We build the inference layer Europe needs.

ARK Labs is a sovereign AI infrastructure company. We build inference software that runs frontier open-source models on any GPU or CPU an enterprise can put to work — owned, leased, or repurposed — in their data center, their region, under their own rules. Our customers are the banks, insurers, telcos, hospitals, and government agencies that can’t send their data to someone else’s cloud.

Founded
Szczecin, Poland
Team
5 · growing to 10
Stage
Commercial anchor live
Focus
Regulated enterprises, EU

Core principles

1
The model is not a monolith.
2
The cluster is fluid.
3
GPUs are shared resources, not fixed containers.

Inference is becoming infrastructure. Infrastructure requires control.

Our mission and vision are short on purpose. They describe the world we are building toward, and the work we do every day to get there.

M

Our mission

Enable any enterprise to run secure, high-performance inference on their own terms.

ARK Labs enables enterprises and AI builders to run secure, profitable, high-performance inference across any combination of cloud, on-prem, edge, or decentralised hardware — while maintaining privacy, resilience, and full control.

V

Our vision

Run AI inference on any hardware, anywhere — without compromising security or performance.

We’re building the inference substrate of the distributed AI era: a world where any organisation can run frontier models on any hardware they own, with the same performance and control as the biggest cloud providers.

A small team, pointed at a large problem.

ARK is led by two founders with complementary backgrounds in enterprise go-to-market and inference systems, supported by a team of engineers building the runtime day by day.

AH
CEO · Co-Founder

Andreas Hestbeck

Runs commercial strategy, enterprise partnerships, and fundraising. Leads ARK’s go-to-market across Western Europe and drives the company’s public narrative.

LZ
CTO · Co-Founder

Lukasz Zbrzeźny

Leads technical vision, runtime architecture, and engineering. Owns the performance, reliability, and security posture of the ARK platform across all three deployment tiers.

AN
Principal Engineer

Andre Nowak

Key technical contributor on the ARK runtime. Focused on multi-GPU scaling, heterogeneous hardware support, and the Supervisor Node orchestration layer.

··
Engineering

Platform & Runtime Team

The engineers behind the API overhaul, Mistral text component, and multi-GPU scaling work. Based across Poland and Western Europe.

+
We’re hiring

Hardware Specialist

Senior infrastructure engineer to lead the scale-up of our data centre and manage upgraded GPU fleet. Above-market comp, Warsaw or EU remote.

+
We’re hiring

Executive Account Manager

Technical seller with enterprise relationship skills. Owns 10–20 regulated-sector accounts across DACH and the Nordics.

We’re growing from 5 to around 10 people across engineering, GTM, and infrastructure — always technical, always senior.

See open roles

What we’ve been building, saying, and shipping.

Company updates, technical milestones, and coverage from our partners and the wider European AI infrastructure community. The full archive lives in the ARK Newsroom.

Inside ARK Labs.

A look at the team, the hardware, and the events where we meet the enterprises we build for.

Backed by operators who’ve built infrastructure at scale.

Building the sovereign inference layer for Europe.

ARK is supported by early investors, strategic partners, and a distribution alliance with Asseco that connects us to the regulated and public-sector client base across Central and Eastern Europe.

If you invest in European deep tech, sovereign AI, or enterprise infrastructure, we’d like to talk.

EU
Primary focus
Asseco
Commercial anchor
Talk to us about investing

Let’s talk about your inference stack.

Whether you’re evaluating a sovereign inference platform, exploring a partnership, or reporting a security issue — we read every message and reply within one business day.

Send us a message

We only use your details to reply. No mailing list, no third-party sharing.