Where your AI cluster actually lives.
AI hardware asks things of a building that most IT environments were never designed to give. Power density five to ten times what your racks draw today. Cooling technology your team has probably never had to buy before. Floor loads, service access, fibre pathways, grid headroom.
We run European-owned datacenters with capacity available now, and an engineering team that designs the power and cooling around your specific hardware. So the building never becomes the bottleneck.
One facility, four stages of density.
Most enterprise AI follows the same path: start small, prove the value, scale. The
infrastructure requirement changes at every step — and the point of getting the facility right
the first time is that you never have to move.
Tier 1 • Validating
Your first GPU servers
For teams testing use cases before committing to a facility project.
- Up to 80 kW/rack
- High-spec air cooling, with a liquid upgrade path
- Dual power feeds
Tier 2 • First cluster
Moving from pilot to production
For teams with workloads that now run continuously.
- 80–100 kW/rack
- Liquid-to-air direct liquid cooling
- High-capacity power distribution
Tier 3 • At scale
Production AI, contained rows
For businesses where AI has become core to the product.
- 100–150 kW/rack
- Liquid-to-liquid DLC, N+1 across power and cooling
- 8–18 racks, expandable in place
Tier 4 • Headroom
What's coming next
For national programmes, research institutions, and roadmaps that run past the current hardware generation.
- 150–300+ kW/rack
- Advanced direct-to-chip cooling
- Modular power substations
Is your infrastructure ready for AI?
This is not a guide to building an AI gigafactory. It's about what happens when a normal organisation needs to run serious AI — next to everything else.
Power, cooling, floor loads, connectivity, compliance, and the configuration questions our engineering team works through with every deployment. Written for the person who has to make it physically work.
The AI Infrastructure Guide will be available for download soon.
- 100% European-owned. Not European data residency — European ownership, European jurisdiction, European law. The question of where your training data lives has a straightforward answer here.
- Capacity available today. Grid connections are the scarcest resource in digital infrastructure, and no budget can fast-track them. Ours are already contracted, in facilities that are already running.
- Engineering, not a portal. Every deployment above roughly 80 kW is designed around your hardware, with our engineers in the room. Cooling configuration, power distribution, demarcation, deployment logistics — worked through before anything is racked, not during the incident.
- Built for both. Your AI cluster will sit alongside your conventional IT. Our facilities are designed to run air-cooled and liquid-cooled environments in the same building, at a design PUE of 1.2.
- Carrier-neutral. Your networks are your choice — not ours. Redundant meet-me-rooms, diverse fibre paths, and the pathway capacity that dense interconnect needs.
European-owned. Modern facilities. Capacity now.
Sofia, the Flemish national supercomputer.
Your first AI deployment will almost certainly be smaller than Sofia's 13,580 TFLOPS. That's what makes it useful.
Sofia runs in our BRU01 facility in Brussels on a dedicated 1,000 kW feed, with 5,500 kW of expansion available in the same building. Its GPU nodes required direct liquid cooling for both CPUs and GPUs — a configuration our engineering team designed with the VUB before deployment, and then rolled out across the rest of the facility so that it's available to the next customer who needs it.
"GPUs represent a significant investment, so it is important to get the power and cooling configuration — as well as the managerial roles and responsibilities — right from the outset."
– Harmen Laan | Head of Engineering, Penta Infra
The Flemish Supercomputer Case Study will be available for download soon.
Let's look at your hardware.
Bring us the configuration you're planning — the OEM, the density, the cooling requirement, the timeline. We'll tell you what it needs and where it fits.
