AI FACTORY
REFERENCE
ARCHITECTURE

REFERENCE D60-W
WATER-COOLED CHILLERS WITH DRY COOLERS

 

A REFERENCE ARCHITECTURE FOR SCALABLE, HIGH-PERFORMANCE AI FACTORIES

WHAT IS A REFERENCE ARCHITECTURE?

A reference architecture is an engineering blueprint that defines how equipment, controls and systems work together to meet the requirements of a specific application. It establishes design conditions, equipment configurations, redundancy strategies and operating modes, providing a foundation for project-specific integration, reducing engineering time and supporting reliable system performance.

This guide presents Daikin’s cooling reference architecture for a 60 MW AI factory built around the NVIDIA DSX Platform and Vera Rubin NVL72 across six deployable units. It integrates liquid cooling, dedicated air cooling and facility-wide heat rejection into a coordinated system designed to support high-density AI workloads. The following sections detail the design conditions, equipment configuration, resiliency strategy and operating modes that engineers can use as a foundation for project-specific implementation and future scalability.

 

DESIGN BOUNDARY CONDITIONS

Site and Weather Icon

SITE AND WEATHER

ParameterReference Basis
Location / StationAshburn, VA (WMO: 724030)
Summer Dry Bulb39.8°C (103.6°F)
Summer Mean Coincident Wet Bulb28.1°C (82.6°F)
Winter Dry Bulb-12.9°C (8.8°F)
Temperatures Icon

TEMPERATURES

CircuitSupply / EWTReturn / LWT
Facility Liquid37°C (98.6°F)47°C (116.6°F)
Rack Liquid41°C (105.8°F)51°C (123.8°F)
RCU Water22°C (71.6°F)33.11°C (91.6°F)
Condenser Water44.8°C (112.6°F)50.4°C (122.7°F)
IT Air Temperatures30°C (86°F)44°C (111.2°F)
60 MW IT Load Icon
60 MW IT Load
NVIDIA DSX Icon
NVIDIA DSX AI Factory Platform
Vera Rubin Icon
Vera Rubin NVL72
Deployable Units Icon
6 Deployable Units
Reference Positions Icon
768 Reference Positions

COOLING SPLIT

92%
8%
92% Liquid Cooling
8% RCU Air Cooling
Deployable Unit Architecture Diagram

DEPLOYABLE UNIT ARCHITECTURE

13.8 MW DU Reference Capacity Icon
13.8 MW DU reference capacity
10 MW Plant Operating Allocation Icon
10 MW Plant operating allocation
128 Racks Icon
128 racks 4 Scalable Units
7 x 2.3 MW CDUs Icon
7 × 2.3 MW CDUs 6 duty + 1 standby
Liquid and Air Cooling Split Icon
9.2 MW liquid cooling 0.8 MW RCU air cooling Operating split: 92% liquid / 8% air

AI FACTORY ARCHITECTURE

Six Deployable Units with Dual Temperature Loops & Multiple Operating Modes

Heat rejection plant supports 60 MW IT load; additional plant capacity is required to fully utilize the total DU thermal capacity.

MODES OF OPERATION
Dry Cooling Icon
Dry Cooling
Hybrid Cooling Icon
Hybrid Cooling
Mechanical Cooling Icon
Mechanical Cooling
AI Factory Architecture Diagram
Mode 1 Dry Cooling Diagram
Equipment Configuration

WME Liquid-Cooling Plant

Operating Range

Dry cooler leaving fluid + HX approach ≤ 41°C (105.8°F)

MODE 1: DRY COOLING

Dry Cooling: Enabled when the dry cooler leaving-fluid temperature, plus the economizer HX approach, is at or below the facility supply-temperature setpoint. In this mode, the chiller compressors are off. During full economization only, facility supply resets up to 41°C, providing 45°C rack supply through the CDU. Hybrid and mechanical operation modes restore 37°C facility supply.

Mode 2 Hybrid Cooling Diagram
Equipment Configuration

WME Liquid-Cooling Plant

Operating Range

41°C (105.8°F) < dry cooler leaving fluid + HX approach < 47°C (116.6°F)

MODE 2: HYBRID COOLING

Hybrid Cooling: Enabled when the dry cooler leaving-fluid temperature, plus the economizer HX approach, is above the facility supply setpoint but below the facility return-water temperature. The economizer precools the return water, and chillers provide the remaining cooling to reach the supply setpoint, significantly reducing energy consumption.

Mode 3 Mechanical Cooling Diagram
Equipment Configuration

WME Liquid-Cooling Plant

Operating Range

Dry cooler leaving fluid + HX approach ≥ 47°C (116.6°F)

MODE 3: MECHANICAL COOLING

Mechanical Cooling: Enabled when the dry cooler leaving-fluid temperature, plus the economizer HX approach, is at or above the facility return-water temperature, leaving no useful economizer cooling. In this mode, the economizer is bypassed and the chillers maintain the facility supply setpoint.

60 MW Central Plant

Load (MW)Per DU AllocationCentral Plant
Liquid9.255.2
RCU air0.84.8
Total1060

OVERALL SYSTEM LOADS & REDUNDANCY STRATEGY

Deployable Unit Redundancy

  • CDU 6+1
  • RCU fans and coils: 2N design target

Central Plant Redundancy (N+2)

  • WME Quad (High Temperature Loop) 6+2
  • Dry Coolers (High Temperature Loop) 21+2
  • WME Single (Low Temperature Loop) 4+2
  • Dry Coolers (Low Temperature Loop) 2+2

COOLING DESIGNED TO WORK TOGETHER

High-performance data centers demand more than individual pieces of equipment. We apply a systems-level approach to cooling, prioritizing reliability, efficiency, and coordinated operation across the thermal infrastructure.

01

UNIFIED PERFORMANCE

Systems engineered to operate together deliver better cooling stability, efficiency, and scalability.

02

RESILIENCY BY DESIGN

Equipment capacity redundancy and coordinated controls support availability.

03

FUTURE-PROOF FLEXIBILITY

Air and liquid cooling paths designed to evolve with rising compute density.

04

SUSTAINABILITY AT SCALE

Non-evaporative heat rejection, low-GWP refrigerants and heat reuse options reduce environmental impact.

05

FOSTERING STRONG PARTNERSHIPS

Taking ownership and driving measurable results through a proactive, hands-on approach focused on innovative, tailored solutions.

06

FASTER DEPLOYMENT

Reference architecture accelerates design, integration and commissioning.

07

PROTECTIVE MAINTENANCE PLANS

Comprehensive plans designed to keep critical assets operating at their best.

08

24/7 SERVICE & SUPPORT

From seasonal tune-up, critical repairs, and system diagnosis to comprehensive project management and readily available, trained field service personnel for on-site support.

KEY EQUIPMENT

Daikin Coolant Distribution Unit

Daikin Coolant Distribution Unit

  • Total Capacity: 7 × 2.3 MW deployable units
  • Redundancy: 6 duty + 1 standby
  • Approach Temperature: 4°C
  • Operating Pressure: 50 psi (3.45 bar)
  • Controlled Rack Conditions: Regulates liquid supply temperature, flow and pressure.
  • Redundant Pumps & Filtration: Supports continuous uptime with failsafes designed to counter hardware failure.
  • Load Sharing & Operating-Hour Balancing: Uses Daikin proprietary control logic to maintain uptime and maximize efficiency.
  • Hot-Swappable Components: Enables maintenance and repairs without data center downtime.
  • Filtration: 25 μm, Beta 75
  • Approach Temperature: 4°C
  • Capacity Range: 1 MW to 3.5 MW
  • Materials Compliance: Wetted-material compliant
DDC S-Series Platform

DDC S-Series Platform

  • System Type: Rack Containment Units (RCU)
  • Deployment Capacity: 128 RCUs per deployable unit
  • Cooling Approach: Captures heat at each rack and transfers it directly to the water loop.
  • Integrated Protection: Fire suppression and water-loop protection built into every rack.
  • High-Density Cooling: Captures heat at the rack to help maintain neutral room temperatures and minimize hot spots.
  • Hybrid Cooling Compatibility: Supports air-cooled, direct liquid-cooled (DLC) and hybrid racks in the same row.
  • Scalable Deployment: Expand rack capacity within existing halls as power and chilled water become available.
  • Liquid-to-Chip Ready
  • Dynamic Density Control™
  • Biometric Rack Access
  • Telemetry + DCIM Integration
Magnitude WME-C Quad

Magnitude® WME-C Quad Magnetic Bearing Chiller

  • System Configuration: 8 Magnitude® WME Quad water-cooled chillers
  • Capacity: 10 MW each
  • Redundancy: 6 duty + 2 redundant
  • Technology: Oil-free magnetic bearing centrifugal chillers designed for AI factory cooling.
  • Operational Benefit: Supports rapid cooling recovery without oil-related startup delays.
  • Compact Footprint: Simplifies installation and maximizes usable floor space to help reduce costs.
  • High Efficiency: Oil-free magnetic bearing technology reduces energy costs and improves PUE.
  • Dry-Cooling Applications: Eliminates the need for cooling towers, reduces water usage and simplifies infrastructure.
  • Maximum Uptime: RideThrough® and RapidRestore® support cooling continuity and rapid recovery.
  • Load Sharing: Distributes cooling loads across all available chillers.
  • Refrigerant Options: R-513A or R-515B
  • Capacity: 10 MW per chiller
  • Oil-Free Operation
Magnitude WME-A Single

Magnitude® WME-A Single Magnetic Bearing Chiller

  • System Configuration: 6 Magnitude® WME single water-cooled chillers
  • Capacity: 1.4 MW each
  • Redundancy: 4 duty + 2 redundant
  • Application: Dedicated rack containment unit (RCU) cooling loop.
  • Water Supply: Maintains a reliable 22°C (71.6°F) water supply through oil-free technology.
  • Compact Footprint: Simplifies installation and maximizes usable floor space.
  • High Efficiency: Oil-free magnetic bearing technology reduces energy costs and improves PUE.
  • Dry-Cooling Applications: Eliminates the need for cooling towers and reduces water usage.
  • Maximum Uptime: RideThrough® and RapidRestore® support cooling continuity.
  • Load Sharing: Distributes cooling loads across all available chillers.
  • Refrigerant Options: R-513A or R-515B
  • Capacity: 1.4 MW per chiller
  • Oil-Free Operation
Data Center Dry Cooler

Data Center Dry Cooler

  • System Configuration: 27 dry coolers
  • Nominal Capacity: 3.0 MW
  • Liquid Cooling: 21 duty + 2 standby
  • Air Cooling: 2 duty + 2 standby
  • Application: Heat rejection for liquid and RCU cooling plants.
  • Availability: 2027 H2
  • Water Conservation: Avoids evaporative heat rejection and cooling-tower makeup water.
  • Efficiency & Sustainability: Designed for modern, high-density AI server architectures.
  • Lower Maintenance: Eliminates cooling-tower blowdown while maintaining closed-loop water quality.
  • Low Noise: Supports site-specific acoustic requirements.
  • Approach Temperature: 5°C
  • Noise Level: 70 dBA
  • Footprint: 8 ft 5 in × 53 ft 11 in
  • Seismic Certification

iDCM SUPERVISORY SOLUTION

iDCM Supervisory Solution Architecture Diagram

Load sharing, operating-hour balancing and three-mode plant control.

Optimization for energy efficiency.

Advanced predictive maintenance.



Need a partner that delivers end-to-end data center cooling?

Connect with our team to learn how Daikin’s innovative data center cooling solutions and maintenance services are engineered to help your data center stay cool and efficient. From standard deployments to highly complex AI data center operations, connect with Daikin to get started.

Contact Your Local Sales Rep →