SETEK / AI PREFABRICATED MODULAR DATA CENTER

AI infrastructure.
Engineered as a system.

High-density computing works only when rack layout, protected power, liquid cooling, networking and operational monitoring are designed together. SETEK AI PMDC brings these interfaces into one factory-integrated concept, configured around each project.

Open-side SETEK modular AI data center concept with server racks

MODEL OPTIONS

Start with the workload.
Then select the enclosure.

Model 20 and Model 40 are project configuration families, not fixed turnkey specifications. The usable load depends on the servers, rack arrangement, cooling infrastructure and service clearances.

COMPACT FOOTPRINT

Model 20

20-foot platform considered for edge AI, pilots or limited-footprint installations. Rack quantity, IT power and cooling approach are defined through project engineering.

Ideal when space and site interfaces are the primary constraints.
HIGH-DENSITY IT MODULE

Model 40

40-foot high-cube IT-module planning baseline with 10 rack positions for D2C-oriented concepts, coordinated with separate Power and Cooling Modules where needed.

Final rack density and IT load require engineering verification before proposal.

THREE MODULES · ONE COORDINATED DESIGN

Clear responsibilities.
Connected interfaces.

Separating compute, protected electrical power and thermal plant can create more usable IT space and clearer maintenance boundaries—but only with correctly engineered interconnections.

01 / IT MODULE

Compute

GPU servers, racks, internal network fabric, liquid manifolds, residual air, cable paths and service access.

  • Rack density & clearance review
  • InfiniBand or Ethernet/RoCEv2 options
  • Rack-level monitoring interfaces
02 / POWER MODULE

Power

Utility input, UPS, energy storage strategy, protective distribution, rack PDUs and metering.

  • Peak and continuous power design
  • UPS autonomy & redundancy choice
  • Selective protection & earthing
03 / COOLING MODULE

Thermal

CDUs, pumps, heat exchangers and external heat rejection matched to the IT loop and ambient conditions.

  • Temperature & flow setpoints
  • Leak detection & redundancy
  • Dry cooler / other project solution

DIRECT-TO-CHIP (D2C)

Follow the heat.
Protect the compute.

Cold plates draw heat from liquid-cooled components; the CDU manages the interface between the IT coolant circuit and facility-side heat rejection. Remaining non-liquid heat still requires suitable air cooling.

01GPU / CPU cold platesHeat captured at the component
02Rack manifoldsDistribution to liquid-cooled equipment
03CDU & heat exchangerControlled transfer to facility loop
04Heat rejectionSite-sized outdoor thermal plant
ILLUSTRATIVE HEAT BALANCE · NOT A PRODUCT RATING

800 kW IT heat load

If a specified server platform transfers 90% of heat to liquid, an indicative split is 720 kW liquid-side plus 80 kW residual air-side. Pumps, CDU losses, design margins and actual equipment conditions are evaluated separately.

Liquid · 720 kW Residual air · 80 kW

INTERACTIVE / CONCEPT PLANNING TOOL

Understand rack density.
See the thermal consequence.

Explore how a planning rack load scales with IT power and residual air cooling. This calculator is a conceptual heat-balance illustration—not an equipment specification, BOM or certified PUE.

20–120 kW · illustrative planning range
10 positions is the Model 40 D2C concept baseline
Depends on the selected IT hardware and operating point
LIVE ILLUSTRATIVE HEAT BALANCE
800kW IT heat
Liquid circuit720 kW
Residual air80 kW

External heat rejection, CDU/pump losses, non-IT loads, redundancy and temperature margins need separate engineering.

MONITORING, SAFETY & EFFICIENCY

The infrastructure must
also be operable.

Good design supports diagnostics, maintenance and safe intervention throughout the lifecycle.

01

Electrical resilience

Power-source monitoring, UPS status, battery alarms and rack distribution metering.

02

Environmental awareness

Temperature, humidity, coolant leak, smoke and access signals placed where they support intervention.

03

Cooling control

Coolant temperatures, flow, pressures, pump health and heat-rejection conditions as needed for the chosen platform.

04

Meaningful PUE

PUE is total facility energy divided by IT equipment energy. Any target must state measurement boundaries and real operating conditions.

ENGINEERING TO COMMISSIONING

Factory-integrated.
Site-validated.

Factory work and site preparation can run in parallel, but delivery depends on approved design, equipment availability, regulatory needs and test scope. No universal lead time is assumed.

01

RFI & Basis of Design

Define workloads, rack-level power, thermal interfaces, availability objectives and site constraints.

02

Preliminary engineering

Prepare architecture, module boundaries, outline layouts, major BOM and a budget proposal.

03

Manufacturing & FAT

Coordinate assembly, inspection, functional sequences and recorded factory acceptance checks.

04

Installation & SAT

Connect utility and coolant interfaces, verify safety and controls, and document site acceptance.

Lead time, availability level, energy efficiency, usable rack power, certifications and final scope are established in each project proposal. All capacities and performance commitments require project-specific validation.

DOBE COMPUTING PARTNER REFERENCE CASE STUDY

AI in the field.
One mobile deployment pattern.

This reference case from our Korean partner shows how AI compute, command operations and field utilities can be combined in a ruggedized container platform. It is presented as an example of deployment logic—not as a standard SETEK configuration.

REPUBLIC OF KOREA ARMY

AI Border Surveillance System

Reference platform: a custom 40 ft Edge Mobile Data Center arranged for AI-powered defense operations. The concept separates command activities, server infrastructure and mechanical support areas within an ISO-format transportable enclosure.

ClientRepublic of Korea Army
Product concept40 ft Edge Mobile DC – Custom
Design intentRapid deployment with remote monitoring
  • Command center: workstations, large display walls, cabinets and operator-support environment.
  • Server room: cooling-rack-based AI/IT equipment, protected distribution and internal separation.
  • Mechanical room: condensing units and backup generation support outside or adjacent to the IT area.

FAQ

Questions that define
the right design.

Is direct-to-chip cooling the baseline?

For SETEK's high-density AI PMDC concept, D2C with a CDU and residual air cooling is the default planning approach. Immersion, rear-door cooling or all-air designs are project-specific alternatives rather than assumed equipment.

Does a Model 40 always deliver a fixed IT capacity?

No. The 10-rack concept is a space-planning reference, not a fixed MW rating. The chosen servers, available electrical capacity, thermal solution and service clearances determine the achievable load.

How do we choose InfiniBand versus RoCEv2?

Start with the server and switch platform, training topology, bandwidth, congestion behavior and software requirements. Network interfaces are evaluated with the intended workload rather than assumed from the enclosure.

Can SETEK guarantee a specific PUE or installation period?

Only after the project has defined its IT load profile, climate and energy boundaries, equipment selections, procurement schedule, site readiness and commissioning criteria.

START WITH YOUR REQUIREMENTS

From “this is what we need”
to “ready for operation.”

Share your GPU platform, target IT capacity, rack density, redundancy, site location and preferred deployment date. SETEK can coordinate an initial technical review and a clear next-step proposal.