How to Estimate the Power and Water Needs of a Hyperscale Data Center

There is no meaningful single number for a hyperscale data center’s power or water use. A defensible estimate reports four separate outputs: coincident peak facility demand in megawatts (MW), annual electricity in megawatt-hours or terawatt-hours (MWh/TWh), direct water consumed at the site, and the additional water consumed to produce the electricity. Build the estimate from the planned IT equipment and its hourly operating profile, then apply condition-specific facility and cooling assumptions.

The most important discipline is keeping unlike quantities separate. Peak MW determines grid-interconnection and infrastructure capacity; annual MWh measures energy consumed over time. Site water is not the same as the water footprint of the electricity supply. National studies provide context, but a campus estimate requires its own equipment plan, weather data, cooling design, PUE, WUE and electricity-supply assumptions.

#	Preview	Product	Price
1		Spanish -- Thermal Guidelines for Data Processing Environments, 4th ed.		Buy on Amazon
2		Thermal Guidelines for Data Processing Environments by TC9.9 MISSION CRITICAL FACILITIES...	$133.63	Buy on Amazon

As an Amazon Associate I earn from qualifying purchases.

What a complete estimate should report

Prepare a results sheet with at least these four lines:

• Peak facility demand (MW): the coincident maximum of IT load plus cooling, power-conversion and other facility loads under the stated design or operating condition.
• Annual facility electricity (MWh or TWh): the time-integrated consumption over a defined year, including seasonal and workload variation.
• Direct or site water (liters or cubic meters): water consumed at the facility, primarily for cooling, within a clearly stated withdrawal and consumption boundary.
• Indirect or source water: water consumed in producing the electricity used by the facility, calculated from the electricity supply’s water intensity rather than from water delivered to the data center.

These outputs answer different planning questions. A utility needs the peak MW and its duration; an energy buyer needs the annual MWh and load shape; a water planner needs seasonal direct consumption and the site’s source; and a sustainability report may need both direct and electricity-related water.

#1 Best Overall

Spanish -- Thermal Guidelines for Data Processing Environments, 4th ed.

Buy on Amazon

Start with the IT load, not the building

List the equipment that will actually be installed: general-purpose servers, high-power AI accelerators, storage, networking and any specialized appliances. For each class, record nameplate or measured operating power, idle power, expected utilization, quantity, operating lifetime and the refresh or growth schedule. AI accelerators can materially change the result because their power and utilization assumptions differ from those of conventional servers.

Build an hourly or sub-hourly load profile

For every time interval in the reporting period, estimate the number of active devices and their power at the expected workload. Include low-utilization and idle states, batch workloads, seasonal changes and planned expansion. A simple annual average hides the peaks that drive electrical infrastructure, while a nameplate sum can greatly overstate normal consumption.

• Average IT load: the mean IT power over the reporting period.
• Peak IT load: the highest coincident demand from the equipment population and workload.
• Ramp and duration: how quickly a workload reaches its peak and how long the peak persists, which affects UPS, generators, transformers and cooling response.

Use project-specific equipment and operating data whenever available. Lawrence Berkeley National Laboratory’s national method combines equipment shipment information, per-device annual electricity models, cooling simulations, facility types and locations; those national assumptions are not a substitute for a campus load profile.

Apply PUE correctly

Power Usage Effectiveness (PUE) is defined as:

PUE = total data-center electricity demand ÷ IT-equipment electricity demand

The ratio is dimensionless. It captures cooling and other overhead such as UPS losses, power transformation and distribution, fans and pumps. A PUE of 1.25 means that each 1 kWh delivered to IT requires 1.25 kWh for the whole facility under the stated conditions.

Annual energy calculation

For each interval, multiply IT energy by the PUE appropriate to that interval, or use an annual PUE that represents the facility and period:

Annual facility electricity (kWh) = annual IT electricity (kWh) × annual PUE

If the cooling system performs differently in summer and winter, a single annual ratio should be the weighted result of those conditions, not an arbitrary design value.

Peak demand calculation

For peak planning, estimate the coincident peak rather than multiplying an annual-average PUE by the peak IT value:

Peak facility MW = coincident peak IT MW × design or operating PUE at that condition

Cooling equipment, pumps, fans and power-conversion losses can be highest during a hot-weather peak. An annual average PUE therefore may not represent the condition used to size the substation or interconnection.

Separate direct water from source water

Site Water Usage Effectiveness (WUE) is commonly expressed as:

WUE = site water consumed (liters) ÷ IT electricity (kWh)

For a defined period:

Direct site water (liters) = IT electricity (kWh) × site WUE (L/kWh)

State whether the boundary covers only consumptive use or also withdrawals, make-up water, blowdown and wastewater. Cooling-tower systems consume water through evaporation and require blowdown to control dissolved solids. Weather, heat load, setpoints, cycles of concentration and operating practice all affect the result.

Indirect water from electricity generation

Electricity production can consume water at power plants. Estimate this separately by multiplying the facility’s electricity use by a water-intensity factor for the actual or modeled grid supply:

Source water (liters) = facility electricity (kWh) × electricity-supply water intensity (L/kWh)

The factor depends on the grid mix, generation technologies, plant cooling systems and the accounting method. Source water is not water drawn at the data-center site. If a report presents a combined footprint, label the two components and state the grid assumptions and time boundary before adding them.

Cooling architecture is the main water decision

Evaporative or cooling-tower systems

Cooling towers reject heat partly by evaporating water and also need blowdown. The U.S. Department of Energy Federal Energy Management Program explains: “A cooling tower system by necessity uses an extensive amount of water because the warm water from the chiller’s condenser water loop is cooled by evaporating water into the surrounding atmosphere.” Consumption rises with heat rejection, hot or dry weather and operating choices. A tower design can reduce mechanical-cooling energy, but its direct-water requirement must be modeled for local weather and the facility’s load.

Air-cooled systems

Air-cooled equipment can be relevant where minimizing site water is a priority. It may have a different energy and peak-power profile, especially in hot weather, so compare it using the same IT load, weather year, reliability criteria and reporting boundary. “Lower site water” does not automatically mean lower total electricity or lower source-water use.

Economizers and hybrid designs

Air-side economizing uses suitable outside air instead of mechanical cooling when temperature, humidity, contamination and air-quality conditions permit. Water-side economizing can bypass or reduce chiller operation. Availability depends on climate, control limits, filtration and the equipment’s allowable operating envelope.

Rank #2

Thermal Guidelines for Data Processing Environments by TC9.9 MISSION CRITICAL FACILITIES (2004-06-30)

$133.63

Buy on Amazon

Operational choices that change both energy and water

Temperature and humidity setpoints, airflow management and economizing influence cooling energy and tower water use. Separating hot and cold aisles limits mixing between server exhaust and supply air. DOE FEMP says the practices described in its guidance can enable higher chilled-water temperatures and reduced airflow, citing 20% less chiller energy from the referenced best-practices design guide. That figure is a cited chiller-energy result under the described practices, not a guaranteed whole-facility reduction.

• Containment and airflow: keep supply and exhaust paths separated, then verify fan speeds and control sequences.
• Setpoints: select temperature and humidity limits compatible with the installed equipment and the applicable ASHRAE Thermal Guidelines for Data Processing Environments, 4th Edition.
• Economizer controls: document the outdoor conditions at which economizing starts, stops or transitions to hybrid operation.
• Water treatment: quantify evaporation and blowdown. DOE FEMP cautions that side-stream filtration by itself does not reduce facility power or water use unless other technologies or operating changes reduce cooling demand.

Worked example: an illustrative campus estimate

The following is an arithmetic example, not a measured facility result. Assume a campus has a 100 MW coincident peak IT load, a design PUE of 1.30 at that peak, a 60 MW annual-average IT load, an annual PUE of 1.25 and a site WUE of 0.30 L/kWh.

1. Peak facility demand: 100 MW × 1.30 = 130 MW. This is the design-condition facility peak, not annual energy.
2. Annual IT electricity: 60 MW × 8,760 hours = 525,600 MWh.
3. Annual facility electricity: 525,600 MWh × 1.25 = 657,000 MWh.
4. Direct site water: 525,600,000 kWh × 0.30 L/kWh = 157,680,000 liters, or 157,680 cubic meters.

The example uses annual-average IT energy for WUE, as the definition specifies. A project should replace every assumption with measured or engineered values, test seasonal weather and workload scenarios, and calculate source water from the electricity supply separately.

How to choose and validate input data

Use a documented hierarchy of evidence. Measured rack or building data is preferable for an operating site; commissioning tests and vendor curves can support a new system; design ratings and nameplate values are appropriate for early planning but should be marked as such. Record the unit, time basis, operating condition and whether a value is measured, modeled or design-rated. Keep IT electricity and total-facility electricity on the same meter boundary so the PUE calculation is internally consistent.

For water, identify every make-up line, cooling-tower blowdown stream, treatment discharge and reclaimed-water source. A withdrawal limit and a consumptive-use estimate answer different permitting questions. Report monthly values when supply is seasonal: an annual total can conceal a summer shortage even if the arithmetic is correct. If reclaimed or municipal water is substituted, state that change without assuming it eliminates the underlying consumption.

Requirements for a decision-grade model

A utility or permitting review normally needs a coincident peak, expected peak duration, ramp rate, redundancy configuration and expansion stages. An annual sustainability inventory needs a full-year load shape, meter boundary, PUE and WUE definitions, source-water factor and reporting year. A water-supply study additionally needs monthly evaporation, blowdown, make-up demand, source reliability and the withdrawal-versus-consumption boundary.

Model each build phase separately when racks, cooling plants or substations come online at different dates. Show how a partial-load phase affects PUE; fixed pumps, fans and transformers can make early phases less efficient than the eventual campus. For AI deployments, include accelerator count, expected operating life, idle draw and utilization in every scenario. LBNL’s 2025 national sensitivity cases specifically vary equipment installations, specialized graphics-chip shipments, AI-chip operating life and AI-server idle power and utilization, illustrating why one point estimate can be misleading.

Compare sites or designs on the same axes

When evaluating alternatives, hold the IT workload and reporting period constant. A comparison that gives one site a peak PUE and another an annual PUE is not like-for-like.

Use low, base and high cases for equipment count, utilization, idle power, cooling performance, weather and grid assumptions. There is no universally best cooling option: the appropriate choice depends on climate, water constraints, operating requirements and engineering design.

What national figures can—and cannot—tell you

Lawrence Berkeley National Laboratory’s 2025 electricity update provides scale for the United States, not a design value for an individual campus. Its 2030 reference case estimates 649 TWh of data-center electricity, equal to 11.8% of total U.S. electricity use. The compounded uncertainty range is 521–843 TWh, or 9.5%–15.3% of total U.S. electricity use. The range reflects sensitivity to equipment installations, specialized graphics-chip shipments, AI-chip operating life and AI-server idle power and utilization.

The LBNL 2024 United States Data Center Energy Usage Report projects 60–124 billion liters of direct water in 2028 for U.S. hyperscale data centers. This is a national projection, not a per-campus allowance. It comes from an older report than the 2025 electricity update, so the two figures should not be presented as one synchronized forecast.

Turn the estimate into a planning model

1. Define the boundary and year: state whether electricity is IT-only or whole-facility, whether water is consumed or withdrawn, and which calendar or operating year is modeled.
2. Create the equipment register: list device types, quantities, power states, utilization and refresh dates.
3. Generate the IT profile: calculate hourly or sub-hourly average and coincident peak load, including idle and growth cases.
4. Select cooling and infrastructure assumptions: model chillers, towers, dry coolers, economizers, UPS, transformers, fans and pumps for the local climate.
5. Calculate PUE and WUE by condition: use seasonal or operating-condition values where performance changes, then aggregate to annual results.
6. Calculate source water: apply a documented water-intensity factor to facility electricity and identify the grid or contract represented.
7. Stress-test the result: run low, base and high cases for AI utilization, equipment growth, weather, cooling efficiency and grid mix.
8. Document every assumption: retain units, measurement basis, source year, boundary and whether each value is modeled, measured, design-rated or promotional.

Common estimation mistakes and fixes

• Confusing MW with MWh: report peak capacity in MW and energy over time in MWh or TWh; never convert a peak number into annual consumption without a load profile.
• Using annual PUE for the peak: use a coincident, design-condition PUE for interconnection planning.
• Counting site water as the whole footprint: add a separately labeled source-water calculation when the question concerns electricity-related water.
• Copying a national average: replace shipment-based assumptions with the project’s equipment, utilization, location and cooling design.
• Ignoring idle power and AI growth: model utilization, idle draw, accelerator deployment and refresh schedules explicitly.
• Reporting one weather condition: test hot, humid and representative seasonal conditions, because cooling performance and tower evaporation vary.
• Assuming filtration alone saves resources: identify the mechanism that changes heat load, cooling energy, evaporation or blowdown before claiming a reduction.
• Hiding the boundary: state whether water figures include only consumption, include withdrawals, or include wastewater and off-site electricity production.

Technical reference for operating limits

DOE FEMP points operators to the ASHRAE Thermal Guidelines for Data Processing Environments, 4th Edition for equipment operating-condition guidance. Treat that publication as a technical reference, not as a substitute for project engineering. Confirm the applicable edition, equipment warranties, control strategy and local requirements before fixing temperature or humidity limits.

Frequently Asked Questions

What is the fastest way to estimate a hyperscale data center’s peak power?

Estimate the coincident peak IT load from the equipment profile, then apply a PUE representing that same design or operating condition. Do not apply an annual-average PUE to a hot-weather peak.

How do I convert IT electricity into annual facility electricity?

Multiply annual IT electricity in kWh by an annual PUE that represents the facility and reporting period. If PUE varies materially by season, calculate each interval and sum the results.

What is the difference between direct and indirect data-center water?

Direct water is consumed at the facility, mainly for cooling. Indirect, or source, water is consumed at power plants producing the electricity and must be calculated from the electricity supply’s water intensity.

Does a lower-WUE cooling system always have a lower environmental footprint?

No. Air-cooled systems can reduce site water but may require more electricity, particularly in hot weather. Compare direct water, annual energy, peak demand, source water, reliability and local conditions together.

Why are AI workloads difficult to estimate?

Accelerator power, idle draw, utilization, operating life and deployment growth can differ sharply from conventional servers. Include those variables explicitly and test low, base and high cases.

What water boundary should a project disclose?

State whether the result covers consumptive use, withdrawals, make-up water, blowdown, wastewater and any off-site electricity-generation water. These boundaries answer different planning and reporting questions.

What national benchmarks are available for context?

LBNL’s 2025 update gives a 2030 U.S. data-center electricity reference case of 649 TWh, with a 521–843 TWh uncertainty range. Its 2024 report projects 60–124 billion liters of direct water in 2028 for U.S. hyperscale centers. Neither is a campus design input.

Can side-stream filtration alone reduce data-center water use?

DOE FEMP cautions that filtration alone does not reduce facility power or water use unless another technology or operational change reduces cooling demand, evaporation or blowdown.

The Bottom Line

Estimate hyperscale resource needs from the IT equipment and its time-varying load, then model facility overhead and cooling under the local climate. Report coincident peak MW, annual MWh/TWh, direct site water and electricity-related source water as separate, clearly bounded results. Use condition-specific PUE and WUE, test AI and weather uncertainty, and treat national figures as context rather than as a substitute for project engineering.

Quick Recap

Bestseller No. 1

Spanish -- Thermal Guidelines for Data Processing Environments, 4th ed.

Bestseller No. 2

Thermal Guidelines for Data Processing Environments by TC9.9 MISSION CRITICAL FACILITIES (2004-06-30)

$133.63

Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.