✦ PMDC · Prefabricated Modular Data Center

AI Data Centers,
Built in Months

Factory-built IT, power and cooling modules — integrated, FAT-tested, then commissioned on your site. A Tier III Ready reference design engineered for NVIDIA GB300 NVL72.

Factory-built PMDC module

7,764 kW

IT capacity

142 kW

per-rack density

1.15 –1.25

design PUE, 100% load*

N+1

power & cooling redundancy

Why Modular

Built in the factory, proven before it ships

Conventional data center construction takes 24–36 months of sequential site work. PMDC moves most of it into the factory — and runs it in parallel with your civil works.

Speed

Module fabrication and integration testing proceed while the site is prepared, compressing the overall schedule by up to half.

Flexibility

Deploy on idle land, in warehouses, or on constrained plots. Capacity scales in 2,588 kW units without disrupting operations.

Predictability

Standardized factory production with full FAT before shipment — the quality, availability and budget you approved is what arrives.

Conventional — sequential Design Civil & building works MEP fit-out & commissioning 24–36 mo PMDC — in parallel Design Factory build + FAT Civil works (parallel) Ship · install · SAT up to 50% shorter
What is PMDC?

More than a container — the whole system, modular

PMDC (Prefabricated Modular Data Center) is a construction method where IT data halls, power and cooling systems are pre-fabricated as modules in the factory, integration-tested (FAT), then assembled and commissioned on site.

How is this different from a "container box"?
Beyond power and cooling, fire suppression, security and monitoring are engineered into the modules as one system. You receive a verified data center — not a kit of parts.

System Composition

Three module types, one data center

IT Data Hall Module

The compute space

Power Module

The powertrain

Cooling Module

The heat-rejection plant

IT Data Hall 72 racks · 7,764 kW Power Modules 3 MW × 4 · N+1 Cooling Plant Free cooling + TES
AI-Ready

Engineered for GB300-class density

142 kW per rack — over 15× the density of a typical data center. The PMDC reference design engineers power, cooling and airflow for NVIDIA GB300 NVL72 systems from the ground up.

48

GB300 NVL72 systems (max)

142 kW

TDP per GPU rack (123.5 liquid + 18.5 air)

2,588 kW

IT load per scalable unit

8

power circuits per GPU rack (4 busways × 2)

Cooling Architecture

87% direct-to-chip liquid cooling

A dual-loop chilled water system: free-cooling chillers serve the air-side load, while dry coolers feed the DLC loop through CDUs. TCS supply is held at 35 °C for GB300-class racks, and a thermal storage tank guarantees uninterrupted cooling through any power transfer.

DRY COOLERN+2 · 30/42°C FC CHILLERN+1 · 20/30°C CDUFWS→TCS FAN WALLN+1·AIR 13% GPU RACKS142 kW · 87% DLC 30/42°C20/30°C TCS 35/45°C TES TANK
Power Architecture

Designed for Concurrent Maintenance

A distributed-redundant power architecture lets any single component be shut down for maintenance without touching IT operations. A fault in one powertrain never interrupts the data center.

UTILITY GENSET ATS POWER MOD ATX·UPS 3MW POWER MOD BTX·UPS 3MW POWER MOD CTX·UPS 3MW POWER MOD D4-TO-MAKE-3 BUSWAY A→ RACKS BUSWAY B→ RACKS
Deployment Process

Factory and site, moving at the same time

1 Design & layout 2 Order & procure 3 Factory build + FAT Civil works — parallel on site 4 Ship & deliver 5 Site installation 6 SAT & go-live

The key is step 3 — while modules are built and verified in the factory, civil works proceed in parallel on site. The waiting time of conventional builds is engineered out.

Reference Specifications

The 7.8 MW reference design

Overall
IT capacity7,764 kW = 3 × Scalable Units (2,588 kW) — adjustable per requirement
Design basisUptime Institute Tier III Ready · concurrent maintainability
Annual PUE1.15 – 1.25 (100% load, design simulation)
Total footprint≈ 2,030 m² (power 445 / cooling 869 / IT hall 714)
Ambient designASHRAE N=20-year extreme: 37.8 °C DB / 28.7 °C WB
IT & Racks
Rack configuration72 racks = 48 GPU (142 kW) + 12 switch (56 kW) + 12 management (23 kW)
Supported platformNVIDIA GB300 NVL72, up to 48 systems · room-based air + DLC in parallel
Rack liquid interfacePG25 coolant · 1.3 lpm/kW · FD83 couplings · 20 psid @ 45 °C
Airflow designFull hot-aisle containment (auto doors) · rack-on-slab · dual A/B busway · overhead trunking
Cooling
Heat capture ratio87% liquid : 13% air
Water temperaturesTCS 35/45 °C (PG25) · FWS 30/42 °C · air-side CHW 20/30 °C
Heat rejectionFree-cooling air-cooled chillers (N+1) + dry coolers (N+2) + TES tank
Room unitsFront-flow modular fan walls (N+1) · in-row CDUs (N+1)
Power
RedundancyDistributed redundant · UPS 4-to-make-3 (N+1) · power modules 3 MW × 4 · cooling on 2N feeds
Rack feedsGPU racks: 4 busways × 2 circuits = 8 feeds (3P+E 63A) · management racks on dual-source PDUs
BackupDiesel generator support · 12-hour on-site fuel autonomy · dedicated 1,000 kW UPS for critical cooling
Grid interfaceCustomized per site (reference: 22.9 kV / 380 V / 60 Hz)

All figures reflect the 7.8 MW reference design and are adjusted to site conditions and requirements. PUE values are design simulations.

Scope of Delivery

Clear from day one — what we deliver, what you prepare

✓ Delivered by HeTone

Prepared by the client (typical)

The exact scope is finalized in each project proposal. The split above reflects the typical baseline of the reference design.

Why HeTone

Vendor-neutral integration, Asian supply-chain speed

One-stop AI infrastructure

From GPU servers and racks to power, cooling and construction — liquid cooling technology and global computing infrastructure under one team.

Proven global architecture

Built on proven module architecture from global infrastructure partners, applying Uptime Institute Tier III Ready design criteria.

Taiwan & SEA supply chain

Manufacturing bases across Taiwan and Southeast Asia shorten lead times and diversify supply-chain risk.

Turnkey, single accountability

Design → factory build → FAT → shipping → installation → SAT, all under one contract and one responsible team.

FAQ

Frequently asked questions

A Prefabricated Modular Data Center: the IT hall, power and cooling systems are built as factory modules, integration-tested (FAT), then assembled and commissioned on your site. Unlike a simple containerized server box, the entire facility system — including fire suppression, security and monitoring — is engineered as one modular product.

Because factory fabrication runs in parallel with your civil works, overall schedules are typically up to half of a conventional build. The exact timeline depends on capacity and site conditions — we provide a stage-by-stage schedule at consultation.

CAPEX consists of the three module systems (IT / power / cooling), plus transport, installation and commissioning. It varies mainly with rack density, redundancy level and site conditions. Request a consultation for a budgetary estimate based on your requirements.

Tier III Ready means the module design follows Uptime Institute Tier III criteria — concurrent maintainability with N+1 redundancy. Site-level certification is a separate process performed on the completed facility; our design basis is intended to support that path, and we can assist with the certification process.

Often yes — via staged retrofit paths (e.g. rear-door heat exchangers as a transition, then DLC), or by adding PMDC modules alongside an existing facility to expand without major building works. A site assessment determines the best route.

The reference design targets NVIDIA GB300 NVL72 (142 kW/rack, 87% liquid capture). The cooling architecture’s temperature envelope (TCS 35/45 °C) is designed to accommodate future high-density platforms as well.

Get in Touch

Planning an AI data center?

Tell us your project scale and timeline — our engineers will follow up with materials based on the 7.8 MW reference design.