What Is a Modular Data Centre? Three Definitions, Three Very Different Buildings

"Modular data centre" (often written "modular data center") covers a number of different approaches: factory-built components inside a conventional building, a complete system in one or more shipping containers, and custom steel-frame modules engineered to a client's specification and scaled in increments. The industry has not agreed a single definition, and which one a supplier means has a direct effect on programme, compliance and how much density a facility can absorb.

Video coming soon

Why is there no agreed definition?

Adaptive MDC's white paper, Modular Data Centre Design: A Philosophy, Not a Delivery Method, addresses this early on: there is significant variation in how the industry uses the term "modular", and it has yet to agree on a single set of definitions. A client comparing two "modular" proposals may be comparing two entirely different approaches without realising it.

What are the three types of modular data centre?

We use three definitions.

Why does the distinction matter?

The three types behave differently under pressure, and the pressures on the sector are specific. A conventional hyperscale-class build typically takes 24 to 36 months from design commencement to practical completion, with critical-path items such as generators and switchgear exceeding 52 weeks. Against that, a single AI deployment can take a designed rack density from 8–12 kW to 80–130 kW.

Containerised designs put IT racks, power distribution and cooling plant into one fixed volume. At low density that competition doesn't matter. As density rises, cooling and electrical distribution grow with the load while the envelope does not, which sets a ceiling on achievable density that no internal rearrangement can fix.

Modular components bring programme benefits for the plant they cover, but the facility around them is still designed once, for one site, at one point in time.

Modular data centres proper add a compliance requirement that a conventional build does not face. A conventional facility's redundancy is designed once, for its final capacity. A modular facility must remain compliant whether it comprises its minimum viable configuration or its maximum planned extent, so N+1, 2N or an equivalent redundancy topology must hold true at every intermediate step of expansion, not solely at final capacity.

How does this compare with other definitions?

Our three types correspond closely to the two-way classification of prefabricated modular (PFM) facilities, containerised and modular, used by Uptime Institute, which also notes that some modular designs are optimised for transport as "flat-pack" data centres. UL Solutions takes a narrower view, describing a modular data centre as a prefabricated structure designed for faster deployment, scalability and efficiency.

Arcadis argues that modularity is better understood as a spectrum than a set of categories. It runs from standardised design components through pod- or module-based white space, kit-of-parts platforms, containerised or edge units and prefabricated MEP skids, to hybrid site-built and modular combinations.

The two views sit comfortably together. A spectrum is a useful way to map the market; clear categories are a useful way to specify. Uptime Institute's classification also separates containerised from modular facilities, which is the line between our second and third definitions.

What should a specifier ask?

When a proposal describes itself as modular, three questions tell you which category you are looking at:

• •Is the modularity part of the design from the outset, or a way of packaging a conventional design for delivery?
• •What is fixed (an envelope, a building shell) and what is free to vary as density and capacity change?
• •Does redundancy and compliance hold at the minimum configuration, at the maximum, and at every step between?