Modular and prefabricated data center deployment field guide

What a modular data center is, and where the work moves

A modular or prefabricated data center is a facility built from factory-assembled modules, the power, the cooling, and the IT space, fabricated and wired in a plant, shipped to site, and connected together instead of being built piece by piece in the field. The module arrives as a finished, tested assembly. The field work shrinks to the pad, the utility stub-ups, the set, and the connections between modules.

Two words get used loosely, so pin them down. Prefabricated means the parts were built off site in a controlled plant. Modular means the design is cut into repeatable blocks that grow by adding more of them. Most real projects are both at once: prefabricated modules deployed on a modular plan, so the same block design lands on the second site the way it landed on the first.

What changes versus a stick-built hall is where the labor and the risk sit. In a conventional build, electricians and pipefitters do the terminations and the welds in place, on a schedule that waits for the building to be weathertight. In a modular build, most of that happens on a factory floor, under cover, by crews that wire the same module over and over. The site keeps the foundation, the rigging, the field joints between modules, and the test that proves the seams. Get the factory part right and the site part is fast. Get it wrong and you have shipped the problem hundreds of miles before anyone finds it.

Why deploy a modular data center instead of building one?

The reason is time, and time is now the constraint that drives the whole market. A conventional 50 MW build averages around 18 months and routinely stretches to 24 or 36 months once delays stack up. A prefabricated approach can move from order to operating capacity in a few months, because the module gets built in a factory while the pad and utilities go in on site at the same time. The two schedules run in parallel instead of end to end.

AI compute is what turned that speed from a nice-to-have into the default. Training clusters need megawatts energized in months, and the gap between when the load is ready and when a stick-built hall could be finished is the whole argument for modular. Vendors are now quoting deployment-time cuts in the range of 50 to 60 percent against conventional construction for prefab AI configurations, and the field has gone from edge novelty to a mainstream way to add capacity.

Speed is not the only payoff. A factory turns out the same wiring, the same pipe, the same panel layout repeatedly, which tightens quality and cuts the field rework that eats conventional schedules. Cost moves too, less from a lower sticker price and more from a shorter schedule, fewer trades stacked on site, and less weather exposure. The honest caveat: modular trades flexibility for speed. You are buying a designed block, so the time to shape it is before the factory cuts steel, not after the module is on the truck.

What are the types of modular data center?

Modular splits into three rough families, and the project usually mixes them. The all-in-one enclosure packs power, cooling, and a small IT space into a single container or purpose-built box. The skid-mounted module is a steel frame carrying one function, a power skid or a cooling skid, pre-piped and pre-wired for fast hookup to the rest of the plant. The separate prefab modules are larger purpose-built rooms, a power module and a cooling module feeding a white-space IT module, set side by side and tied together on site.

The container versus purpose-built choice is the one people get hung up on. An ISO shipping container is cheap to move and standard to handle, but its dimensions box in your rack count and service clearances. A purpose-built module is sized for the equipment instead of the truck, which gives better access and density at the cost of more planning around transport and rigging. For dense AI racks, purpose-built tends to win because the container envelope runs out of room for the cooling.

Match the type to the job. A single edge site wants the all-in-one. A campus adding load in steps wants skids and modules that repeat. A hyperscale or AI build usually runs a hybrid: prefab power and cooling modules feeding a white-space superstructure that may itself be partly conventional. There is no single right architecture, only the one that fits the load profile, the site, and how fast the capacity has to come on.

The factory build and where the quality comes from

The quality case for prefab is the controlled environment. A factory wires the switchgear, pulls the cable, runs the pipe, and sets the equipment indoors, on a bench, with the same crew building the same module to the same drawing again and again. Compare that to a field termination made in a half-finished room in February by whoever was available that week. Repetition plus cover plus fixed tooling is what tightens the work, and it shows up as fewer loose lugs, cleaner pipe, and less field rework.

Repeatability is the real product. The first module of a design is the prototype where the integrator finds the interferences, the access problems, and the sequence that does not match the drawing. Every module after that benefits from those corrections, so unit two is better than unit one and unit ten is better still. A conventional build never gets that learning curve because every room is its own first attempt.

The catch is that a factory will reproduce a bad detail as faithfully as a good one. If the design released to the floor has an undersized neutral or a pipe that fouls a filter change, the plant will build that fault into every module on the line. So the design review and the first-article check carry more weight in modular than they do in stick-built. You are not inspecting one room, you are approving a pattern that is about to be stamped out. Catch it on the prototype or you ship it on all of them.

What is a factory acceptance test?

A factory acceptance test, the FAT, is a structured, witnessed test run at the manufacturer's plant to prove the module meets the approved design and performance requirements before it ships. The owner or the commissioning agent attends, the module is powered and exercised against its load profile, and the result is a signed test record with measured values, pass or fail on each function, and a punch list of what gets fixed before the truck is loaded. The whole point is to catch the problem where remediation is cheapest, on the factory floor, with the panel still open and the right people standing in front of it.

On a modular data center the FAT is not optional, it is the thing that makes it a data center rather than a box of equipment. The plant tests power, cooling, fire detection, monitoring, security, and cabling against a defined load, often with a load bank standing in for the IT heat. A common field rule is blunt: if the vendor cannot show you a factory acceptance test report, you do not have a modular data center, you have a shipping container full of parts and a hope.

Witness it for real. Send someone who knows the sequences, not someone to collect a stamp. The FAT is where you watch the transfer actually happen, watch the alarm actually annunciate, and write the deficiency down while the manufacturer still owns it. The report it produces feeds straight into the site commissioning and the warranty record, so a thin FAT does not save time, it just moves the discovery to your pad. The full structure of a witnessed commissioning program, with the deficiency log and the owner's project requirements behind it, is the subject of the commissioning process guide; the FAT is that program's first hard gate.

Commissioning levels and the FAT-to-SAT handoff

Mission-critical commissioning is usually run in levels, commonly L1 through L5, and modular shifts where those levels happen. L1 is component and factory verification, L2 is delivery, storage, and installation checks, L3 is energizing and configuring each system on its own, L4 is performance under load and fault, and L5 is the integrated test across systems. The exact numbering varies by owner and program, so confirm the scheme against the project's commissioning plan rather than assuming one definition.

The modular move is to pull work forward into the factory. L1 plainly belongs there, and a good plant runs much of what L2 and L3 would normally cover in the field, energizing the module, configuring it, and even pre-running parts of L4 against a load bank before it ships. That is the schedule advantage of prefab applied to commissioning, not just to construction. The module shows up already proven on its own.

What cannot move to the factory is the integration. The seams between modules, the power module feeding the cooling module feeding the white space, only exist once the modules are set and connected on site. So the field still owns the site acceptance test and the L5 integrated systems test, no matter how thorough the FAT was. The handoff to get right is the documentation one: the FAT record has to carry into the site program cleanly so you re-verify what travel and rigging could have disturbed, and you do not re-run from scratch what the factory already proved.

Site preparation: the pad, the utilities, and the structure

The site work that runs in parallel with the factory build is what makes the schedule add up, and the foundation is the long pole. The module sets on a pad or piers engineered for its weight, including the live load of full racks, batteries, and water, plus the seismic and wind demands for the location. A power module loaded with switchgear and a UPS battery string is heavy and concentrated, so the structural design is sized to the real module weights from the manufacturer, not a generic allowance.

Utilities get stubbed up to meet the module where it lands. That means the medium-voltage or low-voltage feed, the conduit and duct bank, the chilled-water or condenser-water pipe, the makeup water and drain, the fire and signal pathways, and the grounding, all brought to the connection points the module drawings call out, at the elevations they call out. The single most common site-prep miss is stub-ups that do not line up with the module's connection points, which turns a fast hookup into field rerouting.

Access has to be designed, not assumed. The module has to reach the pad on a truck and then get lifted into place, so the haul route, the laydown area, the crane pad, and the swing radius are part of site prep, not an afterthought on set day. Crane access and ground-bearing capacity can kill an otherwise sound logistics plan, and you find that out the hard way if you wait until the module is on the road to check whether the crane can stand where it needs to stand and reach where it needs to reach.

Shipping, rigging, and the inter-module connections

Transport and the set are their own discipline. The module ships oversized and heavy, often on a multi-axle trailer, sometimes with a permit and an escort, and it has to survive the road without shaking a termination loose. On site it gets rigged off the trailer and lifted onto the pad, picked from engineered lift points to a marked center of gravity, set level, and anchored per the seismic and wind detail. A module set out of level or off its anchor points is a problem you carry for the life of the building, so it gets shimmed and verified before anyone connects anything to it.

Then come the field joints, which are where modular either pays off or bites. The connections between modules are the bus tie or feeder between the power module and the load, the pipe headers between the cooling module and the white space, the control and data tie-ins, and the structural and weather seals where modules meet. These are the few terminations that were not made in the factory, so they get the attention the factory joints already earned: torqued to spec and marked, pipe pressure-tested, and the seams flashed and sealed so the joint between two weathertight boxes does not become the leak.

Weatherproofing the field joint is the detail that gets rushed at the end of a fast set. Two modules each sealed at the factory still meet along a seam that has to be closed on site against water and air, and if that seam leaks you have undone the envelope on both. Treat the inter-module connections, electrical, mechanical, and the seal, as the punch list that has to close before the integrated test, because the integrated test will find the joint you skipped anyway, just later and with more equipment energized.

Site acceptance test and integrated systems test: what changes on site?

The site acceptance test, the SAT, re-proves on the pad what the FAT proved in the plant. It is not a repeat of the whole FAT, it is the verification that shipping, rigging, and the field connections did not disturb what the factory signed off, plus the checks that only make sense once the module is on real utilities. Read the FAT report, then go confirm the items most likely to move in transit: terminations, alarms, transfers, and anything the connection to site power and water touches. A clean FAT makes the SAT short. A skipped FAT makes the SAT a discovery exercise on your schedule.

The integrated systems test, the IST, is the one that cannot be done anywhere but site, because it tests the modules working together. The IST runs the full failure sequences across the connected plant: utility fails, the generators start and the transfer switches pick up, the UPS rides through, and the cooling holds the inlet temperature as the load banks simulate the IT heat, all measured as one system. This is the L5 test, and it is where the seams between modules either prove out or expose a sequence that worked module by module but not together.

The IST is where modular projects most often get short-changed, because the FAT went well and the modules each tested clean, so the integrated test feels redundant. It is not. The factory proved the modules. Only the IST proves the building. The mechanics of running an integrated systems test as a witnessed, scripted event, who calls the steps and what gets recorded, are covered in the commissioning process and the cooling and airflow guides; the modular point is simply that the FAT does not replace it.

The prefab power module and power skid

The power module is the prefabricated block that carries the critical electrical plant: the medium-voltage or low-voltage switchgear, the transformers, the UPS and its battery or flywheel energy storage, the static transfer and automatic transfer switches, the distribution, and often the paralleling and controls for the generators. It arrives wired, with the bus made up and the protection set, and the field work is the incoming feed, the bus tie to the load, the grounding, and the SAT.

Prefab is a strong fit for power because the work is dense, repetitive, and exactly the kind of termination that benefits from a bench and a fixed crew. Modules in the range up to a couple of thousand kVA are common, and a campus that needs more power adds more modules rather than enlarging one. The factory makes up the bus, sets the protection, and runs the transfer sequences against a load bank, so the module that ships has already shown it can take a utility loss and ride through.

The thing to verify hard at FAT and again at SAT is the protection coordination and the transfer timing, because that is what a load bank in a plant proves and a real outage will test for keeps. The selectivity has to be confirmed against the as-built settings, not the design intent, and the transfer has to actually hold the bus within the UPS ride-through. The deeper conduct of generator paralleling, transfer schemes, and protection sits in the broader power and commissioning coverage; in a module, the difference is that you are buying that engineering as a tested block instead of assembling it in the field.

The prefab cooling module and cooling skid

The cooling module is the prefabricated mechanical plant: the chillers or the pumped refrigerant gear, the pumps, the heat rejection, the coolant distribution for liquid-cooled racks, the headers, the valves, and the controls, built up and pre-piped on a skid or in an enclosure. It ships pressure-tested and charged or ready to charge, and the field work is the pipe tie-in to the white space, the power and control connections, and proving the loop holds and carries heat under load.

The cooling side is where prefab pays off in pipe quality and where it gets caught in field integration. A factory welds and tests the pipe under cover, which beats field welding a chilled-water header in a congested mechanical room. But the loop only closes once the module is connected to the rest of the plant on site, so the field joints in the piping, the flushing, and the air purge are real work that the FAT cannot finish. A cooling module that tested clean in the plant still has to prove delta-T and flow once it is feeding the actual load.

Watch the interface between the cooling module and whatever it serves, because that handoff is the common failure. The module's design flow and temperature have to match the white-space demand and the rack-level cooling, and a mismatch shows up as a hall that cannot hold inlet temperature even though every component tested fine. How cooling actually carries load, the ASHRAE inlet envelope, containment, delta-T, and the proof under load, is the subject of the cooling and airflow guide; the modular angle is that the plant is now a tested block you connect rather than a system you build in place.

AI density and the prefab liquid-cooling module

AI racks broke the assumption that a container's air handling could cool whatever you put in it. A single AI training rack can pull well past what air can carry, so high-density modular now ships with liquid cooling built in, and the prefab liquid-cooling module is its own block. It carries the coolant distribution unit, the CDU, the manifolds, the pumps, and the standardized liquid and electrical interfaces, sized to feed direct-to-chip cold plates or rear-door heat exchangers at the rack.

Prefabricating the liquid loop is attractive for the same reason as the pipe: the plant builds and leak-tests the manifolds and the CDU connections under controlled conditions, and the field work becomes the tie-in and the fill. CDUs are now shipping at multi-megawatt capacity for AI and HPC clusters, and rack-level integrated units combine the CDU, the rack space, and the manifolds in a single enclosure for the densest deployments. The interfaces are standardized so the module is closer to plug and connect than to a custom field build.

Leak integrity is the line you do not cross on a guess. A water or two-phase coolant joint over energized AI hardware is a different risk than a chilled-water main in a mechanical room, so the FAT for a liquid module has to include a real leak test and a proven leak-detection and isolation response, and the SAT has to confirm both survived the trip and the field connection. Liquid cooling as a discipline, the loops, the temperatures, and the commissioning, is covered in the cooling and airflow guide; the prefab point is that the module lets you buy a tested liquid loop instead of plumbing one in the field next to live servers.

Standardization, the parallel schedule, and single-source delivery

The quiet advantage of modular is the documentation reuse. When the same module design lands on three sites, the drawings, the FAT scripts, the commissioning procedures, and the as-builts carry from one to the next with edits, not rewrites. The crew that sets the second module already knows the connections. The operator who runs the third site sees the same panel layout as the first. That repeatability compounds across a fleet in a way a portfolio of one-off stick-built halls never does.

The schedule math is parallelism. The factory builds the module while the site crew pours the pad and runs the utilities, two efforts on the clock at once instead of trades waiting their turn in one building. That overlap is most of where the deployment-time savings come from, so the planning job is to keep the two tracks in step: the pad and stub-ups ready when the module arrives, not weeks before it sits exposed or weeks after it waits on a trailer.

Single-source delivery is the other lever, and it cuts both ways. Buy the module as an integrated, tested product from one integrator and you get one warranty, one party who owns the FAT, and one phone number when the integrated test finds a fault. Split the scope across vendors and you save on procurement but you own the integration risk and the finger-pointing when a seam fails. The trade is real: single-source costs more up front and buys you a clear line of responsibility, which on a fast schedule is usually worth paying for.

Can you add capacity to a modular data center later?

Yes, and pay-as-you-grow is one of the strongest reasons to go modular. Because the design is built from repeatable blocks, you can deploy the capacity the load needs now and add power, cooling, and white-space modules as demand grows, instead of building out a full hall on day one and paying to cool empty space. The capital follows the load rather than running ahead of it.

Phasing only works if the first phase was planned for it. The site, the utility service, the central pipe and bus, and the controls have to be sized or stubbed for the modules that come later, or the second phase turns into a teardown of the first. The practical move is to build the shared infrastructure for the end-state capacity, or at least to leave the connection points and the space, then add modules into that frame. A phase-two module that has nowhere to connect is not a phase, it is a new project.

The repeatability is what makes later phases fast. A module added in year three is the same design, the same FAT, the same connection detail as year one, so the second deployment is quicker and less risky than the first. That is the difference between modular phasing and just building another building: the block is known, so adding it is an exercise you have already run.

The edge and micro data center

At the small end, the same idea shrinks to a single enclosure. A micro data center is a self-contained, prefabricated unit, sometimes a single cabinet, that packs the racks, the UPS, the power distribution, the cooling, the monitoring, and the security into one box, built and tested before it leaves the factory. It is modular thinking at the scale of a closet or a cabinet instead of a campus.

The reason edge and micro deployments lean so hard on prefab is the location. These units land in retail back rooms, telecom closets, cell-tower bases, and remote sites with no on-site staff, so they have to arrive pre-tested and run unattended. The design target is lights-out: remote monitoring and management so nobody has to drive out to a site for routine work, and an enclosure ruggedized for an environment that is not a clean data hall. Prefabricated and pre-tested is not a preference here, it is the only way the model works.

The FAT matters even more on a unit that ships to a place nobody visits. A micro module that fails in the field is a truck roll to a remote site, so the factory test and the burn-in are the difference between an asset that runs for years untouched and one that becomes a recurring service call. Test it hard before it ships, because the whole premise is that it runs without hands on it.

Handover and the as-built record

The turnover for a modular build is a stitched record, factory plus field, and it has to read as one document set the operator can run against. The factory contributes the FAT report, the module's as-built drawings, the equipment submittals, the protection settings, and the configuration snapshot. The field contributes the SAT and IST results, the inter-module connection records, the pipe and bus test results, and the corrections from the deficiency log. The operator should not have to guess which drawing matches the module that actually got set.

The gap to watch is the seam between the factory record and the field record. A FAT report that never makes it into the site commissioning package, or a module that shipped with a revision the as-builts do not reflect, leaves the operator running a building against documents for a different building. Tie the records together at turnover so the as-built shows the module as built and as connected, including any field change made during the set.

Operations and maintenance documentation, the sequences, the spares, the maintenance intervals, has to cover the module as an integrated system, not just the loose equipment manuals stacked in a binder. How the turnover record gets assembled and what the owner should insist on, the commissioning report and the systems manual, is the close of the commissioning process guide; for a modular build the only twist is that part of that record was written on a factory floor and has to be carried forward intact.

Does a modular data center still have to meet local code?

Yes. A module built in a factory in one state and set on a pad in another still has to satisfy the building, electrical, mechanical, and fire codes the local authority having jurisdiction enforces, and it still gets inspected. Prefab changes where the work is done, not whether the work meets code. The AHJ holds the final say, including the basic question of whether a given unit is treated as equipment or as a building, which changes which code paths apply.

There is a listing path that helps. UL 2755, the Outline of Investigation for Modular Data Centers, lets a prefabricated unit be evaluated as a single integrated product, and it is referenced in NFPA 70, the National Electrical Code, at Article 646 covering modular data centers. A unit evaluated to that listing arrives with safety evidence that smooths the conversation with local inspectors, but the listing is not a building permit and does not override the AHJ. Confirm the article and the adopted code edition against the jurisdiction, because the numbering and the adoption vary by location and cycle.

Engage the AHJ early, before the factory cuts steel, not after the module is on the pad. The inspector cannot watch the work get done in a plant in another jurisdiction, so the path to approval has to be agreed up front: what gets witnessed at the factory, what gets inspected on site, and how the listing and the FAT records satisfy the local process. Leave that conversation until the module arrives and you risk a unit that is built, set, and stuck waiting on an inspection nobody planned for. Beyond the safety listing, Uptime Institute Tier criteria and TIA-942 still frame the reliability and the infrastructure expectations the way they would for any data center.

What to document

The modular record has a structure the field record alone does not, because it spans the factory and the site. For each module, you want the identity and design revision, the factory test result, the site re-test, the integrated test, and the connections that joined it to the plant, all traceable to the unit that actually got set. The table below is the minimum set that keeps the as-built honest when someone has to troubleshoot the building a year later.

Tie every row back to the specific module serial or tag, not the design in general, because two modules of the same design can ship with different revisions and different field corrections. The record that cannot tell you which module you are looking at is the record that fails you on the day you need it.

Common mistakes

• Skipping or thinning the factory acceptance test, so the defect ships hundreds of miles before anyone finds it.
• Site not ready when the module arrives, leaving it exposed on a trailer or the pad waiting on stub-ups.
• Stub-ups and connection points that do not line up with the module drawings, turning a fast hookup into field rerouting.
• Inter-module connections left untested, so the seam fails during the integrated test with equipment energized.
• Skipping the integrated systems test because the FAT went well, then discovering the modules do not work together.
• Pad or structure sized to a generic allowance instead of the real module weights, including batteries and water.
• Crane access and ground-bearing not verified before set day, stranding a good module on a trailer.
• AHJ engaged late, leaving a built and set module stuck waiting on an inspection nobody planned for.
• Factory test records that never make it into the site commissioning and as-built package.

Field checklist

Standards and references

The commissioning framework comes from ASHRAE Guideline 0, which defines the commissioning process the levels and the FAT, SAT, and IST sit inside, run against the owner's project requirements. The reliability and infrastructure expectations come from the Uptime Institute Tier system and from TIA-942 for data center telecommunications infrastructure, neither of which a module escapes by being prefabricated.

On the safety and code side, NFPA 70, the National Electrical Code, governs the electrical installation, and recent editions address modular data centers directly at Article 646, which references the UL 2755 Outline of Investigation for Modular Data Centers as the listing path for evaluating a unit as an integrated product. The exact article numbers and the edition in force vary by jurisdiction, so confirm them against the adopted code and any local amendments before citing them on a submittal.

Above all of these sit the project specification, the manufacturer's and integrator's requirements, and the authority having jurisdiction. The standards give the framework; the contract documents and the listed equipment requirements control the actual acceptance, and the AHJ controls the inspection and the classification. Cite the standard that governs the point, and let the project spec and the listing override a rule of thumb whenever they are stricter.

Units and terms

Modular deployment borrows vocabulary from construction, commissioning, and manufacturing, so the same idea shows up under different names across a submittal, a factory report, and a field package.

Prefabricated and modular are often used interchangeably but are not identical: prefabricated means built off site, modular means designed in repeatable blocks. The module functions are the power module or power skid, the cooling module or cooling skid, and the IT or white-space module, with the all-in-one enclosure combining them. The tests run FAT at the factory, SAT on site, and IST across the connected plant, mapped onto commissioning levels commonly numbered L1 through L5. Capacity is measured in kW or MW of IT load, power equipment in kVA, and the AHJ is the authority having jurisdiction that inspects and classifies the result.

Prefabricated vs modular

Prefabricated means built off site in a plant; modular means designed in repeatable blocks that scale by addition

FAT

Factory acceptance test, the witnessed test of a module at the manufacturer's plant before it ships

SAT

Site acceptance test, the on-site re-verification after the module is set and connected

IST

Integrated systems test, the L5 test proving the connected modules work together under load and fault

Power module / skid

Prefabricated electrical plant: switchgear, transformer, UPS, transfer switches, and distribution

Cooling module / skid

Prefabricated mechanical plant: chillers, pumps, heat rejection, and coolant distribution

CDU

Coolant distribution unit, the prefab block feeding liquid to direct-to-chip or rear-door cooling at the rack

AHJ

Authority having jurisdiction, the local body that inspects, classifies, and approves the installation

Related tools

Calculators and readiness checks for this work

Compare your options

FAQ

People also ask

• What is commissioning in construction?
• What is bypass airflow in a data center?
• What is an edge data center?
• What is data center commissioning?
• How long does data center electrical commissioning take?