Someone says a 100 MW AI data center uses “as much electricity as a city.”
That sounds dramatic.
But which city?
A city of 100,000 people?
A city full of offices and hospitals?
A manufacturing city?
And are we comparing peak power or annual electricity?
The useful comparison is not “How many people live there?” It is “How much electricity does the city actually use in a year, under a comparable boundary?”
By the end of this article, you will be able to read a “100 MW data center” headline, translate it into annual electricity use, compare it with a real city, and spot when the comparison is misleading.
Using the same model as the previous article, a 100 MW IT data center with an 85% average load factor and PUE of 1.30 uses about 0.97 TWh per year.
Now we can compare that number with citywide electricity data.
The comparison is about annual electricity, not population. A city and a data center also place very different shapes of demand on the grid.
First, Rebuild the 0.97 TWh Number
The model starts with 100 MW of maximum IT capacity.
100 MW × 0.85 load factor × 1.30 PUE = 110.5 MW average facility load
Then:
110.5 MW × 8,760 h = 967,980 MWh ≈ 0.97 TWh/year
The previous article, How to Understand a 100 MW Data Center Power Model, explains why IT capacity, load factor and PUE must be kept separate.
Here we use the resulting 0.97 TWh as an intuition-building benchmark.
The Contexta City Comparison Integrity Test
Before comparing a data center with a city, check four things.
- Same quantity: annual electricity vs annual electricity—not MW vs TWh.
- Known boundary: city proper or local authority, not an undefined metro area.
- Comparable coverage: domestic and non-domestic consumption should be clear.
- Comparable year: same-year data are much stronger than mixing old and new city totals.
This matters because a city comparison can be numerically correct and still be misleading if the boundaries or years are inconsistent.
The Cleanest Comparison: Six UK Cities, One 2024 Dataset
The strongest comparison in the original article is the six UK local authorities because they come from the same 2024 Department for Energy Security and Net Zero dataset.
That dataset reports domestic and non-domestic metered electricity consumption at local-authority level. The official 2024 release remains the latest available as of September 30, 2026; DESNZ says 2025 data are due in December 2026.[1]
| Local authority | 2024 electricity | 0.97 TWh data-center share |
|---|---|---|
| Bristol | about 1.63 TWh | 59% |
| Liverpool | about 1.65 TWh | 59% |
| Sheffield | about 1.90 TWh | 51% |
| Manchester | about 2.40 TWh | 40% |
| Leeds | about 2.86 TWh | 34% |
| Birmingham | about 3.74 TWh | 26% |
So the same illustrative AI data center would use annual electricity equal to roughly:
- three-fifths of Bristol or Liverpool
- half of Sheffield
- two-fifths of Manchester
- one-third of Leeds
- one-quarter of Birmingham
That is a much stronger statement than saying “it powers a city of X people.”
Population alone does not tell us how much electricity a city uses.
Why Population Is a Weak Shortcut
A public data-center discussion asked whether a fixed formula could turn data-center MW into an “equivalent city population.”
The problem is that cities with similar populations can have very different electricity demand.
Hospitals matter.
Universities matter.
Offices matter.
Factories matter.
Transport and digital infrastructure matter.
Housing stock and climate matter.
The relevant denominator is the electricity actually measured inside the chosen boundary—not population by itself.
The Contexta Same Energy, Different Grid Shape Map
There is another reason the city analogy has limits.
Annual energy can be similar while the shape of demand is completely different.
| City load | Large AI data-center load |
|---|---|
| Spread across homes, offices, shops, hospitals and industry | Concentrated at one campus |
| Distributed across many feeders and substations | May require dedicated high-capacity connections and substations |
| Demand varies across many independent users | Large correlated load can persist for long periods |
| Expansion occurs through many separate decisions | Campus expansion can add another large block of MW |
That creates the key infrastructure lesson:
The grid challenge is not only producing 0.97 TWh over a year. It is delivering roughly 100 MW or more to one location whenever the compute load needs it.
This is why annual-energy comparisons build intuition, while grid planning still has to focus on location, connection capacity, peak conditions and reliability.
What the Existing Map Shows—and What It Does Not
The six UK comparisons use the same 2024 DESNZ dataset. Boston remains on the legacy map as an older cross-country reference and should not be read as an equal-year comparison.
The map is useful for scale.
It is not a ranking of cities.
It also does not mean the data center would literally replace the electricity system of a city.
The comparison only asks:
What share of this city’s annual electricity equals the model’s 0.97 TWh?
Boston Is Useful as Context—but It Is an Older Comparison
The original article used an official 2016 Boston electricity figure of about 6.50 TWh, making the modeled data center roughly 15% of that total.
That comparison is still useful as an order-of-magnitude reference.
But it should not sit on equal footing with the six UK 2024 comparisons.
Boston’s current climate page reports that both residential and commercial electricity use fell from 2022 to 2023, but the page does not present the same simple citywide TWh figure used in the original comparison.[4]
So this rebuild keeps Boston as a clearly labeled older cross-country reference rather than pretending all seven city numbers are synchronized.
The Bar Chart Is an Intuition Tool, Not a League Table
Read the UK bars as the matched-year comparison. Read Boston separately because its source year is older.
A smaller percentage does not make one city “more efficient.”
A larger percentage does not make another city “wasteful.”
City electricity totals reflect economic structure, building stock, local boundaries and many other factors.
The Official Dataset Has Limits Too
The 2024 DESNZ local-authority statistics are accredited official statistics based on meter-point data.[1]
But the methodology matters.
Electricity consumed directly from onsite generation is not captured by the meter-based dataset, and Central Volume Allocation users—large industrial consumers connected through the transmission system under different arrangements—are excluded from the subnational dataset.[2]
That is another reason to use words such as about, roughly and under this model.
Change the Model and Every City Percentage Moves
The city comparison inherits all the assumptions from the data-center model.
If PUE falls from 1.30 to 1.20, annual facility electricity falls.
If average load factor rises above 0.85, annual electricity rises.
If the 100 MW number turns out to describe total facility demand instead of IT capacity, the entire calculation changes.
The AI Data Center Power Cost Calculator in Python lets readers change those assumptions directly.
The Contexta City-Scale Load Test
When you see a claim such as “this data center uses as much electricity as a city,” ask:
- What data-center boundary is being used?
- Is the comparison power or annual energy?
- What load factor and PUE are assumed?
- Which city boundary is being used?
- Are the city data from the same year and methodology?
- Does the comparison acknowledge concentrated vs distributed load?
If those six answers are visible, the analogy can be useful.
If they are hidden, the city comparison may be more rhetorical than analytical.
What Should You Watch Next?
- 2025 UK local-authority data: DESNZ says the next release is due in December 2026.
- 100 MW vs 1 GW: How quickly does city-scale intuition break when campuses grow by an order of magnitude?
- Load concentration: How much transmission and substation capacity is needed at one site?
- Grid-cost allocation: Who pays for upgrades triggered by large new loads?
- Local acceptance: How do communities weigh electricity, water, noise, jobs and tax revenue?
- Global scale: How many 100 MW-equivalent AI campuses would 2030 electricity forecasts imply?
The Main Idea
A 100 MW AI data center sounds abstract.
Under the model used in this series, it becomes about 0.97 TWh per year.
Against a matched 2024 UK local-authority dataset, that is roughly:
59% of Bristol, 59% of Liverpool, 51% of Sheffield, 40% of Manchester, 34% of Leeds, and 26% of Birmingham.
But the deeper point is not the percentages.
It is the combination of scale and concentration.
A large AI data center can use city-scale annual electricity while concentrating that demand at a single site. The annual-energy analogy builds intuition; the concentration explains the grid problem.
Continue Reading
- How to Understand a 100 MW Data Center Power Model — see how the 0.97 TWh number is built.
- AI Data Center Power Cost Calculator in Python — change PUE, load factor and price yourself.
- How Many 100 MW AI Data Centers Will the World Need by 2030? — move from one campus to global scale.
Key Terms
- citywide electricity consumption: electricity recorded within a defined city or local-authority boundary
- local authority: an administrative geographic area used for local government and official statistics
- concentrated load: large electrical demand located at one site rather than distributed across many customers
- distributed load: electricity demand spread across many buildings, customers and network connections
- annual energy: the total electricity used over a year, usually expressed in MWh, GWh or TWh
- meter-point data: electricity-consumption data aggregated from customer metering systems
- measurement boundary: the geographic or technical boundary defining which electricity use is included
- Central Volume Allocation user: a large UK electricity user with transmission-level arrangements that are excluded from the DESNZ subnational meter dataset
- city-scale load: a single load large enough that its annual electricity can be compared with a significant share of a city’s total
Sources
- UK Department for Energy Security and Net Zero — Regional and local authority electricity consumption statistics — official 2024 local-authority electricity dataset.
- DESNZ — Subnational methodology and guidance — coverage, onsite-generation and Central Volume Allocation limitations.
- DESNZ — Subnational electricity and gas consumption summary report 2024 — meter-based methodology and current release context.
- City of Boston — Boston’s Carbon Emissions — current citywide inventory context and 2023 electricity-use trend.
Status checked September 30, 2026. The City Comparison Integrity Test, Same Energy–Different Grid Shape Map, and City-Scale Load Test are The Contexta analytical frameworks. UK city totals are rounded values from the 2024 DESNZ local-authority dataset. Boston is retained only as an older cross-country reference. The 0.97 TWh data-center value is an illustrative model output based on 100 MW IT capacity, 0.85 load factor, and PUE 1.30.