With IT load held at 10 MW for every hour of a 365-day year, PUE 1.2 gives 105.12 GWh of annual facility energy; PUE 1.4 gives 122.64 GWh. The difference is 17.52 GWh. These are calculated scenarios, not observed consumption at a real data center. If 10 MW is installed capacity rather than continuous demand, annual energy needs another assumption about how that capacity is used.
Start with the same IT-energy boundary
PUE compares total facility energy with IT-equipment energy over the same reporting period. The US Department of Energy’s PUE explanation expresses it as an annual-energy ratio. Rearranging the definition gives facility energy = IT energy × PUE. Comparing two PUE values is meaningful here because both scenarios use exactly the same IT energy.
Our base case assumes 10 MW of constant IT input, 8,760 hours, and hypothetical annual PUE values of 1.2 and 1.4. It excludes changes in the amount of IT work, changes to the IT equipment and differences in the reporting boundary. Neither PUE value is claimed to be typical, guaranteed or achievable by a particular cooling product.
The 10 MW, full-year calculation
Convert megawatts to kilowatts before calculating kWh: 10 MW = 10,000 kW. IT energy is 10,000 × 8,760 = 87,600,000 kWh, or 87.60 GWh. Multiply that number by each PUE; subtract IT energy from facility energy to find the non-IT portion.
| Annual quantity | PUE 1.2 | PUE 1.4 |
|---|---|---|
| IT energy | 87.60 GWh | 87.60 GWh |
| Total facility energy | 105.12 GWh | 122.64 GWh |
| Non-IT energy | 17.52 GWh | 35.04 GWh |
| Average facility power over this year | 12 MW | 14 MW |
The difference is 87.60 × (1.4 − 1.2) = 17.52 GWh. The average powers in the final row are annual facility energy divided by 8,760 hours. They are not a calculation of the maximum cooling-day demand or the short-duration electrical peaks that supply equipment may need to support.
Why this is not a 20% reduction in the total bill
PUE falls by 0.2, but the percentage depends on the denominator. Moving from the 1.4 case to the 1.2 case reduces total facility energy by 17.52 ÷ 122.64 = approximately 14.29%. The same change cuts the modeled non-IT energy from 35.04 to 17.52 GWh, which is 50%.
Likewise, at PUE 1.2 the overhead is 20% of IT energy but approximately 16.67% of total facility energy. At PUE 1.4 those figures are 40% and approximately 28.57%. A claim such as “20% overhead” should identify whether it means a share of IT energy or a share of the whole facility.
If average IT electrical load is 70%
Now treat 10 MW as the modeled IT envelope and assume the annual average electrical load is 70% of it. IT energy becomes 10,000 × 0.70 × 8,760 = 61.32 GWh. Holding the same assumed annual PUE values gives the following sensitivity case.
| Annual quantity at 70% average load | PUE 1.2 | PUE 1.4 |
|---|---|---|
| IT energy | 61.320 GWh | 61.320 GWh |
| Total facility energy | 73.584 GWh | 85.848 GWh |
| Non-IT energy | 12.264 GWh | 24.528 GWh |
The modeled difference is now 12.264 GWh, not 17.52 GWh. The 70% input is an electrical average-to-envelope ratio, not a GPU utilization percentage. It also does not establish that power infrastructure can be sized for only 70% of the IT envelope. Energy forecasting and capacity planning answer different questions.
Where a constant PUE assumption becomes weak
A selected annual PUE can be useful for a sensitivity exercise, but it should not hide seasonal cooling changes, partial-load behavior or a changing facility boundary. The DOE’s 2024 data center design guide, section 8.1, explains why annual measurement matters and why PUE does not capture the efficiency of useful IT work.
When interval energy data is available, add total facility energy across intervals and divide by the sum of IT energy. Do not simply average interval PUE values when the IT energies differ. For example, 1 MWh of IT energy at PUE 1.4 and 3 MWh at PUE 1.2 produce (1 × 1.4 + 3 × 1.2) ÷ 4 = 1.25 overall, not the simple average of 1.3. Google’s operator PUE reporting illustrates the use of energy-weighted reporting across its published periods.
Translate the scenario into your own inputs
Use the PUE calculator when you have facility and IT energy totals. Use the AI power calculator to build a server-based planning envelope, then test annual electrical-load assumptions separately. Keep the original IT load and PUE distinct when continuing to the capacity planner.
For an energy-only cost comparison, open the electricity cost calculator and enter 2,000 kW, 8,760 operating hours and 100% average load for the full-load difference. Change only average load to 70% for the lower-load case, and enter your own energy tariff. These inputs reproduce differences of 17,520,000 kWh and 12,264,000 kWh respectively before multiplication by the tariff. A time-varying tariff needs interval calculations. Demand charges, taxes, fixed fees and investment cost are not represented by this multiplication. The useful result is a transparent sensitivity range, not a promised saving or a complete project business case.
Sources & further reading
- US DOE: Cooling Water Efficiency Opportunities for Federal Data Centers ↗
- US DOE: Best Practices Guide for Energy-Efficient Data Center Design, July 2024 ↗
- Google Data Centers: Power Usage Effectiveness ↗