A 360MW headline can hide a 100MW procurement error.
Before issuing an RFQ for CDUs, dry coolers, cooling towers, pumps, piping, or control systems, buyers must first determine what “360MW” actually represents. Is it reserved utility capacity, total facility power, designed IT load, contracted capacity, or commissioned load?
These numbers are not interchangeable.
In June 2026, Firmus announced a dedicated 360MW NVIDIA DSX AI Factory campus in Batam, Indonesia, developed with DayOne. The announcement covers up to 170,000 NVIDIA AI accelerators across Grace-Blackwell, Vera-Rubin, and Vera platforms through 2027 and 2028.
The scale is impressive. Still, the number alone is not enough to prepare a cooling equipment purchase order.
The real procurement question is this:
How much IT heat must be removed in each deployment phase, at what temperatures, with what flow rate, redundancy, water strategy, and commissioning schedule?
1. Challenge the 360MW Number Before Sizing Equipment
A campus may be marketed as 360MW while only a portion of that capacity is available during the first construction phase.
Buyers should separate at least five capacity definitions:
| Capacity Definition | What It Means | Why It Matters |
|---|---|---|
| Utility capacity | Power reserved or available from the grid | May include future phases and non-IT loads |
| Facility input | Total power entering the data center | Includes IT, cooling, pumps, fans, lighting and losses |
| Designed IT load | Maximum planned server and network load | Main basis for long-term heat-load planning |
| Contracted capacity | Capacity committed to customers | May not yet be installed or commissioned |
| Commissioned IT load | IT capacity tested and ready for operation | Best basis for near-term equipment deployment |
If 360MW represents total facility input rather than IT load, the required IT cooling capacity will be lower.
For example, a facility operating at a PUE of 1.20 with 360MW of total facility input would support approximately 300MW of IT load:
IT load = 360MW ÷ 1.20 = 300MW
By contrast, if 360MW is the IT load, total facility input at the same PUE would be approximately 432MW:
Facility input = 360MW × 1.20 = 432MW
The additional 72MW represents total facility overhead. It should not automatically be labeled as cooling power because PUE also includes electrical losses, lighting, controls, pumps and other supporting systems.
Procurement decision: Never size the cooling plant from the campus headline. Request a phased IT-load schedule.
2. Convert IT Megawatts Into Heat and Flow
Almost all electrical power consumed by AI servers eventually becomes heat. Therefore, a 360MW IT load requires the cooling infrastructure to remove approximately 360MW of heat under full-load conditions.
That equals roughly:
360,000kW ÷ 3.517 = approximately 102,360 refrigeration tons
For a simplified water-based cooling loop:
Flow rate = Heat load ÷ (fluid density × specific heat × temperature difference)
The approximate full-load flow requirements are:
| Design Condition | 360MW Campus | One 30MW Capacity Block |
| Heat rejection | 360MW | 30MW |
| Refrigeration tons | 102,360 RT | 8,530 RT |
| Flow at 10K ΔT | 30,960m³/h | 2,580m³/h |
| Flow at 15K ΔT | 20,640m³/h | 1,720m³/h |
| Flow at 20K ΔT | 15,480m³/h | 1,290m³/h |
These figures are simplified heat-balance examples based on water-like fluid properties. Actual flow depends on glycol concentration, elevation, pipe pressure drop, heat-exchanger approach temperature, fouling allowance and equipment operating limits.
A wider temperature difference reduces required flow, pipe diameter and pumping energy. However, the server, cold plate, CDU and heat-rejection system must all support the selected supply and return temperatures.
Do not select ΔT only to make the pump smaller.
Pro Tip: Ask server suppliers for minimum flow, maximum pressure drop, allowable facility-water temperature and validated return-water temperature at full GPU load. A brochure showing only “liquid-cooled” is not a cooling design envelope.
3. Freeze the Server and Rack Design Envelope
Cooling procurement should not begin with the dry cooler or CDU. It should begin with the compute equipment schedule.
The buyer should confirm:
Server and GPU platform
Maximum rack power, not average rack power
Number of racks in each phase
Direct-to-chip or immersion architecture
Minimum and maximum coolant temperature
Rack-level flow requirement
Maximum allowable pressure
Coolant type and water-quality specification
Quick-disconnect and manifold requirements
Expected load variation during training and inference
The rack-density range also matters. A campus containing 80kW racks requires a different manifold, pump and air-cooling strategy from one designed for 250kW or 500kW racks.
Future GPU platforms may also change the cooling requirement before the campus reaches full capacity. For that reason, the design envelope should include an agreed upgrade range rather than one fixed rack value.
4. Confirm the Liquid Heat-Capture Ratio
Direct-to-chip cooling does not always remove 100% of server heat through liquid.
GPUs and CPUs may be connected to cold plates, while memory, power supplies, switches, storage and other components continue rejecting heat into the room. The remaining air-side load can still be substantial.
For illustration, if a 30MW IT block has an 80% liquid heat-capture ratio:
Liquid cooling load: 24MW
Residual air-cooling load: 6MW
That 6MW cannot be ignored. It still requires fan walls, air handlers, rear-door heat exchangers or another properly engineered air-cooling path.
Before equipment procurement, require the server supplier to provide the liquid heat-capture ratio under maximum validated workload. The ratio should be tied to a specific server configuration rather than a general platform family.
5. Size CDUs at Real Operating Conditions
CDU selection is not simply:
IT load ÷ CDU nameplate capacity
The usable CDU capacity depends on facility-water temperature, technology-water supply temperature, approach temperature, coolant type, pump head and heat-exchanger condition.
Suppose the liquid load for a 30MW block is 24MW. If the project uses nominal 2MW CDUs, the arithmetic suggests twelve active units. An N+1 arrangement may require thirteen.
However, that calculation is valid only if each CDU can actually deliver 2MW at the project’s specified temperatures, flow and pressure.
The CDU specification should confirm:
Rated and guaranteed capacity at design conditions
Maximum and minimum flow
Available pump head
Heat-exchanger approach temperature
Pump and electrical redundancy
Filtration and air-removal system
Expansion and pressure-control method
Leak detection
Sensor accuracy and calibration
Local control during BMS communication loss
BACnet/IP, Modbus TCP or other required protocols
Service clearance and component replacement path
Spare-parts availability
Pro Tip: Separate capacity margin from redundancy. A system described as “20% oversized” is not automatically N+1. If one unit fails during peak load, the remaining units must still satisfy flow, pressure and temperature requirements.
6. Divide the Campus Into Hydraulic Zones
One campus-wide cooling loop creates an enormous common failure domain.
A more practical strategy is to divide the 360MW campus into repeatable capacity blocks. One possible planning model is twelve 30MW blocks, with each block divided into smaller CDU and rack-level zones.
The exact block size will depend on building layout, electrical topology and deployment schedule, but the principle remains the same:
Isolate failures. Repeat validated designs. Commission capacity in phases.
A zoned architecture may include:
Campus heat-rejection loop
Building or module facility-water loop
CDU heat-exchanger boundary
Technology cooling system supplying servers
Rack manifold and cold-plate loop
Each zone should include suitable isolation valves, bypasses, drains, fill points, filtration, pressure control, leak detection and maintenance access.
Buyers should also verify whether a failure in one CDU, pump, controller or network switch can affect multiple zones. Electrical and controls common points are often overlooked even when the hydraulic system appears redundant.
7. Match Heat Rejection to the Site
The best heat-rejection system depends on climate, water availability, coolant temperature, electrical price and operating priorities.
| System | Main Advantage | Main Procurement Risk |
| Dry cooler | Closed-loop operation and low direct water consumption | Capacity falls as outdoor temperature approaches coolant temperature |
| Cooling tower | Lower leaving-water temperature based on wet-bulb conditions | Water use, treatment, plume, permits and maintenance |
| Chiller | Stable supply temperature across difficult conditions | Higher power consumption, CAPEX and mechanical complexity |
| Hybrid system | Balances energy, water and peak-weather performance | More complex controls and operating sequences |
Warm-water liquid cooling can increase the number of hours during which dry coolers operate without compressors. Still, dry-cooler capacity must be checked at the site’s actual summer design temperature, altitude and fouling condition.
Cooling towers can achieve lower water temperatures, but WUE, water availability, treatment cost and local environmental requirements become major procurement factors.
Chillers may be used as full-time cooling, peak-temperature support, trim cooling or backup. Buyers should define that operating mode before requesting prices.
Pro Tip: Compare annual operating profiles, not one design-point efficiency. A dry cooler may have excellent WUE but require higher fan power during hot weather. A cooling tower may reduce compressor energy while increasing water and treatment costs.
8. Confirm Water Chemistry and Material Compatibility
Many liquid-cooling failures begin as chemistry problems rather than capacity problems.
The project team should create one approved list of all wetted materials, including:
Copper
Stainless steel
Aluminum
Brass
Elastomers
Gaskets
Hose materials
Heat-exchanger plates
Cold-plate materials
Mixing incompatible metals can create galvanic corrosion. Poor oxygen control, incorrect inhibitor concentration, biological growth or particle contamination can damage cold plates and narrow channels.
The procurement specification should define acceptable limits for pH, conductivity, hardness, chloride, suspended solids, inhibitor concentration and biological activity where applicable.
It should also identify who owns sampling, laboratory testing, fluid replacement and corrective action after handover.
9. Specify Controls, Alarms and Restart Logic
At 360MW scale, cooling performance depends as much on control logic as on mechanical capacity.
The sequence of operations should cover:
Normal startup and shutdown
Lead-lag pump rotation
Partial-load control
CDU staging
Dry-cooler fan control
Cooling-tower or chiller staging
High-temperature response
Low-flow response
Sensor failure
Communication loss
Pump or fan failure
Utility power loss
Emergency shutdown
Controlled restart after power restoration
A CDU should retain enough local control to protect the servers if communication with the campus BMS or DCIM platform is interrupted.
Alarm priorities, delay times and escalation paths should be agreed before FAT. Otherwise, the project may generate thousands of alarms without giving operators a clear response order.
10. Define Redundancy at the Capacity-Block Level
“N+1 cooling” is incomplete unless the procurement document defines what the “N” represents.
It may refer to pumps, CDUs, heat exchangers, dry coolers, cooling towers or complete cooling trains. These are different levels of redundancy.
Buyers should confirm:
Which equipment is N+1
Which systems require 2N
Whether maintenance can occur at full IT load
Whether shared headers create a common failure point
Whether controls and communications are redundant
Whether standby equipment starts automatically
Whether the system has been tested during actual failover
For high-value AI training clusters, the financial impact of a thermal shutdown can exceed the cost of additional cooling equipment. Still, adding redundant hardware without isolation and validated controls will not create real resilience.
11. Make FAT and SAT Part of the Purchase Order
Cooling equipment should not leave the factory based only on visual inspection.
The Factory Acceptance Test should verify:
Pump flow and head
Heat-exchanger performance
Sensor calibration
Valve operation
Filter differential pressure
Leak detection
Alarm and interlock logic
Lead-lag rotation
Redundant power transfer
Communication with BMS or DCIM
Power-loss behavior
Automatic restart sequence
Documentation and nameplate accuracy
Where practical, load simulation should be used to verify thermal performance across partial and full-load conditions.
After delivery, the Site Acceptance Test should confirm flushing, pressure testing, water quality, balancing, communication, failover and operation with the actual site heat-rejection loop.
An integrated systems test should then verify how the cooling, power, controls and IT systems respond together during abnormal conditions.
Pro Tip: Put acceptance criteria, calibrated instrument requirements and test-report formats into the purchase order. “FAT included” is too vague to protect the buyer when performance disputes occur.
12. Procure Repeatable Capacity Blocks
Ordering equipment for the entire 360MW campus on day one may lock the project into technology assumptions that change before later phases are built.
A phased strategy can reduce this risk:
Standardize a validated cooling block
Complete factory integration and FAT
Deploy the first operational phase
Record temperature, flow, pressure, PUE and WUE
Correct the design before repeating it
Reserve physical and electrical space for future expansion
Standardized modules can shorten onsite installation and commissioning while improving quality control. However, interface management remains essential. Pipe sizes, electrical connections, network protocols, structural loads and field responsibilities must be fixed before shipment.
13. Calculate ROI From Commissioned Capacity
The lowest equipment price does not always produce the lowest project cost.
A more useful procurement metric is:
Cooling TCO per commissioned IT MW =
Equipment + piping + electrical integration + logistics + installation + commissioning + energy + water + maintenance + spares + expected delay exposure
÷
Commissioned IT capacity
Delay exposure should also be included:
Monthly delay exposure = unavailable IT MW × expected contribution per MW-day × delay days + financing and contractual costs
This framework explains why factory-tested modular cooling equipment may produce a better return even when its initial equipment price is higher. Bringing revenue-generating compute online earlier can be more valuable than saving a small percentage on cooling hardware.
14. Final 360MW AI Campus Cooling RFQ Checklist
Before asking suppliers for a binding quotation, send the following information:
Definition of the 360MW capacity
Deployment schedule by phase
IT load per building or module
Server and GPU models
Rack count and maximum rack power
Liquid heat-capture ratio
Facility-water supply and return temperatures
Technology-water supply and return temperatures
Coolant type and water-quality limits
Required flow and allowable pressure drop
Site climate and altitude data
Water availability and discharge restrictions
Selected heat-rejection strategy
PUE and WUE targets
Redundancy and maintainability requirements
BMS and DCIM communication protocols
Available electrical voltage and frequency
FAT, SAT and integrated-test requirements
Shipping dimensions and delivery limitations
Spare-parts and service expectations
A supplier cannot provide a reliable fixed price when these inputs remain undefined. The result will either contain a large risk premium or a long list of exclusions.
Final Procurement Verdict
For a 360MW AI campus, cooling procurement should begin with capacity definitions and deployment phases, not equipment catalogs.
Confirm the IT load. Freeze the server design envelope. Establish the liquid heat-capture ratio. Select temperatures and flow rates. Divide the campus into isolated hydraulic zones. Then procure CDUs, heat-rejection equipment, controls and commissioning services as repeatable capacity blocks.
Buy equipment for the load you can commission, while reserving interfaces for the load you plan to build.
ACT-Boxes supports modular AI data center infrastructure, including CDUs, dry coolers, liquid-cooling loops, power distribution, controls, factory testing, delivery and onsite commissioning coordination. A useful cooling proposal starts with the phase-by-phase IT load, rack design, temperature requirements, site climate, redundancy target and deployment schedule.
FAQ
Does a 360MW AI campus need 360MW of cooling equipment immediately?
Not necessarily. The project may be deployed in phases, and the 360MW figure may represent utility, facility, designed or future capacity. Cooling procurement should follow commissioned IT load by phase.
How many CDUs are required for a 360MW AI campus?
It depends on the liquid heat-capture ratio, CDU capacity at actual operating conditions, hydraulic zoning and redundancy requirement. Dividing 360MW by the CDU nameplate rating is not sufficient.
Are dry coolers suitable for a 360MW campus?
They can be suitable when coolant temperatures, climate and available installation area support dry operation. Capacity must be verified at the site’s summer design temperature and altitude.
What temperature difference should the liquid loop use?
A larger ΔT reduces flow and pumping requirements, but it must remain within server, cold-plate, CDU and heat-rejection equipment limits. The final value should be validated as one system.
What is the difference between FAT and SAT?
FAT verifies equipment performance and controls before shipment. SAT confirms that the equipment operates correctly after installation with the actual site piping, power, water and control systems.
