GPU Racks
Standard rack positions supporting cold-plate liquid-cooled server installation and front/rear service access.

The IT module is based on a standard container structure, arranged internally with dual-row racks and a central service aisle, with network cabinet positions reserved at the ends. Racks, cable trays, cold-plate piping and monitoring points are all factory-installed.
Power-on, network connectivity and load tests are performed before shipment, so on-site work only requires connecting liquid cooling supply/return lines, electrical busway and uplink network before commissioning.
The module uses a unified definition for rack and piping interfaces, so adding more modules later requires no changes to the internal design of existing units.
Key equipment and systems are factory-installed, piped and tested.
Standard rack positions supporting cold-plate liquid-cooled server installation and front/rear service access.
Switching and uplink equipment centralized at the module ends for easy cross-module networking.
Main supply/return lines and rack-level manifolds, quick connects and shut-off valves run along the side walls.
Busway feeds rack PDUs, supporting dual-path power configurations.
Fiber and copper are routed in separate trays to reduce the risk of future changes.
Temperature/humidity, leak detection, access control and smoke sensors are pre-installed and connected to monitoring.
A quick look at how IT, cooling, and power modules differ and where each applies; specific parameters should be confirmed with site conditions and project evaluation.
| Dimension | IT MODULEIT Module | COOLING MODULECooling Module | POWER MODULEPower Module |
|---|---|---|---|
| Core role | Houses GPU racks, network equipment, and cold-plate liquid cooling piping — the direct carrier of compute capacity. | Collects heat from the IT side and rejects it outdoors, maintaining the temperature conditions equipment requires. | The complete chain from utility grid connection to rack power supply, ensuring continuity and metering of power. |
| Key components | Racks, cable trays and structured cabling, cold-plate manifolds, rack PDUs, environmental monitoring points. | Pump sets, heat exchangers, filtration and fluid make-up units, outdoor heat rejection equipment, water quality and flow monitoring. | MV/LV switchgear, transformers, UPS and batteries, busways and backup power transfer devices. |
| Key parameters to consider | Rack power density, number of racks, cooling method (cold-plate/air-liquid hybrid), network architecture. | Heat exchange capacity, supply/return water temperature and flow, free-cooling hours, PUE design target. | Incoming voltage level, total capacity, redundancy level, backup duration and transfer method. |
| Typical applicable scenarios | Added compute capacity, GPU platform upgrades, rack re-planning for training or inference scenarios. | High power-density liquid cooling deployments, free cooling in cold-climate sites, retrofits where existing cooling capacity is insufficient. | Access to renewable or hydro power, phased campus expansion, retrofits of existing facilities with limited power distribution capacity. |
| Interfaces with other modules | Connects to the cooling module's secondary supply/return water and the power module's busway feed, with unified monitoring integration. | Supplies water to the IT module, draws power from the power module, and feeds cooling capacity and water temperature data into monitoring. | Supplies power to the IT and cooling modules, with metering and protection data fed into monitoring. |
| Expansion approach | Add module units as compute demand grows, keeping the internal design of existing units unchanged. | Add pump sets and heat rejection units per thermal load, or add cooling modules. | Add power distribution units or increase incoming capacity as needed, releasing supply capacity in phases. |
Choose the enclosure form factor, compute workload, cooling and power scheme to generate a recommendation summary and matching interface list.
Enclosure form factor: 20ft enclosure
Uses a 20ft enclosure with fewer racks per module, suited to space-constrained sites or phased expansion, with lighter transport and lifting constraints.
Compute workload: Large model training
Planned for training workloads, with rack power density and network oversubscription prioritized for the low-latency, high-bandwidth needs of the compute fabric.
Cooling method: Full cold-plate liquid cooling
All racks use cold-plate liquid cooling; secondary-loop piping is factory-installed and pressure-tested, so only supply/return connections are needed on site.
Rack power supply: Single-path supply
Racks are supplied by a single PDU path, keeping the design simple; suited to clusters with lower availability targets or that rely on upper-layer scheduling for fault tolerance.
The above is directional guidance based on your selections; final capacity, equipment models and parameters need to be confirmed against site conditions and project evaluation.
The summary below covers the module's main interfaces and configurable range; the full "Interfaces & Technical Parameters" document (in Chinese) is available to download.
Selected per site space and transport conditions
Network rack positions reserved at the ends
Determined by equipment type
Determined by availability target
Includes pressure, temperature and flow measurement points
Planned per training or inference scenario
Factory pre-installed and connected to the cluster control plane
On-site work limited to pipe/cable takeover and commissioning
Document No. XA-IT-IFC | Rev. A | Updated 2026-09
The document (issued in Chinese) covers module composition, external interface definitions, configurable ranges and on-site interconnection points, for design and engineering teams to review.
Download PDF (Chinese)Parameters in the document describe the configurable range; specific values need to be confirmed against site conditions and project evaluation.
The answers below are based on publicly available product information; specific parameters need to be confirmed against site conditions and project evaluation.
Need a solution and parameters for a specific site? Pleasebook a technical discussionor visit thefull FAQ pageto learn more.
Schematic and physical views side by side for engineering review.


The following parameters are determined during project evaluation based on site conditions and business scale.
Modules connect via standard interfaces, requiring no on-site redesign.