The basics

Modular data centers,
explained.

A modular data center moves most of the build into a factory and delivers capacity in repeatable blocks. This guide explains the types, the design, the trade-offs against traditional construction and what drives cost.

Discuss a project

What is a modular data center?

A modular data center is a data center facility assembled from prefabricated modules that are designed, built and tested off site, then shipped to the deployment site and connected. Each module carries part of the physical infrastructure: IT racks, power distribution, cooling equipment, fire protection or monitoring systems. Modules can be combined and added over time, so data center capacity grows in standard steps rather than through one large building project. For data center operators, that turns a new site into a series of smaller, repeatable projects.

The term covers a wide range of products, from a single rack-sized enclosure to multi-megawatt prefabricated buildings. What they share is a modular approach to delivery: standardized components, factory integration, and site work reduced to foundations, utility connections and commissioning.

In short, a modular data center does what any data center does—house IT equipment, supply clean power and remove heat—but it gets there through manufacturing instead of conventional on-site construction. The result is digital infrastructure that can follow demand for computing capacity more closely.

Types of modular data centers

Suppliers use overlapping names. Four broad types cover most modular data center solutions on the market.

Containerized data center

IT, power or cooling equipment is installed inside ISO shipping containers or container-sized enclosures. Shipping containers are easy to transport and stack, which suits remote locations and phased sites. For example, GRC describes its ICEtank as a 40-ft containerized immersion data center with eight 42U racks and up to 480 kW. See the containerized data center guide for detail.

Skid-mounted modules

Power modules and cooling modules—switchgear, UPS systems, transformers, chillers or pumps—are mounted on steel skids for installation outdoors or in a shell. Skids are common for the electrical and thermal plant even when the IT space is a conventional building.

Prefabricated modular building

A prefabricated modular data center can also be a complete building made of volumetric modules: structure, walls, roof and fitted-out white space. These modular facilities look like a conventional data center once assembled but were built largely in a factory. Vertiv, for example, describes OneCore as factory-assembled prefabricated infrastructure for 5 to 50 MW in a single block.

Micro and edge modules

Micro data centers put one rack or a few racks in a self-contained cabinet with integrated cooling and monitoring. They serve edge computing in branch sites, factories and telecom locations. Our edge data center guide covers this category.

Sources: GRC edge products; Vertiv, August 5, 2025. Reviewed November 2, 2026.

How a modular data center works

Modular data center design splits the facility into functional blocks that can be built, tested and replaced independently.

  • IT modules hold racks, containment, cabling and the network infrastructure that connects servers to each other and to carriers.
  • Power modules carry the critical electrical equipment. They take the utility power supply through transformers, switchgear, UPS systems and power distribution to the racks. Some designs add generators or battery storage for ride-through and backup.
  • Cooling modules remove heat from the IT load. Air-cooled designs use in-row units or air handlers; high-density designs use coolant distribution units, liquid loops and external heat rejection.
  • Monitoring and control systems track temperatures, flows, electrical loads, leaks and access, giving operations teams a single view of the critical components.

Because each block is defined in advance, the interfaces between them—electrical connections, pipe sizes, control signals—are standardized. That is what makes repeatable modular deployments possible. It also means the interfaces must be specified carefully, because a mismatch between a power module and an IT module is discovered on site, not in the factory.

Modular data center vs traditional data center construction

Traditional data center construction builds the structure and installs the mechanical and electrical systems in sequence at the site, using local general contractors and specialist trades. Modular construction moves most of that work into factories running in parallel with site preparation, so modular delivery changes the order of work as much as the method.

FactorModular buildTraditional construction
Where work happensMostly off site, in controlled environmentsMostly on site, exposed to weather and trade sequencing
ScheduleFactory and site work run in parallelLargely sequential; longer average development cycle
Capacity stepsAdd modules as demand growsLarge up-front shell, filled over time
Quality controlRepeat assembly and factory testingDepends on site supervision and commissioning
Design flexibilityConstrained by module sizes and transport limitsHighly customizable
Site needsFoundations, utilities, crane access for heavy equipmentFull building works
Best fitFast, phased or remote capacityLarge, long-lived campuses and hyperscale data centers

The comparison is not all-or-nothing. Many new data center projects combine a conventional building with prefabricated components such as skid-mounted electrical rooms or prefabricated modules for cooling.

Advantages of modular data centers

The benefits modular data centers offer come mainly from moving work into the factory. Compared with traditional data centers, the gains show up in deployment speed, predictability and the ability to add capacity in steps.

  • Faster deployment. Rapid deployment is the most cited reason to go modular. Factory integration overlaps with site preparation, which shortens deployment timelines. Vertiv, for example, claims up to 50% faster deployment for its MegaMod CoolChip than an onsite build; Schneider Electric reports cutting on-site installation per pod from 75.5 to 15 hours in its RapidScale portfolio. These are vendor figures, not independent benchmarks.
  • Better quality control. Repeated assembly in one place, with factory acceptance tests, can catch defects before shipment.
  • Phased capital. Deploying capacity in blocks lets an operator match spending to demand instead of building the whole facility at once.
  • Less on-site construction. Fewer trades on site can reduce congestion, safety exposure and dependence on local labor.
  • Repeatable design. Standardized components and identical modular units make multiple builds easier to plan, spare and maintain.
  • Site flexibility. Modules can go where available power exists, including industrial land, existing powered properties and remote locations near renewable energy sources.

Sources: Vertiv, July 25, 2024; Schneider Electric, October 21, 2026. Reviewed November 2, 2026.

Disadvantages of a modular approach

Modular delivery has real limits, and an honest plan should account for them.

  • The site is still the critical path. A factory can build a module quickly, but utility capacity, permits, foundations and network connections still take time. A module waiting for power earns nothing, so total deployment time is set by the slowest of the factory and the site.
  • Transport sets the envelope. Road and crane limits cap module size, which constrains layouts and maintenance clearances.
  • Less customization. A standard product may not fit unusual redundancy, security or architectural requirements without engineering changes that erode the schedule benefit.
  • Interface risk. When power, cooling and IT modules come from different suppliers, responsibility for the connections must be explicit.
  • Approvals vary. Inspectors may treat a module as equipment, a building or both. Factory certification does not remove local zoning and permit requirements.
  • Long-term cost. For a very large, stable campus, a conventional building can be competitive over its life cycle, and building life cycle assessment results depend on the design and materials.

Modular data center design for AI and GPU density

AI training and inference for large language models need far more computing power per rack than conventional IT, and have pushed rack power far beyond what most legacy rooms were designed for. At these densities, air alone struggles to remove heat, and thermal management becomes a primary design constraint. Most high-density modular designs now use direct-to-chip liquid cooling with a smaller residual air load; some use immersion.

Published examples show the range. HPE’s AI Mod POD offers up to 1.5 MW per module with a hybrid air and liquid cooling system. Armada describes its Leviathan as a liquid-first unit with 1.77 MW of IT capacity. Crusoe says its Spark Liquid supports up to 3.5 MW of compute per modular unit with closed-loop, non-evaporative direct liquid cooling.

For GPU modules, check the cooling system against the platform’s coolant temperature and flow requirements, the outdoor design conditions, and the thermal performance at peak ambient. Confirm how much power the cooling equipment and other data center equipment draw, because that sets the facility load you need for a given IT load. Our liquid-cooled GPU container guide covers this in more depth.

Sources: HPE AI Mod POD; Armada Leviathan; Crusoe Spark Liquid. Reviewed November 2, 2026.

What does a modular data center cost?

There is no reliable single price per megawatt, because quotes cover very different scopes. A bare enclosure, a fitted module with power and cooling, and a commissioned facility with GPU hardware are not comparable numbers.

The main cost drivers are:

  • IT load and rack density, which set the electrical and cooling capacity
  • cooling method and the climate at the deployment site
  • redundancy and resilience targets, including UPS systems, generators and battery storage
  • site works: utility connection, foundations, drainage and crane access
  • compute and network hardware, often the largest single line for AI projects
  • operations: energy use, maintenance, staffing and connectivity

Our modular data center cost guide breaks a budget into packages and includes an illustrative 1 MW energy calculation.

Who builds modular data centers?

Three groups supply most modular data center deployments. Electrical and thermal OEMs such as Vertiv, Schneider Electric and Eaton offer prefabricated power, cooling and white-space modules. IT vendors such as Dell and HPE package modules around their own servers. Specialist developers—Armada, Crusoe, Mod42, Flexnode and others—build modular AI data centers for their own clouds or for customers.

Compare suppliers on the scope they deliver, not the headline capacity. Our modular data center companies guide lists vendors by type, and the comparison pages for Vertiv, Dell and Armada go deeper.

Common questions about modular data centers

What is a prefabricated data center?

A prefab data center is one whose major systems are built and tested off site and assembled at the deployment site. “Prefabricated” and “modular” are often used interchangeably.

What are the main types of data centers?

Data centers are commonly grouped as enterprise (owned by one organization), colocation (shared space leased to tenants), hyperscale data centers (very large campuses run by cloud providers) and edge data centers (small sites close to users). A modular data center can serve any of these roles.

Are modular data centers future proof?

Partly. Adding or replacing modules is easier than rebuilding a hall, but power and cooling margins chosen today still limit tomorrow’s density. Size the site’s electrical and heat-rejection capacity for the next hardware generation, not only the first one.

Is a containerized data center the same as a modular data center?

A containerized data center is one type of modular data center. Not all modular designs use containers; many use skids or volumetric building modules.

Where QIT fits

QIT develops liquid-cooled modular GPU data centers planned around roughly 1 MW IT blocks, with phased 5–10 MW sites in Texas. We prefer existing powered properties, where usable power is already in place, and install each factory-integrated container in 15 hours once the site is prepared. Power, permits and site works still determine when that site is ready.

To go further, read about the GPU shipping container, the deployment plan and Texas site selection. If you have a site or a capacity requirement, contact QIT.

Prepared by QIT · Updated November 2, 2026

Explore all field notes

Let’s put your
power to work.

Start with your site, your workload and the compute you need.

Discuss a project
hello@qit.energy