What happens when a data center moves to town?

Add a 300 MW data center to Prairie Town’s busiest day and its whole year, count the homes it equals, the wait to connect, and who pays

F8 of Energy Systems, Step by Step: add a 300 MW data center to Prairie Town’s demand on a summer and a spring day and over a year, compare it with homes, read real requests to connect, time the wait for grid service, and see who decides, pays and benefits.
  • Foundations · L1
  • 15 min guided · about 10 min practice
  • After lesson 2 and F6

The idea in one table: a 300 MW data center joins Prairie Town (illustrative)

The town's demand on two real-shape days, and with the campus's steady 240 MW (this lesson's assumption) added to every hour.

Day Town's busiest hour Town's quietest hour With the campus: busiest and quietest Campus share of the day
29 July 2025 450 MW 298 MW 690 MW and 538 MW 39%
17 April 2025 300 MW 216 MW 540 MW and 456 MW 48%
  • Its contract. Up to 300 MW: on 29 July 2025 the busiest hour could reach 750 MW.
  • Homes. 240 MW all year is 2,102,400 MWh: as much as about 195,000 average U.S. homes buy in a year (10,791 kWh each), and 85% of the town's own use.
  • The year. With the campus, demand is above the town's 2025 record in 8,690 of 8,760 hours; the load factor rises from 63% to 76%.
  • Requests. ERCOT counted about 474 GW of large loads asking to connect (June 2026), 5.2 times its record peak of 91,134 MW: proposals, not committed demand.
  • The wait. Built in 24 months, connected after 36: 12 idle months.
  • Cooling. Evaporative heat rejection: uses water that evaporates in a cooling tower; less electricity for cooling. Dry, air-cooled heat rejection: little operational water on site; may need more electricity, depending on the weather and the design.
  • A closed loop. A closed water loop inside the building can still send its heat out through an evaporative cooling tower (U.S. Department of Energy).
  • Jobs and taxes. 1,000 workers for 24 months make 2,000 job-years, then 50 permanent jobs; $1,000M of property at 1.0% pays $10.0M a year in local tax (illustrative).
  • Who pays. If the utility builds wires for the campus and the company cancels, the cost can land on other customers, unless a large-load tariff makes the company pay.

With JavaScript on, you work through this step by step: the day, a spring day, the homes, the year, the requests, the wait, water, jobs and taxes, and who pays.

  • Illustrative. Prairie Town and its data center are fictional: the 300 MW contract, the steady 240 MW, the 24-month build, the 36-month wait, the jobs, the taxes and the cooling choices.
  • Assumed. The campus draws a steady 240 MW in every hour; real campuses vary with their computing work and with cooling.
  • Real hourly patterns. The town's demand follows the Southwest Power Pool's 2025 hours (EIA-930, accessed 2 November 2026), scaled so its highest hour is 450 MW, as in lesson 2.
  • Real numbers. Homes from EIA (2022 data) and requests from ERCOT (June 2026), each dated in the sources below.

Need a refresher? Lesson 2 reads demand curves and load factors; F6 follows power from the plant to the plug.

What you just did, as diagrams

A load, not a town
A large load is one customer that draws as much as a town: here, a steady 240 MW against Prairie Town's busiest hour of 450 MW.
Flat, around the clock
Assumed steady here, it adds as much at night as at the busiest hour, so the town's quiet hours rise most in share.
Energy, not peak
240 MW all year is as much electricity as about 195,000 average U.S. homes buy in a year, but that compares energy, not power in any hour.
Requests, not demand
ERCOT's requests to connect were 5.2 times its record peak in 2026: they are proposals, not committed demand.
Who pays
The utility builds first; a large-load tariff decides how much the campus pays, and how much other customers do.
Go deeper · AnalystMW of computers, MW from the grid
  • Power usage effectiveness (PUE) is a data center’s total electricity divided by its computing equipment’s electricity.
  • For example, 200 MW of computers at a PUE of 1.2 need 200 × 1.2 = 240 MW from the grid.
  • Sizes in MW-IT count the computers only; the grid must plan for the whole campus.
  • Cooling works harder on hot days, so a real campus is not perfectly flat; this lesson keeps it flat to show the idea.
Go deeper · AnalystRequests, forecasts and ranges
  • A request is a place in line, not a building: ERCOT’s 474 GW of requests is 5.2 times its record peak of 91,134 MW.
  • A range, not a number. Berkeley Lab’s scenarios put U.S. data centers at 325 to 580 TWh in 2028, 6.7% to 12% of U.S. electricity; planners study the whole range.
  • Read next. Why connecting takes years: Analyst lesson A6. Who pays for a wait: case file CF7.
Go deeper · AnalystAdding a flat load to a load-duration curve

A flat load of \(c\) MW shifts every hour up by \(c\), so the load factor becomes

\[\text{LF} = \frac{\bar d + c}{d_{\max} + c} = \frac{281.9 + 240}{450 + 240} = 0.756\]

  • \(\bar d\) is the town’s average hour of 2025 (MW) and \(d_{\max}\) its busiest hour (MW); a flat load pushes LF toward 1.
  • A load that is not flat must be added hour by hour before sorting: the busiest hours of the sum need not be the busiest hours of either part.
  • Next. The Grid Room sandbox (SB2) adds a large load to a regional grid; resource adequacy (A5) asks whether supply is enough in every hour.
Go deeperWhat this lesson leaves out
  • Whether the wires can carry the campus: voltage, lines and stability are Analyst lesson A17.
  • Which plants run for it, and at what price: the merit order is lesson 8.
  • The emissions it causes or may claim: Analyst lesson A10.
  • Shifting some of its work in time: flexible demand is A7.
  • Real campuses' changing draw, and the real numbers for water, jobs and taxes, which differ by place.

Check yourself

Practice, about 10 minutes: three myths, then five questions.

Myth or fact? “A data center uses its power mostly at night, so it does not raise the busiest hour.”

  • Myth
  • Fact
Answer

Myth. It runs around the clock: in this lesson it adds 240 MW in every hour, so the town's busiest hour on 29 July 2025 rises from 450 to 690 MW.

Myth or fact? “Requests to connect show how much new demand will arrive.”

  • Myth
  • Fact
Answer

Myth. ERCOT's requests were about 474 GW in June 2026, 5.2 times its record peak. Requests are proposals, not committed demand: projects may change or be cancelled.

Myth or fact? “1,000 people building a campus for two years means 1,000 permanent jobs.”

  • Myth
  • Fact
Answer

Myth. That is 2,000 job-years, which end when building ends; the campus then needs its own, smaller permanent staff.

A data center averages 120 MW all year. About how many average U.S. homes (each buying 10,791 kWh a year) buy as much electricity in a year?

  • About 9,700
  • About 97,000
  • About 120,000
  • About 970,000
Answer

About 97,000: 120 MW × 8,760 h = 1,051,200 MWh, ÷ 10.791 MWh a home = 97,415.

At 03:00 on 17 April 2025, Prairie Town draws 216 MW. Add the campus's 240 MW. Is the total above the town's 2025 record of 450 MW?

  • Yes: 456 MW
  • No: 216 MW
  • No: exactly 450 MW
Answer

Yes: 216 + 240 = 456 MW, above the record, at the quietest hour of a mild day.

ERCOT counted about 474 GW of large-load requests; its record peak is 91,134 MW. About how many times the peak is that?

  • About 2
  • About 5.2
  • About 10
  • About 52
Answer

About 5.2: 474 GW = 474,000 MW, and 474,000 ÷ 91,134 = 5.2.

$800M of taxable property, at a local tax rate of 1.5%, pays how much a year?

  • $1.2M
  • $12M
  • $120M
Answer

$12M: 800 × 1.5% = 12. It is a transfer from the owner to local budgets.

Which term of a large-load tariff makes a customer that cancels pay toward wires built for it?

  • An exit fee
  • A time-of-use price
  • Net metering
Answer

An exit fee; a minimum bill does the same while it stays connected (A4).

Numbers, sources and what’s next

Numbers to remember

Number What it means Source
about 4.4% Share of U.S. electricity used by data centers in 2023 (about 176 TWh). U.S. Department of Energy (Berkeley Lab report) · published 20 December 2024; data for 2023
6.7–12% Berkeley Lab's range for data centers' share of U.S. electricity in 2028 (325 to 580 TWh): a range of scenarios, not one forecast. U.S. Department of Energy (Berkeley Lab report) · published 20 December 2024
about 474 GW Large loads asking to connect to ERCOT's grid in Texas, about 90% of them data centers: requests, not demand. ERCOT, update to the Texas Senate, slide 6 (PDF) · 29 July 2026; data as of June 2026
91,134 MW ERCOT's all-time peak demand, on 22 July 2026 (preliminary). ERCOT, all-time peak demand records · page updated 21 August 2026
10,791 kWh Electricity bought by an average U.S. home in a year. EIA, Frequently Asked Questions · data for 2022; page updated 8 January 2024

Keep going

Back to the course map

In the real world

Waiting for grid service costs money, and the cost grows with every month.

For a 300 MW-IT facility, a 36-month delay costs the developer on the order of $0.7 billion in present value.

Staged energization can recover a large share of that cost.

Approved excerpt of: Li et al. (2026), accepted paper, 2026 North American Power Symposium (NAPS); DOI pending.

MW-IT counts computing equipment only; this lesson's 300 MW is the whole campus's grid contract. Present value states future costs in today's dollars.

The full case: case file CF7 (what waiting for a grid connection costs, and who pays).

Sources and further reading

Spotted an error, or stuck? Write to hlli@ksu.edu. Corrections are made in place and dated here.

Text licensed CC BY 4.0, code MIT: teachers and students may reuse and adapt it, with credit to Honglin Li.