The Power Plant Next Door: the data-center series

We broke it into nine parts because the buildout is really several different arguments that usually get argued as one. Seven of the parts each take one of them: the bill, the water, the jobs, the veto, the claims on both sides of the debate, the grid queue, and a checklist to take to a hearing. Then comes the editor’s opinion, and last a part of rechecked figures on bills and tariffs. Each part answers its own question, so read straight through or start with the argument you are having in your own town. The summaries below give one finding from each part, and the table after them sets a question, a source and a part side by side. Each part checks its own claims against its own sources and carries its own dates, and this page only summarizes them.

Parts 1 to 7

Part 1 starts on your electric bill. PJM’s independent market monitor attributes $29.4 billion of capacity charges, 46.2 percent of the total, over the last four auctions to data-center load. That is a with-and-without comparison, not a count of dollars moved from one household to another. In the federal table of average residential prices for August 2025 to August 2026, the median change was +9.8 percent in the eight places wholly or mostly in PJM and +4.6 percent in the 37 states outside it (Part 1 also shows six partly-PJM states separately: 8 + 6 + 37 = 51), and New Jersey, in PJM, fell 0.4 percent. A rise in PJM states does not by itself show data centers caused it. Pennsylvania’s default-service generation prices, which changed on June 1, 2026, moved from +0.05 percent to +19.8 percent across 11 utilities (median +6.9 percent).

Part 2 follows the water. Google’s 2024 table puts The Dalles, Oregon at 461.1 million gallons withdrawn and 361.4 consumed, and the city’s records, as OPB reported them, list 434.4. The part computes the gaps and shows why Google’s withdrawal, discharge and consumption and the city’s billed-use records are different measures that a single total hides.

Part 3 counts the jobs. Data center representatives told Virginia’s legislative auditors that a typical 250,000-square-foot data center may have “approximately 50” full-time workers, about half of them contractors; the auditors’ own estimate was over 8,000 data center jobs in the state in fiscal 2023. The part sets construction crews, permanent jobs and tax-break costs side by side in their own units, with the questions that turn a jobs promise into a number a county can hold a developer to.

Part 4 asks who holds the veto. Nobody keeps one official count of local pauses and bans, and the published counts run from at least 76 (National League of Cities, August 28, 2026) to 632 jurisdictions with concrete action toward a pause, which GatherGov defines to include a staff study as well as an ordinance (GatherGov, September 10), because each counts something different. Part 4 lists each count with its date, and shows who can say no in seven places. New York’s Executive Order 62 of July 14, 2026 was the first statewide pause, covering state environmental permits for data centers that consume or can consume 50 megawatts or more, according to Executive Order 62. Part 4 also lists the 41 local measures on the November 3 ballot. No federal moratorium has passed.

Part 5 sets real statements from both sides beside the records. It uses the federal AI Action Plan of July 2025, the presentation a company gave Guadalupe County, Texas commissioners on April 21, 2026 and the public comments at the same meeting, and tests each against the county’s agreements. It also reports the PJM market monitor’s cost-shift figures with PJM’s warning that its $16.4 billion auction total is not the total cost to customers. The cost-shift evidence is strongest inside PJM, where the market monitor’s figures are, and it thins out in markets with different utilities and fuel costs.

Part 6 checks what a grid-queue figure actually measures. Berkeley Lab’s 61-month median is for generation and storage projects built in 2025, not for data centers. This part gives a five-row records test a resident can use at a hearing and lists what Virginia’s GS-5 order says about contract length, exit fees, collateral and how much contracted capacity a customer can shed.

Part 7 is the scorecard. It lists ten questions on power, rates, water, jobs and what happens if a company leaves, with a place to record whether each answer is known, forecast, pending or not disclosed. It is filled in from the documents Guadalupe County, Texas posted for its April 21, 2026 vote: a signed abatement agreement, a development agreement whose posted copy is unsigned, and the company’s application and presentation, which describe the same project three ways. Queue position, a noise limit and the owner of the substation are in neither agreement. A checklist is a starting point for inquiry, not a verdict about a developer.

Parts 8 and 9

Part 8 is where the editor says what he thinks. Build the data centers away from people’s houses, and make each one bring its own new power so it does not draw down the plants we already have.

Part 9 closes with the rechecked figures. They include August 2026 against August 2025 residential prices in three PJM states and Florida, PJM’s 2026/2027 capacity price of $329.17 per megawatt-day, 11.4 times the $28.92 of the 2024/2025 auction (329.17 divided by 28.92), and five large-load tariffs priced for one hypothetical 25-megawatt data center, from 5.44 to 9.43 cents per kilowatt-hour (E3, May 2026), and OPB’s March 31, 2026 report that Amazon agreed to pay $20.5 million to settle nitrate-contamination claims in eastern Oregon. Those are modeled and reported figures, not a substitute for checking the sources.

Where to start

New to the subject? Start with Part 1, about the electric bill, and read forward, in the order they were written. In a hurry because your own county has a proposal on the agenda? Go straight to Part 7, use the hearing questions, and come back for the parts that explain why each one matters. Either way the point is the same. The cloud is a building. It has an address, a meter and a water line, and local approval procedures depend on where it is proposed.

Update, October 29, 2026: The part summaries were rechecked against their articles. Corrections: the Part 4 summary now gives the dated counts (76 to 632) in place of an unsourced “more than 100”; the Part 2 and Part 3 summaries match their articles, with the 50-worker figure attributed to data center representatives speaking to Virginia’s legislative auditors; the Part 1 summary describes the all-state comparison; and the Part 6 summary and definitions no longer treat generation-queue waits as data-center service waits. Added: the series name in the opening, a table pairing each question with its source and part, and the six partly-PJM states in the Part 1 summary. Later on October 29: the GatherGov count is now dated September 10, the date of GatherGov’s own page, which is linked (September 16 is the date a trade site republished it), and the summary says GatherGov counts staff studies in its 632; the $329.17, $28.92 and 5.44-to-9.43 figures link to their sources at the claim; the source list gains those sources; the Part 5 summary now describes the rebuilt article, which uses real statements in place of composite speeches; the Part 7 summary reflects the signed abatement agreement the county’s minutes attach; the Part 1, Part 4 and Part 9 summaries gain Pennsylvania’s price changes, the November 3 ballot count and the Amazon settlement; and the decoder lost two entries no part uses (advanced packaging and billable demand floor) along with the TSMC source.

Editor written. AI used for spelling and grammar only.