Adoption here sits at 4–6% against 13.5% in Western Europe. The blocker is not demand. It is compliance and infrastructure complexity.

Our answer is boring on purpose: European hardware, open-weight models, open data formats, an API developers already know, and a compliance pack that lets the person signing it off actually sign it off.

COMPUTE

European bare metal Dedicated GPU nodes in DE and FI today; Polish Tier III/IV partner capacity next. No hyperscaler in the substrate.
MODELS

Open weights only Mistral, Llama, Qwen served on vLLM. No proxying to a proprietary US model API — that would break the guarantee.
DATA

Portable and sovereign Iceberg, Parquet, Postgres, S3-compatible object storage. Your data leaves with you, by design.
LAW

GDPR and EU AI Act aligned Purpose, retention and subprocessor documented for every piece of data we touch. EU Cloud Code of Conduct on the roadmap.

04 · ABOUT

Hi, I'm Radek.

Over two decades building cloud infrastructure, services and teams. Started in the UK in 2005 and spent over a decade in London helping grow Databarracks from a backup vendor into a managed cloud provider.

Today I'm Head of Cloud Technology at Databarracks, leading our AWS, Azure and GCP practice for UK public sector and enterprise clients — which is exactly how I know what a regulated buyer needs before they can say yes. Along the way I co-founded Kazoup, unstructured-data discovery for enterprise storage, demoed at Cloud Expo Europe 2015.

Moved back to Poland in 2018 and started NativeKloud. Same job I've always done: storage, then cloud, now sovereign AI infrastructure. The layer changes. The problem — someone else holding your data under someone else's law — does not.

Radek Dymacz — Radosław Dymacz · Head of Cloud Technology, Databarracks · Founder, NativeKloud

2005

Arrived in the UK Linux, Unix, open source. Storage, virtualisation, the unglamorous plumbing of the early cloud.
2010s

Cloud era, London Databarracks: from Linux specialist to Head of R&D. Co-founded Kazoup. AWS, object storage, cloud security.
2018

Back to Poland NativeKloud founded. Independent cloud and data architecture for UK and EU clients.
TODAY

Building Europe's sovereign AI layer NativeKloud Inference — inference, compute and storage that never leave European jurisdiction.

European AI.
European iron.
European law.