How to Build a Procurement-Ready Cooling Plan Before Applying for Data Center Power

The utility application is the wrong place to discover that a 100MW AI campus actually needs 122MW at the meter.

Yet this happens when developers apply for power using only the expected IT load. The cooling plant, pumps, fans, CDUs, water-treatment equipment, UPS losses and auxiliary systems are added later, turning the original power request into an incomplete load forecast.

That mistake is becoming more expensive.

Dominion Energy’s public data center request workflow identifies a site plan, load letter and load ramp schedule among the information customers should provide. Meanwhile, the Federal Energy Regulatory Commission’s June 2026 large-load action focused attention on clearer interconnection processes, network-upgrade costs and flexible service for loads above 50MW connecting above 69kV.

The FERC action is a regulatory proceeding, not a guarantee of faster power. Still, it sends a practical message to data center developers:

Projects that can explain their real load, deployment phases and operating flexibility are better prepared for utility discussions.

A procurement-ready cooling plan gives the utility a defensible facility load while giving equipment suppliers enough information to quote CDUs, pumps, dry coolers, cooling towers, chillers, controls and modular cooling systems.

1. Start With Five Different Power Numbers

“Power requirement” should never appear as one unexplained number in a data center application.

The project team should define at least five values:

The IT design load is not the same as the utility service capacity.

Cooling power also changes with outdoor temperature, IT utilization, liquid-cooling percentage, redundancy state and the number of operating capacity blocks. Therefore, one annual-average PUE value cannot describe every operating condition.

2. Convert IT Capacity Into Facility Power

PUE is calculated as:

PUE = Total facility energy ÷ IT equipment energy

For a simplified 100MW IT campus:

The facility overhead includes cooling, electrical losses, lighting, controls and other infrastructure. It is not cooling power alone.

Google reports a trailing 12-month PUE of approximately 1.09 across its mature hyperscale data center fleet. That is a useful industry reference, but it should not be copied directly into a new project’s power application.

A new campus may operate at partial load for several years. It may also face higher summer temperatures, different redundancy requirements and less optimized equipment sequencing. Those conditions can produce a higher peak PUE even when the long-term efficiency target is aggressive.

Use an operating range, not a marketing target.

The utility-facing load letter should show normal load, seasonal peak load, first-phase load and ultimate build-out load.

Pro Tip: Do not multiply the final IT capacity by the best expected annual PUE and call that the utility requirement. The electrical application should be based on the worst credible coincident facility load, while the business model can use an annual energy estimate.

3. Freeze the Compute and Cooling Design Envelope

A cooling plan cannot become procurement-ready until the project establishes what it is cooling.

The design envelope should include:

• Server and GPU platform

• Maximum rack power

• Number of racks in each phase

• Direct-to-chip, immersion or hybrid cooling architecture

• Liquid heat-capture ratio

• Coolant supply and return temperatures

• Minimum rack flow

• Maximum pressure drop

• Coolant chemistry requirements

• Residual air-cooling load

• Expected training and inference load patterns

• Future hardware upgrade range

Do not use average rack power for infrastructure sizing. AI training clusters can create sustained high loads, while workload changes can also produce rapid power variations.

At the same time, avoid purchasing the entire cooling plant around one server generation. Later deployment phases may use higher-density hardware with different flow, pressure and coolant-temperature requirements.

The goal is to establish a validated range that supports present procurement without preventing future expansion.

4. Select the Liquid-Cooling Temperature Class Early

Coolant temperature is not a minor CDU setting. It influences the entire power application.

ASHRAE’s AI Data Center Energy Performance Framework presents revised liquid-cooling classes including W17, W27, W32, W40, W45 and W+. The number reflects the upper supply-water temperature associated with the class.

A lower coolant-supply temperature may require mechanical chilling for more hours of the year. A warmer loop can increase dry-cooling and free-cooling opportunities, reducing compressor power.

However, the selected temperature must remain within the validated limits of the server, cold plate, manifold and CDU.

The temperature decision affects:

• Chiller capacity

• Dry-cooler size

• Cooling-tower operating hours

• Pumping requirements

• Approach temperature across the CDU

• Annual cooling electricity

• Summer peak power

• Water consumption

• Heat-reuse potential

This is why coolant temperature must be agreed before the utility application is finalized. Changing from a warm-water dry-cooler system to a chilled-water system later may materially increase the requested electrical capacity.

5. Calculate the Liquid and Residual Air Loads Separately

Direct-to-chip liquid cooling may not capture all server heat.

Suppose a 100MW IT deployment has an 80% validated liquid heat-capture ratio:

• Liquid cooling load: 80MW

• Residual air-cooling load: 20MW

Using water-like fluid properties, the approximate facility-water flow for the 80MW liquid load would be:

These are simplified heat-balance calculations. Actual flow depends on glycol concentration, elevation, pipe resistance, fouling, heat-exchanger approach temperature and equipment operating limits.

Increasing ΔT can reduce flow, pipe size and pump energy. However, a larger ΔT is only valuable when the servers, CDUs and heat-rejection equipment can operate at the resulting return temperature.

The remaining 20MW of air-side heat still needs a defined cooling path. Depending on the architecture, that may include fan walls, air handlers, rear-door heat exchangers or other air-cooling equipment.

Pro Tip: Ask the server supplier for a heat-capture ratio tied to the exact server configuration and maximum validated workload. “Liquid-cooled server” is not a complete thermal specification.

6. Build a Seasonal Cooling Power Model

A procurement-ready cooling plan should calculate equipment power under several operating conditions, not only at one outdoor temperature.

At minimum, model:

1. Normal winter operation

2. Normal summer operation

3. Summer design-day peak

4. One cooling train unavailable

5. Partial-load first-phase operation

6. Utility outage and backup-power operation

7. Restart after power restoration

For each condition, calculate the expected power of:

• CDU pumps

• Facility-water pumps

• Dry-cooler fans

• Cooling-tower fans

• Chillers

• Condenser-water pumps

• Air-handling fans

• Water-treatment equipment

• Heat tracing and freeze protection

• Controls and auxiliary equipment

The summer design-day value is especially important for the utility application because dry-cooler capacity and fan power change as ambient temperature approaches coolant temperature.

Chiller power can also become a major peak load when the system transitions from free cooling to mechanical cooling.

The utility does not only need annual energy. It must understand the maximum coincident demand.

7. Compare Heat-Rejection Options Before Freezing the Load Letter

The correct system cannot be selected from PUE alone.

A dry-cooling strategy may reduce WUE but require more fan energy and larger heat exchangers. Evaporative cooling may improve electrical efficiency while increasing water consumption. Chillers can provide temperature certainty, but they may increase both service capacity and standby-generation requirements.

The cooling plan should show the selected architecture, an alternative architecture and the reason the final option was chosen.

8. Define CDU Capacity at Actual Operating Conditions

CDU nameplate capacity is not always the capacity available to the project.

Usable performance depends on:

• Facility-water inlet temperature

• Technology-water supply temperature

• Approach temperature

• Fluid type

• Required flow

• Pump head

• Heat-exchanger pressure drop

• Fouling allowance

• Redundancy arrangement

The Open Compute Project’s CDU guidance emphasizes specification, qualification and operational assurance rather than relying only on nominal megawatt ratings.

For utility planning, CDU power should include active pumps, standby strategy, controls and any operating mode in which additional units start during high temperature or component failure.

For equipment procurement, the RFQ should request guaranteed performance at the project’s actual design conditions.

Pro Tip: Separate cooling-capacity margin from N+1 redundancy. Twenty percent extra nameplate capacity does not guarantee that the system can maintain flow, pressure and temperature after one unit fails.

9. Show the Utility How the Load Will Ramp

A 120MW campus rarely reaches full power on the first day.

The power application should include a realistic load ramp, such as:

The dates and values must match the actual construction and server-delivery schedule.

Applying immediately for ultimate capacity may be necessary in a constrained market, but the project should still explain when each block will energize. Utilities need this information for transmission, substation and generation planning.

A modular cooling architecture makes the ramp schedule easier to defend because cooling equipment can be tied to specific IT capacity blocks.

10. Control Motor Starting and Restart Demand

The utility will care about more than steady-state megawatts.

Large chillers, pumps and fans can create starting current, voltage disturbance and sudden load steps. The project should therefore define:

• VFD use

• Sequential motor starting

• Maximum simultaneous motor start

• Chiller restart delay

• Pump lead-lag sequence

• Automatic load shedding

• Harmonic characteristics

• Power factor

• Reactive power requirements

• Black-start and restoration sequence

After an outage, starting every CDU, pump, fan and chiller simultaneously can create a facility demand spike that exceeds the normal operating load.

The controls narrative should specify how cooling capacity returns in stages while server temperatures remain protected.

11. Quantify Cooling Flexibility Without Risking Uptime

FERC’s 2026 large-load action has increased attention on flexible service and demand-responsive loads. ASHRAE’s grid-interactive framework also identifies cooling optimization, pre-cooling and thermal storage as possible sources of demand flexibility.

That does not mean the entire cooling system can simply be turned off.

A credible flexibility plan should state:

• Maximum cooling load that can be reduced

• Response time

• Maximum event duration

• Maximum number of events

• Required recovery period

• Ambient-temperature limits

• IT-load conditions

• Redundant equipment that must remain available

• Effect on supply temperature

• Effect on service-level commitments

Possible strategies include pre-cooling, chilled-water or ice storage, temporary fan-speed adjustment, pump optimization, warmer coolant setpoints within approved limits and shifting non-critical computing workloads.

Flexibility must be measurable and repeatable. Do not promise a curtailment number that has not been validated through controls testing and thermal modeling.

12. Divide the Project Into Repeatable Capacity Blocks

A single campus-wide cooling system creates a large common failure domain and makes phased procurement difficult.

Instead, align cooling blocks with electrical and IT deployment blocks.

For example, a 100MW campus could be divided into five 20MW IT phases. Each phase could include dedicated or zoned:

• CDUs

• Facility-water pumps

• Heat-rejection equipment

• Electrical distribution

• Controls

• Leak detection

• Isolation valves

• Water-treatment interfaces

• Metering

The exact capacity block depends on the rack layout and redundancy strategy. The principle is to connect each equipment purchase to a specific amount of commissionable IT capacity.

This approach also allows the project team to complete FAT, deploy the first block, measure real performance and correct the design before repeating it.

13. Prepare the Cooling Procurement Package

Before requesting binding supplier quotations, complete the following documents:

Cooling Basis of Design

Define IT load, rack density, cooling method, supply and return temperatures, heat-capture ratio, site climate, water conditions, redundancy and performance targets.

Heat and Mass Balance

Show heat load, liquid flow, residual air load, water consumption and heat-rejection capacity for normal, peak and failure conditions.

Equipment Schedule

List CDUs, pumps, dry coolers, towers, chillers, air handlers, expansion systems, filtration and water-treatment equipment.

Hydraulic Diagram

Define facility-water loops, technology-water loops, pressure boundaries, isolation zones, bypasses, drains and fill points.

Electrical Load Schedule

Provide running power, standby power, starting current, voltage, frequency, power factor and harmonic data.

Sequence of Operations

Explain normal startup, partial-load control, lead-lag rotation, failure response, emergency shutdown and restart.

Interface Responsibility Matrix

State who supplies piping, cables, valves, sensors, network connections, structural supports, insulation, water treatment and onsite labor.

FAT and SAT Plan

Define performance tests, calibrated instruments, acceptance criteria, load simulation, communication tests and required reports.

Pro Tip: “CDU package,” “dry-cooler package” and “factory tested” are not sufficient procurement descriptions. Define the project boundary and acceptance criteria before comparing prices.

14. Align the Cooling RFQ With the Utility Application

The cooling RFQ and utility load letter should use the same baseline.

Check that both documents contain matching values for:

• IT capacity by phase

• Normal facility load

• Summer peak facility load

• Cooling redundancy

• Ultimate build-out

• Energization dates

• Voltage and frequency

• Backup-generation strategy

• Load-shedding capability

• Restart sequence

• PUE assumptions

When the cooling supplier designs around 100MW IT but the utility studies only 110MW total facility power, a summer design-day load can expose the mismatch.

Late corrections may require a larger transformer, additional feeder capacity, a revised substation or another utility study.

15. Calculate ROI From Time to Commissioned Power

Cooling ROI should not be measured only by equipment price.

A more useful model is:

Cooling TCO per commissioned IT MW =

Equipment + electrical integration + piping + logistics + installation + commissioning + energy + water + maintenance + spares + delay exposure
÷
Commissioned IT capacity

The financial impact of a power delay can be estimated as:

Delay exposure = unavailable IT MW × contribution per MW-day × delay days + financing and contract costs

A slightly more expensive modular cooling package may generate a better return when it shortens installation, reduces onsite rework and supports earlier energization.

By contrast, inexpensive equipment with incomplete electrical data or unclear interfaces may delay the utility application and cost more than the original purchase-price saving.

Procurement-Ready Cooling Checklist Before Applying for Power

Confirm the following before submitting the application:

• IT capacity by development phase

• Maximum rack power

• Server and GPU platform

• Liquid heat-capture ratio

• Residual air-cooling load

• Coolant temperature class

• Facility and technology-water temperatures

• Design flow and pressure drop

• Normal and summer peak PUE

• Cooling equipment running power

• Motor-start and restart loads

• Heat-rejection architecture

• Site dry-bulb and wet-bulb conditions

• Water availability and WUE target

• Cooling redundancy

• Maintenance operating mode

• Load ramp schedule

• Minimum critical load

• Curtailable load

• Backup-generation strategy

• Utility voltage and point of delivery

• CDU guaranteed performance

• Controls and BMS/DCIM protocols

• FAT, SAT and commissioning requirements

• Expansion interfaces for future phases

Final Procurement Verdict

Do not apply for data center power using an IT megawatt target and a hoped-for PUE.

Build the cooling basis first. Define the server heat load, liquid-cooling percentage, temperatures, flow, heat rejection, seasonal power, redundancy and restart behavior. Then convert that information into a utility load letter and phased ramp schedule.

The result does not need to be a final construction drawing. It must be detailed enough that the utility can study a credible facility load and cooling suppliers can provide comparable quotations.

Plan the cooling system before the power application. Procure the first capacity block before the promised energization date.

ACT-Boxes supports modular AI data center cooling infrastructure, including CDUs, dry coolers, liquid loops, power distribution, controls, factory testing, delivery and onsite commissioning coordination. A useful proposal begins with the phase-by-phase IT load, server coolant requirements, site climate, PUE/WUE targets, redundancy strategy and utility schedule.

FAQ

Should the data center apply for power before selecting the cooling architecture?

The final equipment model does not have to be selected, but the cooling architecture, temperature range, peak cooling power and redundancy strategy should be defined before submitting the utility load request.

Is the IT load the same as the utility service capacity?

No. Utility service capacity includes IT equipment and facility infrastructure such as cooling, electrical losses, pumps, fans and controls.

Which PUE should be used in the power application?

Use a defensible peak facility-load assumption based on design weather, operating mode and redundancy. Do not rely only on the expected annual-average PUE.

Can liquid cooling reduce the requested power capacity?

It may reduce cooling electricity when combined with suitable coolant temperatures and efficient heat rejection. However, liquid cooling does not automatically guarantee a lower PUE. Pumps, CDUs, residual air cooling and site conditions must be included.

What cooling documents should be sent to the utility?

Requirements vary, but useful documents include the load letter, phase ramp schedule, electrical load schedule, cooling operating modes, motor-start data, backup-generation strategy and demand-flexibility description.