October payrolls rise; earlier gains revised up

Other source limits: Review update: No material numerical change found in the four national figures and existing state/local records reviewed October 29. IEA's October policy review does not replace the U.S. water baseline. March 2026 water analysis retained. Regional figures remain 2025 estimates, not today's meter readings; FY2027 revenue is a forecast, not collected revenue. Grid costs: protective terms are not measured household savings. SCC lacks municipal responses; generator permits are not measured emissions. Virginia's guidance retains exceptions; a final rule published October 5 takes effect November 4. Added DEQ's October 23, 2026 ambient air record; it does not isolate generator emissions or establish illness. DOE's authority is not measured operation or harm. PJM's October 19 advisory, issued October 16, is not renewal. October 15's unused-backup account stays dated. Oregon's adopted-rate counterweight, not measured household savings. The working paper's 2015–2024 data is not a verified local bill decrease.

October 19 addition: national scale, local receipts. PwC estimates $84.2 billion in U.S. state and local tax contribution for 2024, including suppliers and household spending. This May 2026 industry-commissioned model is not an audited total of taxes paid by facility owners or net public benefit. Loudoun's reported fiscal 2026 collections remain a separate local record. See the state/local split, all 50 states and D.C., and the limits →

November 6 update: water and jobs, national first. The cards now show Berkeley Lab's 2023 on-site water model and BLS's preliminary October 2026 sector payroll estimate. State, regional and county records keep their own periods, definitions and gaps. Water evidence → Jobs evidence →

They use real power and can bring real investment. Neither the developer's promise nor the scariest forecast tells you what happens on your block. Start with the records.

Review update: November 9: No material change found in the national electricity, tax, water and jobs figures or published geographic numbers reviewed. All 52 tax-model rows and 56 QCEW records were rechecked. Electricity and taxes retain their 2024 model periods; national on-site water retains 2023. October 2026 sector payrolls remain 454,800, preliminary and seasonally adjusted, from the November 6 release. March 2026 QCEW geography remains separate. PJM warning versions and its dated winter outlook were reviewed; warnings are not outages or backup-diesel operation. DEQ’s November 6 report still ends daily fine-particle averages November 5 and records a November 4 sensor move. Its live values were not audited. Health-literature search cutoffs are unchanged.

November 5 outlook: adequate supply is a forecast, not a guarantee. PJM expects enough resources for expected winter conditions, while warning that reserve margins are tightening and exceptionally high generator outages could leave the system vulnerable. Its higher-stress analysis includes demand-response programs. PJM also says it has never had to use its authority to order large loads onto backup generation. That operator statement is not proof that every generator was idle or that nearby air was unaffected. Read the November 5 outlook and its limits.

November 5 update: Pennsylvania narrows the proposal. The PUC’s clarification stops seeking comments on PJM’s rejected IRAS mechanism while continuing its state emergency-curtailment proposal. The revised appendix keeps limited protection for separately identified essential real-time services when backup fails, fuel runs out or permit limits are reached—not a blanket data-center exemption. This is not a final shutoff rule or a record of outages or household savings. Read the November 5 clarification. Inspect the revised proposed appendix.

November 4 addition / study published October 19. A systematic review of 36 studies found environmental pressures and modeled health impacts, but no observed human-health outcomes in the studies it included. Database searches ended July 31, 2026; supplementary Google Scholar searching was August 10. Those are search dates, not new measurements or proof of safety. The authors warn that some facility-level records may have been missed. Read the risk guide and evidence limits →

November 6 grid check: warnings are not outages. PJM's November 3 DOM transmission warning now records an end time of 00:56 on November 4. A separate DOM warning started at 06:22 on November 4 and now records an end at 21:06 on November 5. November 3 messages explicitly labeled DRILL describe a winter exercise, not actual emergency dispatch. These notices do not establish household outages, backup-diesel operation or a data-center contribution. November 3 warning and end time ↗ · November 4 transmission warning ↗ · Explicitly labeled winter drill ↗

October 14 follow-up: Oregon's governor requests a time-limited permit pause, not an operating-facility shutdown or measured harm. Read the October 9 letter's scope and tradeoff →

October 15 follow-up: PJM's October 14 version 4 list adds opacity to one resource's constraint column. A listed constraint is not a pollution reading or proof of operation. Inspect the dated revision →

October 16 review: PJM's October 15 account reports no use of large-load backup authority and adds preliminary maintenance-outage context. This is the operator's dated account, not proof every generator was idle or that neighbors faced no pollution. Read the updated operating account →

October 16 follow-up: PJM issued a capacity advisory for October 19, citing expected maintenance and construction outages. It warns of tightening capacity, not an inevitable household outage. Forecast demand and unavailable capacity are different measures. The advisory does not itself extend DOE's October 13–16 order. Read the new advisory and retained operating record →

October 19 review: a Pennsylvania proposal, not a shutoff rule. A notice published October 17 carries an October 1 tentative order seeking comments on data-center curtailment rules. Its appendix gives limited protection to separately identified digital services essential to critical facilities when backup fails, runs out of fuel or reaches permit limits. It is not a blanket "all data centers first" rule. The IRAS provisions depend on federal approval. FERC's October 9 order accepted a registry but rejected PJM's linked capacity and curtailment package. The inspected state docket lists a tentative order, not a final rule. Neither record proves an avoided outage or lower bills. Read the notice and appendix. Inspect the state docket. Read the earlier federal decision.

October 18 follow-up: a Pennsylvania cost-allocation dispute. Shapiro's October 15 petition seeks to join the PPL case. PPL's October 16 response cites existing safeguards and a proposed transmission rider. The inspected docket lists an active case. These competing positions are not a new rate ruling or measured household savings. Read both sides and the filing dates.

October 17 follow-up: DOE's inspected October 13–16 operating window has ended; reporting and compliance duties continue. The October 19 advisory is not a renewal. We also added a Stafford noise-baseline case: its April 30 report covers January–February measurements, not operating-facility noise or illness. Read the evidence and its limits →

October 16 correction: Our generator guide described the earlier guidance without the later final rule. Virginia published that rule October 5, with a November 4 effective date. Its text covers data centers applying for permits on or after July 1, 2026. These requirements are not a measurement of pollution reduction. Read the rule and its dates ↗

October 15 follow-up: Texas's attorney general announced water-reporting letters to data-center developments. The announcement is not a water-use measurement, a finding of harm or an adjudication of violations. The letters and responses were not inspected here. Read the office's announcement ↗

Read the full community ledger →

What harm is actually documented? → Noise, pollution, water and land: compare what is recorded with what a claim still needs to prove.

October 13 follow-up: FERC approved a large-load definition and registry, not the linked bring-new-capacity and mandatory-curtailment package. Read what cleared the gate →

October 12 follow-up: PJM received temporary backup-generation authority for October 13–16. Permission is not a measurement of operation or harm. Read the order and its limits →

October 13 update: PJM's revised resource list adds oil-fired units. Its October 12 submission, posted October 13, is not a dispatch or emissions log. Inspect the updated list →

October 15 correction: PJM's updated account reports demand-response actions on October 13 and 14, not use of large-load backup generation as of that account. Earlier we named only October 13. Separate the action from the permission →

Open the source-linked snapshot

MODELED ESTIMATE · UNITED STATES

How much electricity?

Berkeley Lab estimates that U.S. data centers used 192 terawatt-hours in 2024, or 4.7% of national electricity use. That is data centers generally, not an AI-only meter reading or a forecast of your bill.

Read the Berkeley Lab report ↗

REPORTED EXAMPLE · VIRGINIA

How many jobs last?

Virginia's legislative auditors say a typical 250,000-square-foot center may have about 50 full-time workers, according to industry representatives. A building site can have far more workers at its construction peak. Those are different jobs on different clocks, not a prediction for every proposal.

Read the Virginia review ↗

ESTIMATED · REGIONAL WATER

What about water?

Virginia's statewide withdrawal share—below 0.5% in 2023—is not a local safety verdict. ICPRB estimates the Washington-area data-center share at 1% of withdrawals but 9% of annual consumptive use for 2025. Those are different denominators. Cooling design, summer demand and return flows belong in the local plan.

Virginia audit ↗ Regional water analysis ↗

SCENARIO · NOT A HOUSEHOLD BILL

Who pays for new power?

EIA's higher-demand exercise projects possible regional supply and wholesale-price pressure. It does not tell you what one household will pay. That answer needs the local utility's rate case, cost-allocation rules and the project's actual load.

A 2026 working paper estimates lower average rates from 2015–2024 data-center growth relative to a no-growth alternative, not a verified fall in every household's bill. Local terms still matter.

PGE's July 8 adopted change lowered average residential rates 1.3% and raised data-center rates about 29%. Rates are not bills; other adjustments are included. See the Oregon comparison →

Examine the EIA scenario ↗

November 6 air update: DEQ’s new report extends daily fine-particle averages through November 5, 2026. It records a November 4 sensor move from Broad Run High School to Discovery Elementary; site coverage is not continuous or uniform. These ambient measurements do not identify data-center generators’ contribution or establish illness. Read the new report’s scope and limits →

2026 state receipts: Virginia reports about $26.0 million in initial remittances. SCC's October 20 account concerns July–August electricity use; payments lag use and billing. Finance's separate September-only record is about $25.0 million. Neither is an annual total. Do not add them to each other, the 2024 national model or Loudoun's different tax and period. Read the official records and their scopes →

RECORDED LOCAL REVENUE · LOUDOUN COUNTY, VIRGINIA

The benefits can outlast the building crew.

Loudoun County reports $1.2 billion in data-center property-tax revenue in fiscal 2026, about 39% of its budget. The county says that revenue helped fund schools, public safety, libraries and parks while easing some tax rates. That is a real local example, not a promise every community can cash. The county also warns against depending too heavily on one revenue source.

Read the full local ledger ↗

Developer promises and neighbors’ concerns get the same test. These sources answer different questions; none proves the net effect of a particular project. How we check claims ↗