The reliable way to estimate a hyperscale data center’s demand is to model four separate quantities: coincident peak facility power (MW), annual electricity use (MWh or TWh), direct water consumed at the site (usually reported with site WUE in L/kWh), and the water consumed to generate the electricity (source or indirect water). Start with the planned servers, accelerators, storage and networking equipment, apply realistic utilization and idle-power profiles, model cooling and electrical overhead, and state every weather, operating and accounting assumption.
A peak-MW estimate sizes grid connections, substations and backup systems. An annual-energy estimate measures consumption over time. Direct site water is not the same as the broader water footprint of the electricity supply. Keeping those boundaries separate prevents the most common planning errors.
As an Amazon Associate I earn from qualifying purchases.
Define the outputs before doing the arithmetic
A useful estimate has four headline outputs. Publish them together, but never combine them into one ambiguous “power and water” number.
Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Repair Windows errors before they cause bigger problemsFix Now →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →| Output | Unit | What it answers | Primary uses |
|---|---|---|---|
| Coincident peak facility demand | MW | How much electrical capacity the campus may require at its worst coincident operating condition | Utility interconnection, substations, generation, transformers, UPS and emergency generation |
| Annual facility electricity | MWh or TWh per year | How much electricity the facility consumes over the reporting period | Energy procurement, emissions accounting, utility forecasts and operating budgets |
| Direct/site water consumption | Liters or cubic meters per year; often summarized as L/kWh of IT energy | How much water is consumed at the campus, primarily by cooling in the LBNL boundary | Water permits, municipal planning, drought and basin-impact analysis |
| Source/indirect water consumption | Liters or cubic meters per year | How much water is consumed in producing the electricity used by the facility | Fuller electricity-supply water footprint and grid-option comparisons |
Report the direct and source quantities separately. A site WUE value describes water consumed at the facility; it does not include water consumed at power plants. Conversely, source water is not water drawn from the data-center site.
Use an assumption-led workflow
The following sequence works for a new campus, an expansion, or an operating facility with measured data. Replace generic assumptions with project data as soon as it becomes available.
- Set the boundary and period. Specify whether the result covers one building, a campus, or an entire colocated site; identify the calendar or fiscal year; and state whether water means consumption, withdrawal, or both.
- Inventory IT equipment. List server types, CPUs, GPUs or other accelerators, memory, storage, network devices and planned quantities. Record nameplate power and, where available, measured power at idle, typical and maximum utilization.
- Create an hourly or sub-hourly IT load shape. Model workload variation, seasonal effects, batch jobs, training runs, maintenance and growth. Keep average IT load, design IT load and coincident IT peak as different fields.
- Convert IT load to facility electricity. Apply a PUE appropriate to the operating condition. Use an annual PUE for annual energy and a design or coincident peak PUE for peak demand.
- Model cooling and other overhead. Include chillers, cooling towers, pumps, fans, air handlers, economizers, UPS losses, transformers and distribution losses. Simulate setpoints and local weather rather than copying a national average.
- Calculate direct water. Select a cooling-specific WUE for the climate and operating pattern, then multiply it by IT electricity for the same period and boundary.
- Calculate source water separately. Multiply electricity consumed from each supply or grid region by that supply’s water-intensity factor. Document the generation mix and the factor’s accounting method.
- Run scenarios and review the result. At minimum, vary utilization, equipment count, AI-accelerator power, cooling performance, weather and water availability in low, base and high cases.
Build the IT-load model first
IT electricity is the denominator for both PUE and site WUE, so an error here propagates through every result. A credible inventory contains more than a rack count.
Equipment and power fields
- Quantity by server and accelerator model or class.
- Typical and maximum device power, plus idle power.
- Memory, storage and networking power that may not scale linearly with compute utilization.
- Deployment date, operating lifetime and refresh schedule.
- Expected utilization by workload: interactive services, batch analytics, AI training, inference and standby capacity.
- Redundancy capacity and equipment that is powered but normally idle.
High-power AI accelerators can change both the average and peak profile. Treat their operating life, idle behavior and utilization as explicit assumptions. LBNL’s 2025 Update demonstrates why: its national sensitivity cases vary equipment installations, specialized graphics-chip shipments, AI-chip operating life and AI-server idle power and utilization.
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Fix the driver behind crashes, sound loss and screen glitches3Clear out junk files and repair common Windows errorsFrom device profiles to IT energy
For each equipment class, calculate an hourly load such as:
IT power(t) = quantity × [idle power + utilization(t) × (typical power − idle power)]
Use a more detailed curve when measured telemetry is available. Sum all classes, then identify:
- Average IT power: the mean over the reporting period.
- Design IT power: the load used to size a room or electrical block.
- Coincident IT peak: the highest load expected to occur at the same time across the facility.
Annual IT energy is the time integral of that load. With hourly data, sum each hour’s kW multiplied by one hour; with a constant average, multiply average MW by the number of hours in the period. Do not multiply a nameplate maximum by 8,760 hours unless the equipment is actually expected to run at that maximum continuously.
Free tools Windows power users keep installed
One-click scans. No signup required.
Apply PUE correctly for peak and annual electricity
Lawrence Berkeley National Laboratory defines power usage effectiveness as “the total electricity demand of the data center divided by the electricity demand of the IT equipment.” In symbols:
PUE = total facility electricity demand ÷ IT equipment electricity demand
PUE is dimensionless. It captures non-IT electricity such as cooling, UPS conversion, power transformation and distribution, fans and pumps. The value can change with load, weather, maintenance state and economizer operation.
Annual electricity
For a reporting period:
Annual facility energy (kWh) = annual IT energy (kWh) × annual PUE
Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteWindows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallUse an annual or time-weighted PUE derived from the same period. If PUE varies materially by hour or season, calculate facility energy hour by hour rather than multiplying by one average value.
Peak facility demand
For design or operating peak:
Peak facility MW = coincident peak IT MW × PUE at that condition
An annual-average PUE may not represent the peak. A hot afternoon can raise chiller and tower loads while a high-utilization workload raises IT demand. Use a coincident design PUE or a time-series simulation that captures both effects.
Illustrative calculation
Suppose a hypothetical campus has a coincident IT peak of 80 MW and a peak-condition PUE of 1.30. Its estimated peak facility demand is 104 MW. If its average IT load is 50 MW, annual IT energy at 8,760 hours is 438,000 MWh. With an illustrative annual PUE of 1.30, annual facility electricity is 569,400 MWh. These figures are an arithmetic example, not a benchmark or an estimate for a particular operator.
Estimate direct water with a stated site boundary
Site water usage effectiveness is commonly expressed as:
Site WUE (L/kWh) = liters of water consumed at the facility ÷ IT electricity (kWh)
Therefore:
Direct site water (L) = site WUE (L/kWh) × IT energy (kWh)
Use IT energy, not total facility energy, in the denominator when following the LBNL definition. State whether the numerator includes only cooling consumption or also potable uses, humidification, leaks and other site processes. The 2024 LBNL report’s national modeling treats site water separately from source water associated with electricity generation.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Cooling-tower systems
Evaporative towers consume water as heat is rejected to the atmosphere and also require blowdown to control dissolved solids. As 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.” Actual consumption depends on heat load, tower cycles of concentration, weather, setpoints, makeup-water quality and operating practice.
Air-cooled and hybrid systems
Air-cooled systems can reduce direct site-water demand where water availability is constrained, but they may use more electricity during hot conditions or require a different equipment footprint. Hybrid designs can switch between evaporative and dry modes. Compare them using modeled peak MW, annual MWh, direct water, reliability and equipment operating limits rather than choosing on WUE alone.
Illustrative WUE calculation
Using the hypothetical 438,000,000 kWh of annual IT energy above and an illustrative site WUE of 1.2 L/kWh gives 525,600,000 liters, or 525,600 cubic meters, of direct water for the year. The result changes linearly with IT energy and WUE; it is not a prediction for a real site.
Keep source water separate from site water
Electricity generation can consume water even when a data center uses little water on site. To estimate that component, divide the facility’s electricity by supply or grid region and apply an appropriate water-intensity factor:
Recommended Free Tools
Source water (L) = electricity supplied (kWh) × generation water intensity (L/kWh)
Document whether the factor represents withdrawal or consumption, which generation technologies it covers, and the year and geography of the electricity mix. If a campus contracts multiple supplies, calculate each share separately. A combined water footprint can be reported as direct site water plus source water only after both boundaries and units are stated.
Inputs that most change the result
| Input | Why it matters | How to represent it |
|---|---|---|
| Server, storage and network inventory | Sets the base IT load and the scale of growth | Quantity, device class, power curve, deployment and refresh dates |
| AI accelerator mix | Can raise rack density, average power and coincident peaks | Accelerator count, power range, utilization and idle behavior |
| Workload utilization | Separates a mostly idle fleet from a continuously busy one | Hourly or sub-hourly profiles by workload |
| Cooling architecture | Drives PUE, WUE, reliability and temperature limits | Chillers, towers, dry coolers, liquid loops, economizers and redundancy |
| Weather and location | Changes economizer hours, chiller lift and evaporation | Hourly temperature, humidity, air quality and design extremes |
| Setpoints and airflow control | Influences fan, chiller and tower operation | Supply temperature, humidity band, containment and control sequences |
| Electrical infrastructure | Adds losses outside the IT rooms | UPS mode and efficiency, transformers, distribution, fans and pumps |
| Water accounting boundary | Determines what the WUE numerator means | Consumption versus withdrawal; cooling-only versus all site uses |
| Electricity supply | Determines indirect water | Grid mix, contracted generation and water-intensity factors |
Use national figures as context, not design inputs
National forecasts illustrate the scale and uncertainty of the sector, but they cannot replace a project model.
| Figure | What it represents | Qualification |
|---|---|---|
| 649 TWh in 2030 | LBNL 2025 Update reference case for U.S. data-center electricity | National modeled value, not a campus forecast |
| 521–843 TWh in 2030 | LBNL compounded uncertainty range around the reference case | Reflects scenario uncertainty in equipment, AI and operating assumptions |
| 11.8% of total U.S. electricity in 2030 | LBNL 2025 Update reference-case share | The scenario range is 9.5%–15.3%; it is not a measured current share |
| 60–124 billion liters in 2028 | LBNL 2024 projection of direct/site water for U.S. hyperscale data centers | National projection, not a per-campus value; it comes from an older report than the 2025 electricity update |
The 2025 electricity figures and the 2024 hyperscale-water projection should not be presented as one synchronized forecast. Use each for the year and boundary it actually describes.
Compare candidate sites and cooling designs on the same axes
When evaluating alternatives, hold the IT workload and reporting period constant, then compare the following fields.
| Comparison axis | Questions to answer |
|---|---|
| Capacity and energy | What are coincident peak MW, average MW and annual MWh? Is the peak based on the same equipment plan as the annual case? |
| Cooling performance | What PUE and site WUE result under local weather, design-day conditions and annual operation? |
| Water boundary | How much water is consumed on site, and how much source water is associated with the electricity supply? |
| Water availability | What are the permitted source, seasonal limits, drought restrictions and treatment requirements? |
| Reliability and operating envelope | Can the system maintain required server inlet temperatures and humidity during equipment failures and extreme weather? |
| Scenario resilience | How do low, base and high utilization, equipment growth, cooling efficiency and weather cases change the ranking? |
No cooling architecture is universally best. Climate, water constraints, operating requirements, reliability design and the facility’s engineering details determine the outcome.
Operating practices that can lower energy or water
Contain airflow and manage setpoints
Separating hot aisles from cold aisles limits mixing between server exhaust and supply air. DOE FEMP says the cited practices can enable higher chilled-water temperatures and reduced airflow; its summary of the referenced FEMP design guide reports 20% less chiller energy under the described air-management practices. Treat that as a reported chiller-energy result, not a guaranteed 20% reduction in whole-facility electricity or water. Model the control sequence, fan response and local climate.
Use air-side and water-side economizing where conditions permit
Air-side economizers use suitable outside air in place of mechanical cooling. Water-side economizers bypass or reduce chiller operation through a heat exchanger or related loop. Their value depends on temperature, humidity, air quality, filtration, control limits and the number of hours available. Include their effects in the hourly PUE and WUE model rather than applying a fixed annual credit.
The Tool Desk
Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Check the mechanism of water interventions
A measure should be credited only if it changes heat load, cooling energy, evaporation or blowdown. DOE FEMP cautions that side-stream filtration by itself does not reduce facility power or water use without other technologies or operating changes that reduce cooling demand. Filtration may support water quality and cycles of concentration, but the model must show how those changes alter makeup water.
Build low, base and high cases
At minimum, vary the assumptions that have the largest leverage:
- Low case: slower equipment deployment, lower utilization, favorable weather and efficient cooling operation.
- Base case: the approved equipment schedule, expected workload profile and design weather year.
- High case: faster growth, higher accelerator utilization, hotter or more humid conditions, lower equipment efficiency and restricted economizer operation.
For each case, report peak facility MW, annual facility MWh, direct water and source water. Add sensitivity columns for PUE, WUE and grid water intensity so a reviewer can see whether the conclusion is driven by IT demand, cooling assumptions or the electricity supply.
Quality checks before publishing or submitting an estimate
- Confirm that peak MW is not being labeled as annual MWh or TWh.
- Check that peak PUE reflects the same coincident condition as peak IT load.
- Verify that annual energy uses an annual or time-weighted PUE, not a nameplate maximum.
- Ensure WUE uses IT electricity in the denominator and that the numerator’s boundary is explicit.
- Keep direct/site water and source water in separate columns.
- Convert liters and cubic meters consistently: 1,000 liters equals 1 cubic meter.
- Document the weather file, setpoints, economizer rules and cooling redundancy.
- Record whether water figures mean consumption, withdrawal or both.
- Reconcile modeled loads with utility bills, submeter data and commissioning measurements when available.
- Flag any national average or scenario range as context rather than a site design value.
Common estimation failures and how to correct them
Using rack count as a power estimate
Rack count does not reveal accelerator density, utilization or idle power. Replace it with an equipment inventory and load profile.
Applying one PUE to every condition
Cooling and electrical losses vary with weather and load. Use separate annual and peak-condition values or an hourly model.
Calling site WUE the total water footprint
Add a separately calculated source-water component only when the electricity mix and water-intensity boundary are documented.
Multiplying maximum device power by all hours
This overstates annual energy when equipment idles or workloads fluctuate. Use measured or assumed utilization curves.
Quoting a national projection as a campus requirement
The LBNL values of 649 TWh, 521–843 TWh, 11.8% and 60–124 billion liters describe U.S. scenarios and projections. A project still needs its own equipment, cooling, weather and water analysis.
Claiming a universal cooling winner
Dry, evaporative, hybrid and liquid-assisted approaches trade water, energy, reliability and climate performance differently. Compare them on identical loads and local operating conditions.
Technical references for design teams
DOE FEMP points to ASHRAE Thermal Guidelines for Data Processing Environments, 4th Edition for equipment operating-condition guidance. Use it as a technical reference and verify the applicable edition and availability for the project; it does not substitute for current engineering analysis.
For a defensible submission, preserve the input workbook or model, assumptions register, weather data, equipment schedule, PUE and WUE derivations, source-water factors, scenario outputs and review sign-offs. That audit trail is as important as the headline numbers because changing one assumption—especially accelerator utilization, cooling mode or water boundary—can materially change the result.
The practical verdict is straightforward: estimate IT load first, separate coincident peak from annual energy, apply condition-appropriate PUE, calculate direct site water with a cooling-specific WUE, and model source water as a distinct electricity-supply effect. A number without those boundaries is not a dependable hyperscale data-center estimate.
The Tool Desk
Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Quick Recap
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.
