You read that a company plans to build a 100 MW AI data center.
What exactly is 100 MW?
Is that what the servers use?
Is it what the whole building draws from the grid?
Is it an amount of electricity used in one year?
Or is it simply the maximum capacity the project hopes to support someday?
A “100 MW data center” is not a complete energy statement until we know the boundary, the load factor, the PUE, and the time period.
That is why the same 100 MW label can lead to several very different numbers.
This article translates the label step by step—without requiring Python.
The first job is not calculation. It is deciding what the 100 MW number refers to.
Why “100 MW” Causes So Much Confusion
Public discussions about data-center power repeatedly run into the same problem.
People ask whether a data center can really “use 100 MW.”
Others immediately point out that MW is power, not energy.
Then another distinction appears: the 100 MW figure might describe IT nameplate capacity, utility connection capacity, contracted demand, or total facility demand.
These are related numbers.
They are not interchangeable.
So before doing arithmetic, we need a boundary check.
The Contexta Boundary Check
Whenever you see a data-center MW number, ask:
- Where is the measurement boundary?
IT equipment, the whole facility, or the utility connection? - Is it maximum capacity or average operating load?
A project can be designed for 100 MW without averaging 100 MW every hour. - Is the number already adjusted for cooling and electrical overhead?
If not, PUE still has to enter the model. - Is this power or energy?
MW and MWh answer different questions.
Only after those four questions are clear does the calculation become meaningful.
The Contexta 100 MW Translation Ladder
For this article, we define the 100 MW as maximum IT capacity.
Then we translate it through four stages.
100 MW IT Capacity → Average IT Load → Average Facility Load → Annual Energy
We will use four assumptions:
| Input | Value | Role |
|---|---|---|
| IT capacity | 100 MW | Maximum electrical capacity assigned to IT equipment |
| Average load factor | 0.85 | Average IT load as a fraction of maximum IT capacity |
| PUE | 1.30 | Facility energy relative to IT energy |
| Time | 8,760 h/year | Converts average power into annual energy |
These are illustrative assumptions.
The model is useful because each one can be replaced.
Step 1: MW Is Power, Not Energy
A megawatt measures a rate of energy use at a moment in time.
A megawatt-hour measures an amount of energy used over a period.
| Unit | What it measures | Simple analogy |
|---|---|---|
| MW | Power at a moment | Speed |
| MWh | Energy over time | Distance traveled |
A facility drawing 100 MW for one hour uses 100 MWh.
If it averages 100 MW for 8,760 hours, it uses 876,000 MWh in a year.
But our 100 MW figure does not mean the facility averages 100 MW.
It means the IT system can support up to 100 MW in this model.
Step 2: Load Factor Turns Maximum IT Capacity into Average IT Load
The IT system does not have to draw its maximum electrical capacity every hour.
So we introduce an average load factor.
100 MW × 0.85 = 85 MW average IT load
This is an electrical-load assumption.
It is not the same as saying every GPU is “85% utilized.”
Workload utilization, chip activity and electrical draw are related, but they are not identical concepts.
The IEA’s 2026 Energy and AI data product also treats installed capacity, PUE, load factor and electricity consumption as separate variables.[1]
Step 3: PUE Moves Us from IT Energy to Facility Energy
Servers are not the only things using electricity.
The facility also powers cooling, pumps, fans, UPS systems, power conversion, controls and other support systems.
The U.S. Department of Energy defines Power Usage Effectiveness as:
PUE = Total Facility Energy ÷ IT Equipment Energy
DOE also emphasizes that PUE is an infrastructure-efficiency metric. It does not tell us how valuable or productive the computing work is.[2]
In this screening model, we use a representative PUE of 1.30.
Applying it to the average IT load gives:
85 MW × 1.30 = 110.5 MW average facility load
This result can look strange at first.
The data center is called “100 MW,” yet the average whole-facility load is 110.5 MW.
The reason is that the 100 MW label refers to maximum IT capacity, while the 110.5 MW result includes facility overhead.
The Contexta Capacity Multiplier
There is a compact way to see the relationship.
Average Facility Load ÷ IT Capacity = Load Factor × PUE
For this example:
0.85 × 1.30 = 1.105
Then:
100 MW × 1.105 = 110.5 MW
This multiplier explains an important point:
A 100 MW IT facility can average more than 100 MW at the utility boundary if load factor × PUE is greater than 1.
If the load factor were lower, the result could go the other way.
For example, 0.70 × 1.20 = 0.84, so a 100 MW IT design would average 84 MW at the facility level under those assumptions.
The label alone does not tell us the answer.
Step 4: Convert Average Facility Power into Annual Energy
Now we have an average facility load of 110.5 MW.
Multiply by time:
110.5 MW × 8,760 h = 967,980 MWh/year
That is:
967.98 GWh ≈ 0.97 TWh/year
This is the annual energy result for the assumptions in this article.
It is not a universal number for every 100 MW data center.
Step 5: Add a Price Only After the Energy Model Is Clear
If we use an illustrative electricity price of $70/MWh:
967,980 MWh × $70/MWh = $67,758,600/year
That is about $67.8 million per year.
But this is a screening calculation, not a utility bill.
Large-load tariffs can include demand charges, minimum commitments, grid-related charges, taxes, escalation clauses and other contract terms.
That more detailed tariff problem belongs in the calculator model and later project-finance layers.
The Contexta Unit Sanity Test
Before accepting any data-center power claim, run four quick checks.
| Claim | Ask |
|---|---|
| “The data center is 100 MW.” | IT capacity, facility demand, or utility connection? |
| “It uses 100 MW per year.” | Is energy being confused with power? Annual energy should use Wh units. |
| “PUE is 1.2.” | Measured over what boundary and time period? |
| “Load factor is 85%.” | Electrical load factor or application/GPU utilization? |
Most confusion disappears once the unit and boundary are explicit.
Why PUE Needs Careful Reading
PUE is useful because it tells us how much facility energy surrounds the IT energy.
But it should not be treated as a universal score for data-center quality.
DOE recommends considering multiple metrics because total data-center performance can involve water, carbon, utilization, reliability and other factors in addition to PUE.[2]
Berkeley Lab’s 2025 Update also models PUE as one component of a broader bottom-up estimate of data-center electricity demand rather than as a complete performance measure.[3]
That is why our 1.30 value should be read as an example input, not as a claim about the “correct” PUE for AI data centers.
Three Common Mistakes
1. Treating IT capacity as whole-facility demand
If 100 MW is IT capacity, PUE still has to be applied before estimating facility energy.
2. Treating MW as annual energy
MW is power. MWh, GWh and TWh are energy.
3. Treating load factor as GPU utilization
Electrical load factor is a power-model input. It does not directly tell us whether every GPU is busy or whether the compute investment is being used efficiently.
Why This Model Matters Beyond the Arithmetic
The purpose of the model is not to memorize 0.97 TWh.
It is to make large data-center announcements easier to read.
When you see “100 MW,” “500 MW,” or “1 GW,” you can ask:
- Which boundary?
- Maximum or average?
- What load factor?
- What PUE?
- What time period?
- How much energy does that imply?
Only then can we move to the next question:
How large is that amount of electricity compared with something people can picture?
What Should You Watch Next?
- IT vs facility MW: How do companies define capacity in project announcements?
- Load factor: Are AI workloads actually operating near the assumed average?
- Measured PUE: How different is annual operating PUE from design PUE?
- Rack density: Does higher AI rack density change cooling and facility overhead?
- Utility connection: Is the announced MW already secured from the grid?
- City-scale intuition: What does 0.97 TWh look like compared with the annual electricity use of a real city?
The Main Idea
A 100 MW data center is not one number.
It is the beginning of a chain.
Boundary → IT Capacity → Load Factor → PUE → Average Facility Load → Time → Energy
For the assumptions in this article:
100 MW × 0.85 × 1.30 = 110.5 MW average facility load
110.5 MW × 8,760 h ≈ 0.97 TWh/year
The number becomes useful only after we know what it measures, where it is measured, and over what period.
Continue Reading
- AI Data Center Power Cost Calculator in Python — turn the same model into a reusable calculator.
- Why Power, Not Chips, May Limit the AI Data Center Boom — see why deliverable power and connection time matter.
- How Much Power Is 100 MW? An AI Data Center Compared with Entire Cities — translate 0.97 TWh into city-scale intuition.
Key Terms
- IT capacity: the maximum electrical capacity assigned to servers, storage and networking equipment in this model
- facility load: IT load plus cooling, electrical conversion and other supporting infrastructure
- load factor: average electrical load divided by maximum or rated capacity over a period
- PUE: Power Usage Effectiveness, total facility energy divided by IT-equipment energy
- MW: megawatt, a unit of power
- MWh: megawatt-hour, a unit of energy
- GWh: gigawatt-hour, equal to 1,000 MWh
- TWh: terawatt-hour, equal to 1,000 GWh
- measurement boundary: the point around a system that defines which equipment or energy flows are included in a number
- nameplate capacity: a rated or designed maximum capacity rather than an average operating value
Sources
- IEA — Key Questions on Energy and AI data product — installed capacity, PUE, load factor and electricity-consumption variables.
- U.S. Department of Energy — Best Practices Guide for Energy-Efficient Data Center Design — PUE definition and broader efficiency metrics.
- U.S. Department of Energy / Lawrence Berkeley National Laboratory — United States Data Center Energy Usage Report: 2025 Update — current U.S. data-center energy-modeling context.
Status checked September 30, 2026. The Boundary Check, 100 MW Translation Ladder, Capacity Multiplier, and Unit Sanity Test are The Contexta analytical frameworks. The 100 MW, 0.85 load factor, 1.30 PUE, and $70/MWh values are illustrative assumptions rather than universal data-center benchmarks or project quotations.