Modular delivery as a structural advantage in AI capacity

Having led technical and pre-construction teams across more than 1 GW of hyperscale construction in Europe, the Middle East and Africa, I have found that delivery dates are rarely lost in design. They are lost on site: in sequential trades, in weather, in late equipment and in commissioning defects found when they are most expensive to fix.

Qore Modular was designed to remove those variables.

Sequential versus parallel

A conventional data centre is built in sequence on site: groundworks, structure, envelope, mechanical and electrical fit-out, integrated testing. The International Energy Agency puts typical construction at one to two years once power is available, which is realistic for a well-run programme with no surprises.

In Qore Modular, power, cooling and rack infrastructure are integrated in the factory while the plot, foundations, cable routes and grid connection are completed in parallel. On-site time falls by up to 60%. On confirmed power, the platform delivers new capacity in six to nine months.

Cost and quality move upstream

Two further effects matter as much as speed. Because modules are specified and ordered as complete products, supply chain costs are fixed at order rather than accumulating package by package. And because each module is tested in a controlled factory environment, a large share of commissioning defects are resolved before shipment.

Four configurations

Modular is often equated with containers. Containers are one of four Qore Modular formats:

• Container, for rapid deployment and phased expansion
• Internal skid, for integration within existing building envelopes
• External module, for greenfield campuses at scale
• Flatpack, for constrained indoor environments

The flexibility is strategic. Some of the fastest capacity in Europe sits in existing buildings and campuses that already have power.

Designed for density

NVIDIA's Vera Rubin NVL144 systems are expected at around 120 to 130 kW per rack, with Kyber-based systems pointing to around 600 kW. A factory product can be updated between production runs as hardware evolves, rather than being fixed in a building designed years earlier. Qore Modular is designed for direct-to-chip liquid cooling and heat rejection above 100 kW per rack, with heat recovery interfaces in every module.

Where it fits

Modular delivery is particularly suited to AI inference capacity, sovereign and public-sector deployments, enterprise and on-premise capacity, and the first phases of larger campuses that need to be live early. For a hyperscaler deploying its own reference design at campus scale, a powered shell may remain the right answer; the platform offers both.

In a market constrained by power, the building should never be the reason a customer waits. Modular delivery ensures it is not.

Sebastian Murphy

Director of Development and Projects, Qore Infrastructure Partners. Profile