Who said it, and who paid them

Who paid, from the report's own pages: the work was done by Lawrence Berkeley National Laboratory with support from the Department of Energy's Industrial Technologies Office, and the lab is operated for the DOE by the University of California under contract DE-AC02-05CH11231. Two details matter more than the funding line. The input data was provided by three commercial research firms, Omdia Research, S&P Global and the International Data Corporation, so the raw numbers come from private datasets rather than public ones. And the acknowledgements thank experts from organizations representing major data center equipment manufacturers and data center owners who answered questions and commented on the assumptions used in the analysis, so industry helped shape the modeling inputs even though it did not pay for the report. This is still the closest thing to a neutral scorekeeper and the safest number to quote on either side, because the funder has no stake in the answer and the method is published. Know the two caveats anyway. One caution on the numbers: the 2030 estimate is a range, 578 to 782 TWh on the report's single-factor sensitivities and 521 to 843 TWh on its wider compounded uncertainty range, and the authors give a range precisely because a single confident number would not be honest. Do not quote the middle figure as a prediction.

Official federal policy documents issued by the Trump administration on July 23, 2025. Cite them as authoritative statements of what current U.S. policy is, not as research. Their economic assumptions are contested. Anti side: the provision worth targeting is the one exempting these projects from standard federal environmental review, because that review is the step where local objections would normally get heard.

Official federal policy documents issued by the Trump administration on July 23, 2025. Cite them as authoritative statements of what current U.S. policy is, not as research. Their economic assumptions are contested. Anti side: the provision worth targeting is the one exempting these projects from standard federal environmental review, because that review is the step where local objections would normally get heard.

Official communications from the county that has benefited from data center taxes more than anywhere else on earth, published partly to defend a policy that is locally controversial. The tax rate history and revenue figures are public budget documents and are not seriously disputed by anyone. The framing around them is a government advertising its own decisions. Anti side: do not attack the numbers, you will lose. Attack how unusual Loudoun is, and how risky it is to depend on one industry for 39 percent of a budget.

The data center industry's trade association and lobbying arm. Its president, Josh Levi, has led it since 2019, after nearly twenty years at a Northern Virginia technology trade group where he ran policy and founded its political fundraising committee. He testifies regularly before Congress and federal energy regulators. This is advocacy, full stop. Its large national figures for jobs and economic contribution come from studies it paid an accounting firm to produce, and they use a counting method that includes every downstream effect, which inflates the total compared to the careful studies. If you use these numbers, expect to defend the method. You are better off leading with Loudoun County, where the figures are public records.

Two economists. Dany Bahar teaches international and public affairs at Brown University and is also a fellow at the Brookings Institution, a long-established research organization in Washington. Greg Wright teaches economics at the University of California, Merced. Neither one specializes in data centers, and that is exactly why this study is trusted: they applied ordinary economics to the question without a stake in the answer. Their clever move was comparing counties that got a data center against counties where one was announced and then canceled, which compares similar places rather than comparing boomtowns to nowhere. Use in debate: this is the one study both sides can honestly cite. It limits what the pro side can claim about jobs, and it hands the anti side its best evidence that the tax breaks are aimed at the wrong buildings. Brookings put that disclosure on the article itself, which is more than most institutions do, and the money is unrestricted rather than tied to this study. Both facts belong in the same breath: the four biggest data center builders fund the institution, and they did not commission this paper.

No dedicated bibliography paragraph exists for this source in the handbook's "Who these sources are, and who paid them" section.

An energy consulting firm founded in 1989, hired across the industry by utilities, state regulators, government agencies, and environmental nonprofits alike. Disclosure printed in the report itself: it was paid for by the Data Center Coalition, the industry's lobbying group, which reviewed it before publication. E3 states the conclusions are its own. Two things keep it usable anyway: E3 is the same firm Virginia's legislature independently hired for its own study, and this report openly lists its own limits, admitting that only a few studies exist, that most show correlation rather than cause, and that the risk "is not absent." Pro side: disclose the funding before you are asked. Anti side: quote the limitations rather than attacking the firm. It is far more effective. Quoting the report's own acknowledgements: it "was funded by the Data Center Coalition, who reviewed the whitepaper prior to publication," with E3 adding that the conclusions are its own and do not necessarily reflect the sponsor's views. Both halves of that sentence matter. The industry paid for the study and saw it before anyone else did, and the consultants say the findings are theirs. Cite the finding if you want, but never cite it as independent.

Virginia's utility regulator, three commissioners elected by the state legislature, with judge-like authority over what the power company may charge. This is the referee describing its own ruling, so the contract terms are authoritative while the framing is self-congratulatory. Note the split reading, because it is usable from either podium: industry says the November 2025 order proves big customers now pay their own way, while consumer and land conservation groups argued the commission ducked the harder question of who pays for the transmission lines.

An analyst at a Washington think tank that advocates for technology-friendly policy and is supported in part by technology companies. Its starting position is pro-buildout and it does not hide that. Most useful for one specific thing: its hard numbers showing that announced projects consistently overstate what actually gets built. Worth saying plainly: the companies funding the institute are the companies that benefit if its argument wins. ITIF discloses this itself, which is more than most do.

Tyler Norris is a Duke University researcher who previously developed solar projects and advised the U.S. Department of Energy, and he testified on this study before a congressional committee in March 2025. Tim Profeta founded the Duke institute and ran it for almost twenty years, after working as a Senate staffer on climate legislation. Dalia Patino-Echeverri is a Duke professor of energy systems. University research, not paid for by industry, and the most downloaded thing this institute has published in twenty years. Its headline number depends entirely on data centers agreeing to cut back at peak times, which most contracts do not currently require. That is where the anti side will push, and it is a fair push.

A nonprofit research institute founded in 1972 and funded mainly by electric utilities. It is regarded across the energy industry as technically solid and deliberately cautious. DCFlex is the live experiment testing whether data centers will actually reduce power use on command. Note that its 49 participants include the very companies being tested, so treat it as a serious joint engineering effort among interested parties rather than an independent audit. Its value is that it moved this question from theory to real measurements at real sites.

Published in Science, one of the two most prestigious research journals in the world. Eric Masanet is a professor at the University of California, Santa Barbara. Jonathan Koomey is an independent analyst who spent two decades correcting exaggerated forecasts about computers and energy. Arman Shehabi and Sarah Smith are federal lab scientists who also produce the official U.S. estimates today. State the caveat yourself before your opponent does: this covers roughly 2010 to 2018, those efficiency gains have largely been spent, and these same authors now publish the projections showing steep growth. Use it as a reason to be humble about scary forecasts, not as a claim that energy use is flat today.

A company reporting on itself, with a direct financial interest in how it looks. Included because the industry's environmental commitments are part of its argument and deserve to be quoted accurately from the source. Treat the specific figures as self-reported and unaudited. Note that the water research cited above is aimed directly at reports like this one, for counting only the water used at the building.

An analyst at a Washington think tank that advocates for technology-friendly policy and is supported in part by technology companies. Its starting position is pro-buildout and it does not hide that. Most useful for one specific thing: its hard numbers showing that announced projects consistently overstate what actually gets built.

Ari Peskoe runs the electricity law program at Harvard Law School and is one of the most cited experts in the country on how electricity is regulated. Regulators, utilities, and advocates all rely on his work, which makes him hard to dismiss. Eliza Martin was a legal fellow there. Their evidence is a review of nearly fifty real regulatory cases, so the claims rest on the official record rather than on a computer model. This paper is the foundation of nearly every consumer protection argument now being made to state regulators. Pro side: the correct response is that it is an argument about process and fairness rather than a measurement of dollars, and then to agree with them about the secrecy.

An industry news publication reporting the findings of Monitoring Analytics, the independent watchdog for the PJM grid, led since 1999 by Joseph Bowring. This is the credibility point to make out loud: the watchdog is not an advocacy group and not part of the grid operator's management. It is the official referee, created to check whether the market is being run fairly, and it reports to federal regulators. Bowring has publicly criticized the grid operator's own market design as well as data center demand, so this is not a one-sided source. The news article is the accessible version; the original is the watchdog's own market reports.

An industry news publication reporting the findings of Monitoring Analytics, the independent watchdog for the PJM grid, led since 1999 by Joseph Bowring. This is the credibility point to make out loud: the watchdog is not an advocacy group and not part of the grid operator's management. It is the official referee, created to check whether the market is being run fairly, and it reports to federal regulators. Bowring has publicly criticized the grid operator's own market design as well as data center demand, so this is not a one-sided source. The news article is the accessible version; the original is the watchdog's own market reports.

Sean O'Leary is a senior researcher at the Ohio River Valley Institute, a progressive research group founded in 2020 focused on Appalachian Ohio, Pennsylvania, and West Virginia. It does not claim to be neutral. What gives it real weight is a decade of documented work on the natural gas boom, where it showed counties posting excellent economic statistics while actually losing jobs and residents. That track record is the argument, and it is a comparison rather than a direct measurement of data centers. Expect that challenge, and answer it with the Brookings jobs figure.

A nonprofit founded in 1998 that tracks tax breaks given to corporations. It is openly opposed to corporate subsidies and has been for decades, across both parties and every industry, not just this one. Kasia Tarczynska is its senior researcher and the author of its data center work. Its methodological strength is the thing to say out loud when challenged: the revenue-loss numbers come from the states' own legally required financial disclosures, not from this group's estimates. These are the governments' own books. Its limitation is that it counts money given away without fully counting money later collected.

Two economists. Dany Bahar teaches international and public affairs at Brown University and is also a fellow at the Brookings Institution, a long-established research organization in Washington. Greg Wright teaches economics at the University of California, Merced. Neither one specializes in data centers, and that is exactly why this study is trusted: they applied ordinary economics to the question without a stake in the answer. Their clever move was comparing counties that got a data center against counties where one was announced and then canceled, which compares similar places rather than comparing boomtowns to nowhere. Use in debate: this is the one study both sides can honestly cite. It limits what the pro side can claim about jobs, and it hands the anti side its best evidence that the tax breaks are aimed at the wrong buildings. Brookings put that disclosure on the article itself, which is more than most institutions do, and the money is unrestricted rather than tied to this study. Both facts belong in the same breath: the four biggest data center builders fund the institution, and they did not commission this paper.

Virginia's nonpartisan investigative agency, created in 1973, staffed by career analysts and overseen by legislators from both parties. Its studies are ordered by law and it has no position to defend. For this one it hired the University of Virginia for the economic analysis and E3 for the electricity modeling. This is the most authoritative document in the whole debate and the first one you should reach for no matter which side you are on. It studies the place with more data centers than anywhere on earth, and it genuinely supports both cases. Always quote both halves before using your half. Worth carrying into an argument: E3, which modeled the grid and ratepayer effects here for the legislature, is the same firm that separately produced a study funded by the Data Center Coalition. Being hired by a legislature is not the same as being hired by the industry, but if you cite JLARC against someone citing E3, you are partly citing the same modelers.

Shaolei Ren is a professor of electrical and computer engineering at the University of California, Riverside, who studies the environmental cost of computing and was appointed to a United Nations panel on the subject. Mohammad Islam teaches computer science at the University of Texas at Arlington. The other two authors were researchers in Ren's lab. This is peer-reviewed academic work in a major computing journal, and it is the paper that put AI water use into the public conversation. Know its limit before somebody shows it to you: the estimates rely on efficiency figures the companies themselves report and on regional averages, so there is real uncertainty in the numbers, which the authors say plainly.

The largest nonprofit environmental law firm in the country, founded in 1971. It only ever represents the side doing the suing. So this is a legal filing by one party in an active case, not a neutral finding. Cite it for what is formally being alleged and what relief is being requested. Do not cite it as though a court had already ruled. A nonprofit public interest law firm founded in 1986, working across six southeastern states, and the lawyers representing the Memphis NAACP chapter in the air permit fight. They are a party to an active dispute, so their characterizations argue one side. The underlying facts are independently checkable and that is how to cite them: what the permit allows, how many turbines there are, that the county already fails the federal smog standard, and the American Lung Association's 2025 grade.

A project that tracks community opposition to data centers, run by a firm called 10a Labs whose stated clients are AI companies and large technology firms. It does not disclose who funds this specific research. Use in debate: the funding is the argument. A firm that works for AI companies has every reason to make opposition look small, so when their own count shows $130 billion in projects stopped in three months, the pro side cannot easily wave it away. Anti side, say the funding out loud before anyone asks. Take the no-funding claim as the firm's own statement rather than an audited fact, and note the position it sits in: every paying client of the parent company is on one side of the debate its tracker measures.

Bernie Sanders is the independent senator from Vermont. Alexandria Ocasio-Cortez represents part of New York City in the House. This is a political announcement, cited for what the most restrictive position formally asks for. The bill is not expected to pass. Anti side: do not adopt this as your position, it makes you easy to dismiss. Pro side: getting your opponent to defend a total national ban is a very good trap.

New Jersey's state utility regulator, announcing the results of its own annual auction that sets what households pay for electricity supply. This is the referee publishing its own scoreboard, so the percentage increases are authoritative for New Jersey. Used here for one purpose: turning the wholesale capacity price spike into what actually happened to a home's bill. Note that New Jersey was among the hardest-hit states, so treat its 17 to 20 percent as the high end, with PJM's own 1.5 to 5 percent estimate as the low end.

A nonprofit, nonpartisan newsroom covering Southwest and Southside Virginia, reader and foundation funded. Cited for its report on a 2026 state incentive evaluation finding the data center exemption produced $6.10 of labor income per exempted dollar, the strongest pro-side framing of Virginia's tax math. Both sides should hold this next to the 48-cents-on-the-dollar state revenue finding:²² labor income measures wages generated in the economy, mostly during construction; the 48 cents measures what actually came back to the state treasury. Different questions, both honest, and whichever side quotes only its favorite is cherry-picking.

A commercial market intelligence firm that tracks industrial construction projects and sells the data to builders, suppliers, and investors. It has no side in this debate; its customers just need accurate counts of what is actually being built. Cited for the denominator on the opposition point: record actual construction spending of $44.7 billion in Q1 2026 and a full-year 2026 investment forecast near $700 billion, alongside the market analysis putting 30 to 50 percent of the 2026 pipeline at risk of delay or cancellation. Both sides need this source: the pro side to show the buildout is not being stopped, the anti side to show the risk is now large enough that the industry's own trackers price it in.

The index is a tracking site that tallies known AI data center facilities and capacity by country; treat its exact figures as estimates of a fast-moving target, though the rough shape, America holding on the order of three quarters of the world's AI compute against China's roughly 15 percent, is consistent with academic and government estimates. The American Edge Project is a technology industry advocacy group, openly pro-buildout and funded by tech companies, cited here for its own framing of the race: of the four pillars of AI strength, compute is the one America still clearly leads. That framing cuts both ways and both sides should notice: it is the pro side's strongest reason to build, and it is also an admission that everything rests on one pillar.

PwC is one of the world's largest accounting and consulting firms; this study was commissioned by the Data Center Coalition, the industry's lobbying group. It is the source of the 4.7 million supported jobs figure and the claim that each direct job supports 6.5 more. The method is input-output modeling (IMPLAN), which estimates ripple effects through supply chains and household spending rather than counting actual jobs, and it reports gross activity, not net new jobs. Pro side: quote it with the funding and the method disclosed, or it will be done for you. Anti side: do not call it fake, call it a model, and put the measured Brookings numbers next to it. Modeled versus measured is the entire fight over this source. That engagement line matters more than the headline: the 4.7 million jobs figure is a modeled number produced for the industry's own trade association.

A letter published in Fayette County, Georgia's independent community newspaper, relaying the arithmetic of the QTS campus: land that paid about $11,000 a year in county taxes carries a bill projected to exceed $40 million a year at build-out based on current millage rates. A local writer's projection from public tax rates, not a certified county estimate; the page's wording hedges it accordingly.

An independent business journal covering central Iowa. Cited for West Des Moines financing about $300 million of sewers, water lines, roads and bridges through bonds repaid from Microsoft's property tax revenue, infrastructure work that opened roughly 4,500 acres for further development.

DatacenterDynamics is a trade publication covering the data center industry. It is editorially independent but writes for an industry audience. Cited for Mesa's own projections of city, county and state tax revenue from the approved Google campus, alongside the tax breaks granted.

Cardinal News is a nonprofit, nonpartisan newsroom covering Southwest and Southside Virginia, reader and foundation funded; the county page is the local government's own. Cited for Google paying the full cost of the substations and lines its Botetourt campus needs, grid work county officials say also improves reliability for existing residents.

The operators' own community grant program reports: Amazon InCommunities, Meta Community Action Grants (over $94 million across roughly 3,700 local projects, self-reported), Google Data Center Community Grants (local newspapers relaying Google-announced figures), and QTS's $50 million Dane County commitment. These dollar figures are company-published and not independently audited. Real money reaches real local nonprofits; the totals come from the companies themselves. A QTS press release originally cited alongside this no longer resolves, so it has been removed rather than left pointing at a dead page.

A public community college's own program page and a Northern Virginia local news outlet. Cited for the existence of a dedicated data center operations credential program and for IBEW apprentice pay in the DC-area market as InsideNoVa reports it: starting around $26 an hour, with completed apprentices making about $106,000 a year plus benefits.

EPRI is a research nonprofit funded mainly by electric utility members, so treat it as utility-industry-funded research rather than independent academia. Its causal study estimates state residential rates in 2024 would have averaged about 6 percent higher without 2019 to 2024 data center growth, because new demand spread fixed grid costs over more sales. That is a national average; the mid-Atlantic price spike that ran the other way is documented separately by the grid's independent market monitor, source 17. Latitude Media provides independent context on the study's limits.

Trade press, county records and local broadcast news documenting named noise cases: Chandler, Arizona residents reporting a constant hum since 2014 and the city creating a noise-complaint liaison requirement in 2022; Prince William County HOAs formally complaining of round-the-clock industrial noise, with county enforcement for steady tonal noise effective May 1, 2026; and WJLA's reporting on the July 2026 power failure at an Ashburn campus that put the site on diesel backup for about a day and drew a flood of resident complaints about noise and fumes.

Oregon Public Broadcasting is a nonprofit, member-supported public broadcaster. Cited for Google's share of The Dalles' municipal water growing from 12 percent in 2012 to about a third by 2024, at roughly a million gallons a day, as the city looks to pull more water from the Mount Hood forest.

Grist is a nonprofit newsroom focused on climate and environment, funded by grants and donations; its beat is explicitly environmental, so read it as skeptical of the buildout by disposition. The projection it reports comes from Ceres, a sustainability nonprofit: Phoenix-area data center cooling water growing as much as tenfold, from about 385 million to 3.8 billion gallons a year. Grist itself notes the estimate is a worst-case scenario that may run high. This site quotes it with those hedges attached.

Georgia's largest newspaper. Cited for Newton County homeowners near Meta's facility reporting wells running dry and sediment in their water starting during 2018 construction, an active lawsuit against Meta, the water authority projecting a county supply deficit by 2030, and the facility's roughly 500,000 gallons a day, about a tenth of county water use. A lawsuit is allegations, not a ruling; cite it as what is formally claimed.

Trade press and local broadcast news on a public record: Prince William County's Digital Gateway rezoned roughly 2,133 acres of rural land for up to about 23 million square feet of data centers, approved 4 to 3 with one supervisor abstaining, after a hearing of more than 27 hours on December 13, 2023, and a Virginia appeals court later voided the rezoning because public-hearing notice requirements were not met.

Inside Climate News is a nonprofit investigative newsroom focused on climate. The Center for Biological Diversity is an environmental advocacy organization and a party asking regulators to act; treat its petition as advocacy. Cited for how diesel backup fleets work: operators largely self-determine what counts as emergency use and self-report emissions, with limited outside oversight.

The real estate industry's trade association, membership funded. Its 2026 report is cited precisely because it found a mixed result rather than a one-sided one: 25 percent of surveyed agents reported a positive effect on nearby home values and 22 percent a negative one, and residents' top concerns raised with agents were energy costs at 61 percent and water use at 56 percent.

Planet Detroit is a nonprofit local environmental newsroom; Fox Carolina is local broadcast news. Cited for named construction-phase disruption: gravel truck traffic through Saline, Michigan for the Oracle and OpenAI build, and Spartanburg, South Carolina residents reporting construction noise as early as 1 a.m.

Epoch AI is an independent nonprofit research group that measures AI compute trends, funded by philanthropic and research grants rather than industry. Its tracked dataset puts roughly 75 percent of global AI supercomputer compute in the United States, about 15 percent in China, and 10 percent elsewhere, measured by where the hardware physically sits. The most neutral available source for the compute-share claim. Worth noticing across this whole source list: Epoch is paid by two organizations cited separately on this site, EPRI and the AI Index, as well as by the largest AI labs. Epoch says it prices that work at consulting rates so it is not subsidised, and takes no policy positions.

Georgia's elected utility regulator, reported by Georgia Public Broadcasting, a nonprofit public broadcaster. This one source feeds both columns: the commission's approved fuel and storm-cost deal projects the typical residential customer saving about $50 a year, roughly $285 million annually across customers, and the same commission's Public Interest Advocacy staff estimate the current rate structure could raise residential fuel costs 5 to 11 percent as large users' costs spill over.

Federal statistical data from the Bureau of Economic Analysis, as computed by private analysts and relayed by CNBC. S&P Global's analysis of BEA data attributed roughly 80 percent of the rise in first-half 2025 final private domestic demand to data center and related high-tech spending, and Harvard economist Jason Furman computed a related figure near 92 percent of measured GDP growth for that window. CNBC's own framing is the caveat this site keeps attached: across the full year 2025, consumption was the largest growth driver and AI capital spending was second.

A progressive Ohio policy nonprofit that openly opposes the exemption, citing Ohio Department of Taxation data. The underlying numbers are the state's own: more than $1.5 billion forgone in 2025 against an official forecast of $136 million, which the group corrected in its own piece to more than eleven times the projection. Advocacy framing over government figures; quote the Department of Taxation numbers, not the adjectives.

Virginia's nonpartisan legislative auditor again, in its January 2026 incentive report: the data center exemption cost about $1 billion in fiscal 2024, up from $685 million the year before, with cumulative exemptions of $2.7 billion over fiscal 2015 to 2024, more than half of all Virginia incentive spending across that decade. The same agency that found the exemption outperforms the state's other tax breaks; both findings are real and this site cites both.

The corporate subsidy watchdog's Indiana release, built on the state's own newly required disclosures: $655 million in exemptions from 2019 through 2025, with the most recent two years showing over 93 percent of the money going to a single company, Amazon. An advocacy group, openly against these subsidies, reporting government disclosure figures.

A law firm's client summary of a Texas statute; the primary source is the legislature's own bill text and the utility commission's rulemaking. SB6 gives Texas authority to require new loads of 75 megawatts and above, which in practice means data centers, to curtail during grid emergencies, with mandatory protocols for connections after December 31, 2025. The firm calls it a marked policy shift; analysts call Texas a bellwether other states may follow.

Fortune is an established business publication; Circle of Blue is a nonprofit newsroom covering water. Fortune documented a Georgia project drawing over 29 million gallons without proper billing oversight during severe drought and an Arizona contractor moving about 650,000 gallons from Tucson's system without authorization. Circle of Blue covered Ohio's July 2026 decision, after more than 7,000 public comments, to scrap a proposed streamlined statewide data center wastewater permit and keep case-by-case review.

The IEA is an intergovernmental energy agency funded by and reporting to member governments including the United States. It projects global data center electricity emissions peaking around 320 million tonnes of CO2 by 2030, easing to about 300 by 2035, under 1 percent of global emissions, while U.S. power-sector emissions rise modestly in 2026 as natural gas expands to meet data center demand. Gas is currently the largest single power source for U.S. data centers.

The federal energy laboratory's December 2024 report, the water companion to the energy figures this site already cites. U.S. data centers directly consumed about 17 billion gallons of water for cooling in 2023, projected to reach 38 to 73 billion by 2028, plus roughly 211 billion gallons consumed indirectly through the electricity generation powering them. Federally funded, peer reviewed, the neutral scorekeeper on national water totals.

NERC is the federally certified reliability regulator for the North American grid, overseen by FERC; not a government agency and not industry advocacy. Its latest long-term assessment raised the ten-year summer peak demand projection by 224 gigawatts, 69 percent above the prior projection, with data center growth the largest single contributor, and found 13 of 23 assessment areas at elevated or high resource-adequacy risk. It has also issued alerts after real events where sudden data center load losses disturbed grid frequency and voltage.

The BIS is the bank for the world's central banks; the IMF is the intergovernmental fund monitoring global financial stability. Neither is industry funded or an advocacy group. BIS Bulletin 120 documents the shift from cash-flow to debt financing of the AI buildout, warns some structures may mask leverage by moving it off balance sheet, and flags the supply chain's exposure to a capex slowdown; the five largest technology companies plan over $1 trillion of AI spending across 2025 and 2026. The IMF's April 2026 report is more measured, judging the financial stability impact modest so far. This site quotes each at its own temperature.

The Hill is an established Washington news outlet; NRDC is an environmental advocacy group and its framing should be read as advocacy, though the underlying auction figures come from PJM and its independent market monitor. Cited for the scale claim that the capacity price spike reached bills for roughly 65 to 67 million people across 13 states and DC, a spread that reflects different PJM releases, with reported average increases around 30 percent for the 2024 auction cycle.

Louisiana Economic Development is the state agency that recruited the project, so its releases are official government sources with a promotional purpose. Cited for the announcement dates and dollar totals ($10 billion in December 2024, more than $50 billion by July 2026), the 5 gigawatt scale, the state's jobs figures of 7,500 peak construction workers and about 1,000 operating jobs, and the claim of more than $1.6 billion in contracts to Louisiana businesses since groundbreaking.

Meta describing its own project, so every figure is company self-reported. Cited for the financing structure (a joint venture valued around $27 billion in which funds managed by Blue Owl Capital hold about 80 percent and Meta about 20 percent), the $215 million pledged to Entergy bill assistance and efficiency programs, and Meta's report that construction-driven sales tax funded Richland Parish teacher bonuses of up to $50,000.

Entergy Louisiana is the utility building and profiting from the new plants, so its figures are company self-reported and should be read alongside the intervenor sources. Cited for the utility's own account of the August 2025 approval of its Meta-dedicated generation and transmission package, including up to 1,500 megawatts of solar procurement, and for Entergy's claim that its agreements with Meta will deliver about $2.65 billion in combined customer savings ($2 billion over 20 years announced in 2026, on top of $650 million announced earlier), a claim intervenors dispute as unverified. The dollar cost and megawatt figures for the plants themselves are cited to press coverage in source 76.

The Lens is a nonprofit New Orleans newsroom, but read these pieces as advocacy published in a newsroom: the bylines are staff of the Alliance for Affordable Energy and the Union of Concerned Scientists, both formal intervenors against the power deal. Both pieces are labeled Opinion by The Lens, which disclosed each author's employer in the byline. Cited for the core risk math, which covers four Richland units, three Pointe Coupee units and two battery projects rather than the Richland plants alone: the gas plants are built for 30 to 40 year lives against Meta contracts of 15 to 20 years, an estimated $3.4 billion in still-unpaid Richland plant costs at contract end, roughly $10.3 million a year at each of the seven new plants in major maintenance costs recovered through the fuel adjustment clause (which also carries fuel and gas transportation costs) and spread across all Entergy customers, and the authors' conclusion that ratepayers appear to bear the financial risk if Meta leaves early.

The Union of Concerned Scientists and the Alliance for Affordable Energy are the two named intervenors against the power deal, represented by Earthjustice, a public interest law firm, so all three are advocacy sources. Cited for the August 2025 vote breakdown (4 to 1, with commissioner Davante Lewis the sole dissent), UCS's modeled estimates that the first three gas plants would emit more than 300 metric tons of nitrogen oxides and about 4.8 million metric tons of carbon dioxide a year (modeled, not measured), and Earthjustice's January 2026 filing alleging the deal is structured through a holding company in a way that could let Meta exit years early while ratepayers keep paying.

KPLC is a Louisiana broadcast affiliate owned by Gray Media; Straight Arrow News is a national outlet. Both are press coverage with no stake in the docket. Cited for the April 2026 fast-track vote (4 to 1 again) on Entergy's roughly $21 billion proposal for seven additional gas plants, with the commission expected to take up the full application in December 2026, and for the transparency fight: the commission's own chief judge subpoenaed Meta for its electricity demand and permanent job figures, and Meta moved to quash, calling the subpoena unreasonable and oppressive. How that fight ended is documented in source 76.

Floodlight is a nonprofit investigative newsroom covering climate and energy, co-published here with Verite News, Louisiana Illuminator, The Guardian, The Lens and WWNO. Cited for its documented findings that state senator Jay Morris co-sponsored bills enabling the project's land deal and tax breaks while he and business partners bought seven properties within five miles of the site after the announcement, and sold about 300 acres to Entergy roughly four weeks after the August 2025 approval vote, with ethics experts quoted saying this could implicate Louisiana's conflict of interest rules. These are reported findings, not any official finding of wrongdoing.

Federal government statistics republished by Census Reporter, an independent open-source project rather than a Census Bureau product. Cited for the context that makes Richland Parish a test case: about 19,800 residents, a poverty rate of 24.1 percent against 18.9 percent for Louisiana and 12.5 percent for the US, and a median household income of $53,544, about two thirds of the national figure. Survey estimates carry margins of error of a few points.

OpenAI is a company with a direct commercial stake in the data center buildout it is reporting on, so this is self-reported threat intelligence, here paired with independent press coverage. Cited for the June 2026 report that OpenAI banned a cluster of ChatGPT accounts, assessed as likely run from China by a social media team working for provincial government clients, that generated English-language comments and cartoons blaming data centers for rising electricity bills and posted them through inauthentic American personas on X. OpenAI's own report says the activity drew little or no authentic engagement, and that finding is as important as the operation itself.

Alethea is a commercial threat-intelligence firm that sells narrative monitoring and communications counsel, so it has a business interest in influence operations being seen as a serious problem; it does not disclose clients or funders. Cited for its documentation that Russian outlets (including fabricated articles spoofing real tech-news sites), roughly 46 Chinese state media articles, and Iranian state media amplified existing US data center opposition, and for its own two admissions: the report measures amplification, not impact, and foreign actors did not create the grievance, they found it. The handbook cites the same report at the company's research landing page; both links lead to the same document.

X's disclosure is a platform's self-report with no published methodology, so it is cited here only together with the independent check. X said it found a suspected Chinese network of about 200,000 accounts, of which about 200 posted about AI data centers, electricity prices and grid strain under American-sounding names. Darren Linvill of Clemson University's Media Forensics Hub examined the accounts X showcased and found at least 59 had zero engagement across about 90 posts over a few days in January, saying no real humans saw the data center posts and calling the episode a non-story. Both halves matter: the network existed, and its measured audience was roughly nobody.

The Bitcoin Policy Institute is an advocacy nonprofit whose stated mission favors American computing buildout, and it does not disclose its own funders, so read this as the maximal version of the foreign-influence case made by an interested party. Cited for its claims that Chinese and Russian state media campaigned against US data centers, that a network of nonprofits funded by Shanghai-based Neville Roy Singham (about $278 million to six US nonprofits from 2017 to 2023) organized against projects, and that foreign-linked philanthropy flowed downstream to groups opposing data centers, contributing by its count to $23.6 billion in stalled or blocked investment. The report shows no causal link between foreign money and any specific project decision, does not publish its method for attributing dollars, and its congressional echoes are inquiry letters, not findings. The non-disclosure is itself worth quoting back: this is the report arguing that opposition to data centers is foreign funded, published by an organization that will not say who funds it.

Tech Policy Press is a nonprofit technology policy outlet; this is a signed analysis piece by an independent researcher, so treat it as argued analysis rather than neutral reporting. Cited as the strongest published critique of the foreign-influence framing: it documents that the leading reports group fabricated content together with amplified but accurate journalism, quotes Alethea's own caveat that detecting amplification says nothing about the truth of the underlying claim, and notes that the influence-operation framing is convenient for companies and governments facing genuine domestic opposition.

Established press coverage of a court case, cited as the clearest documented example of a foreign-funding accusation collapsing under demand for proof. Data center developer and investor Kevin O'Leary said on national television, across at least ten appearances, that two named Utah community groups opposing his project were acting as proxies for the Chinese government. After a legal demand, he posted that he had no evidence any of the named people or groups were funded by China or the Chinese Communist Party. The groups' founders sued O'Leary and Fox News for defamation in July 2026. The suit is pending; nothing here is an adjudicated finding, but the retraction is in O'Leary's own words.

Gallup is an independent polling organization; this is a national telephone survey of 1,000 adults. Cited for the finding that 71 percent of Americans would oppose an AI data center built in their local area, including 48 percent strongly opposed, with majorities of every party group opposed. Strong opposition runs 56 percent among Democrats and 39 percent among Republicans, so the sentiment leans left but is broadly bipartisan. This is the base rate any foreign-influence claim has to explain: the opposition is large, domestic and measurable without any foreign input.

Two official US government documents that anchor the AI race as stated national policy rather than industry talking point. The NSCAI, a bipartisan commission created by Congress and chaired by Eric Schmidt and Robert Work, concluded in its 2021 final report that the United States was not prepared to compete in the AI era and urged sustained investment in computing infrastructure. The intelligence community's 2026 Annual Threat Assessment continues to identify China as the leading strategic competitor, with artificial intelligence among the central technologies in that competition. Note the commission's chair came from the technology industry, and neither document addresses local data center siting disputes; they establish the strategic frame, not the local tradeoffs. Note for anyone checking: the commission's own site at reports.nscai.gov no longer resolves, the commission having wound up; the report is still widely mirrored, and the ODNI assessment link is live.

Louisiana local press coverage of the two decisive commission moments. KNOE, a Monroe TV station owned by Gray Media, covered the August 20, 2025 approval: a 4 to 1 vote with commissioner Davante Lewis dissenting, three turbines dedicated to Meta with 2,200 megawatts of combined capacity, and Entergy planning to spend about $3.2 billion on generation for the site, with Meta contracted to cover the cost of the three dedicated turbines. WWNO, New Orleans public radio, covered August 12, 2026, when the commission voted 3 to 1, again with Lewis alone in dissent, to overrule its own administrative judge's ruling that would have required Meta to disclose how it calculated its publicized job projections and electricity needs.

The Data Center Coalition is the industry's trade association, funded by member companies; its public member list includes Amazon Web Services, Google, Microsoft, Meta, Digital Realty, Equinix and dozens more, though as a 501(c)(6) the exact dues and donor amounts are confidential under US law. Tech Policy Press documented the coalition's network of state-level advertising campaigns with names like Virginia Connects and Connected Georgia, running near-identical ads in at least eight states, whose spokespeople declined reporters' questions about who pays for the ads. The 990 filing is the primary public record of the group's finances.

The American Edge Project is a 501(c)(4) advocacy group that now runs pro-data-center messaging and names no funders on its own site. Its tax filings, as reported by the Tech Transparency Project and covered by Gizmodo and CNBC, show Meta (then Facebook) provided $4 million in its first year and $34 million the following year, making the company effectively its sole funder in those years. Cited as the clearest documented case of a single company funding a group that presents itself as a broad coalition. As a 501(c)(4) it has no legal duty to disclose donors, so current-year funding is not public.

Nonprofit newsroom coverage of a state ethics enforcement action, the best documented case in this debate of hidden money behind a local pro-data-center campaign. A group called Elevate New Mexico ran mailers supporting Oracle's Project Jupiter data center without registering as a lobbyist; the New Mexico State Ethics Commission sued in April 2026 and the case settled in August 2026. Disclosures from the case showed a project developer paid the group about $3.7 million ($3,667,000, from Yucca Growth Infrastructure LLC), and Oracle separately disclosed about $3.4 million in advertising and lobbying spending over four months, which the commission described as likely the largest lobbying advertising campaign ever disclosed in the state.

The self-published finances of three national groups active against parts of the buildout, cited to show what opposition funding actually looks like where it is disclosed. The Union of Concerned Scientists reports 86.6 percent of its funding from individual donors and 13.2 percent from foundation grants. Earthjustice, which litigates data center air permit cases, publishes audited statements showing 87 percent of revenue from contributions, most of it from individuals. Food and Water Watch, which organized the national moratorium letters, publishes top-line totals but not a current donor breakdown. None of the three publishes any line item earmarked for data center work; it is funded from general budgets.

The documented core of the foreign-linked-money story on the opposition side. A New York Times investigation, later referenced in CBS News coverage of a 2026 federal investigation into the same network, traced funding from Shanghai-based American businessman Neville Roy Singham, whose network reported hundreds of millions of dollars into a handful of US nonprofits including The People's Forum and CodePink, flows corroborated by the groups' own tax filings. CodePink's own site documents its Milwaukee chapter picketing a data center project in May 2026. What is not documented anywhere, including in the advocacy reports that press this theory hardest, is any money trail from this network to the local coalitions fighting specific projects, or to the Party for Socialism and Liberation, which files no public financial disclosures at all.

The two foundations most often named in advocacy claims that foreign-born billionaires bankroll US data center opposition, cited here through their own published grant lists as the check on that claim. The Wyss Foundation's disclosed 2024 grants went to about 120 organizations, heavily land conservation groups; the Oak Foundation's environment programme lists international conservation grantees. Neither list names any data-center-specific opposition group. Some of their multi-issue environmental grantees signed general moratorium letters that mention data centers, which is where the documented trail ends: general support to broad organizations, not money earmarked against data centers.

The International Telecommunication Union is the United Nations agency for digital technology, which says it funds its work through membership fees from member states, the private sector and academic institutions, which is worth knowing because the companies it tracks are also among its members. The World Benchmarking Alliance is a nonprofit that ranks companies on sustainability. Cited for two separate findings: that data center electricity use grew 12 percent a year from 2017 to 2023, four times faster than global electricity growth, and that the 164 digital companies reporting electricity data used 581 terawatt hours in 2023, about 2.1 percent of world electricity, with ten companies accounting for half of it, and that 23 companies ran entirely on renewable electricity in 2023, up from 16 the year before. The 581 figure covers those companies' whole operations, not data centers alone. These are figures the companies themselves reported; the report counts only those that disclosed.

The United Nations University Institute for Water, Environment and Health is a UN research institute, funded by the UN and by donor governments; its specialty is water, which is worth knowing when reading a report whose headline finding is about water. Cited for its 2030 projections: 945 terawatt hours of electricity for global data centers, a water footprint of 9.3 trillion liters, a land footprint above 14,500 square kilometers, and up to 2.5 million tonnes of electronic waste a year. Every one of those is a projection, not a measurement, and the report gives an estimated 448 terawatt hours for 2025 as its starting point. Do not attribute these figures to the UN Environment Programme, which is a different body.

SPAN is a California company announcing and selling its own product, so every figure here is company self-reported and none of it has been independently audited. Cited for what the company says XFRA is: a distributed network of compute nodes in residential and small commercial spaces, built on liquid-cooled Nvidia RTX PRO 6000 Blackwell server GPUs, with Nvidia as launch partner and the homebuilder PulteGroup working with SPAN on the initial rollout. Hosts are offered a SPAN panel, battery backup, optional solar and fixed discounted rates for electricity and internet. The announcement promises deployments beginning in 2026 and a pipeline toward gigawatt scale in 2027. Note what it does not contain: the widely repeated cost and speed comparisons and the $150 a month figure were given to reporters, not published here.

Three trade and consumer outlets covering the same announcement; none is independent technical verification, and much of what they carry is the company restating its own claims to reporters. Several of the most quoted figures were given first to CNBC and relayed by these outlets, so the chain is company to CNBC to here. Realtor.com, a real estate listings site, reports that the units are about the size of an HVAC condenser, that a sample host fee is $150 a month covering electricity and internet (roughly half a typical bill), that PulteGroup calls itself in the early stages of testing, and that the proof of concept covers about 100 homes; it also relays Nvidia and PulteGroup statements originally made to CNBC. Network World independently carries the same SPAN claim to CNBC that it can install 8,000 units about six times faster and at five times lower cost than a centralized 100 megawatt data center, and describes the hardware as liquid cooled and designed to minimize sound. pv magazine, a solar trade outlet, puts the proof of concept at about 100 nodes in a southwestern state, likely Nevada or Arizona; note that it counts nodes where Realtor.com counts homes. Network World also carries the useful dissent: analyst Alex Cordovil argues the economics depend on the nodes consuming locally generated surplus, that servicing a scattered fleet is costly, that hardware turns over quickly, and that a server on a residential wall cannot match a hardened facility's security, concluding it may complement large data centers rather than replace them.

A nonprofit, foundation-funded newsroom that covers climate and energy and has won a Pulitzer Prize. Its sympathies run toward communities and the environment, and this piece is built on resident interviews at Great Oak. The quotes are the residents' own words; the decibel readings are the residents' measurements, later matched by the company's after mitigation. Pro side: do not dispute the experience, dispute what follows from it.

A trade publication for the data center industry, advertiser-funded and generally sympathetic to the business it covers. Cited for the mitigation facts at Great Oak: 424 exhaust tubes replaced and a roughly 10 decibel reduction, confirmed by both residents' and Amazon's measurements. That dual confirmation is what makes the number safe to quote from either podium.

A local weekly newspaper in Prince William County, Virginia, the county with the second-largest data center concentration in the world, which has covered the noise and zoning fights there continuously since 2022. Local, reader-funded and close to the story. Its value is specificity: the proposed 52 and 47 decibel limits, the finding that the Amazon complex would still fail them, the exemption for generators, and the planning staff's objection that viewshed studies could not be trusted without building drawings.

The same county government that publishes the Loudoun tax story, here describing its own rules and their gaps: a 55 dBA limit at residential property lines, no rule for tonal or low-frequency noise, amendments not expected until 2027, by-right approvals ended March 18, 2025, and a Phase 2 standards project approved September 16, 2025 that names aesthetics and compatibility among its subjects. The same page states that Loudoun cannot regulate outdoor air quality, that this falls to the Virginia Department of Environmental Quality, and that "On July 1, 2026, DEQ began mandating Tier 4-equivalent emissions standards for new or modified generators at data center sites," regulating nitrogen oxides, carbon monoxide and particulate matter, while most existing generators in the county are Tier 2. A government admitting a gap on its own website is about as reliable as a source gets; both sides can quote it without a fight over credibility.

A city government describing its own ordinance (No. 5033, effective January 5, 2023), so the provisions are authoritative and the framing is self-congratulatory. Chandler is the longest-running data center noise dispute in the country, with complaints dating to 2014. Both sides use it: the pro side as proof that a workable rulebook exists, the anti side as proof of how long a city can take to write one.

A nonprofit founded in 1987 that buys and preserves Civil War and Revolutionary War battlefield land, funded by members, donors and land-preservation grants. It was a party to the fight against the Digital Gateway, so its account is an advocate's victory announcement. The checkable facts are the project size, the court dates and the withdrawal of the appeal; the characterization of the project as destructive is the Trust's own. Pro side: note that the court voided the rezoning over how the county ran its hearings, not over what the buildings would look like.

SELC is described above; here it represents the plaintiffs, so this is one side's filing. WFDD is the NPR member station in Winston-Salem, North Carolina, publicly and listener funded, reporting a planning board vote. Cite the vote and the lawsuit as events, and the "rural character" objection as what residents argued, not as a finding.

A large law firm's client alert, written for developers who need to know what changed. Neutral on the merits, precise on the mechanics: after March 18, 2025, data centers in Loudoun's industrial districts need a special exception rather than being allowed by right. Use it for the rule, not for any view of whether the rule is good.

An annual report from a university institute, compiled by academics with data from industry and public sources, and the most-cited neutral scorecard of the AI race. Its 2026 edition reports that as of March 2026 the top US model led the best Chinese model by 2.7 percent, that US private AI investment reached $285.9 billion in 2025 against $12.4 billion in China, and that China leads in publication volume, citations, patents and industrial robot installations. Use it to keep "the AI race" concrete: it is several races at once, and the scoreboard differs by event.

A research institute whose stated mission is "investigating the trajectory of AI for the benefit of society." It publishes open databases of AI models (3,200 and counting), training compute, chip shipments and large AI data centers. It takes no position in the data center debate. The most useful single place to check "who has the compute" claims against data. Worth noticing across this whole source list: Epoch is paid by two organizations cited separately on this site, EPRI and the AI Index, as well as by the largest AI labs. Epoch says it prices that work at consulting rates so it is not subsidised, and takes no policy positions.

A Utah daily newspaper, reporting on the foreign-influence question with quotes from both the researchers and skeptics. Cited for balance: it carries the Alethea findings and the caution from a technology trade group president that foreign sources are "amplifying, I don't even know that they're creating it." Use it as the reminder that amplification and origination are different claims.

EDMO is a European Union-funded observatory of fact-checkers and researchers; its Storm-1516 profile summarizes a report by Viginum, the French government's foreign-interference agency, and records that Microsoft named the operation in April 2024 and that European intelligence documents tie its financing to Russian military intelligence. The Justice Department release documents the seizure of 32 domains used by the "Doppelganger" operation, which imitated legitimate news sites at the direction of the Russian presidential administration. These are the two most citable descriptions of how state-run fabrication networks work: a government agency's finding and a court-filed seizure. Neither is about data centers specifically.

The three standing trackers. Meta publishes semiannual Adversarial Threat Reports naming coordinated inauthentic networks it removed and the countries they operated from. Microsoft's Threat Analysis Center publishes attributions of state influence operations (it named Storm-1516). The Foreign Malign Influence Center is the US intelligence community's coordinating office for the subject. Platform reports describe what the platform found on its own service; a government attribution carries legal weight; neither says anything about a story until it names it. Meta's transparency reporting page has moved and the old link returns an error, so check its current threat reporting hub directly rather than the URL as first cited.

The IEA's own world estimate for data center electricity. Note how far it sits from other series: this puts 2025 at 485 TWh while the UN University report at source 84 puts the same year at 448 TWh and the Energy Institute series at source 101 is different again. None is wrong; they count different things. Check which series a number came from before setting two of them side by side.

A long-running energy dataset, now published by the Energy Institute with partners after BP handed it over. Two things to carry: the site reached this figure through Forbes reporting rather than the review itself, so it is a number at one remove, and the underlying data-center series is attributed to S&P Global, a commercial data provider, rather than to public statistics.

The federal government's own energy statistics agency. This is a measured year-to-date figure for commercial electricity sales rather than a data center number, used here as the closest measured proxy, so do not quote it as a data center total.

The later quarter of the same tracker cited at source 25, counting projects blocked or delayed. Its totals are cumulative and change between reports, so a figure from one quarter is not comparable to another without saying which report it came from.

Who paid, from the document itself: PJM produced this in house, funded by the fees its members pay. That makes it authoritative on PJM's own system, because nobody else has better data on it, and interested at the same time, because a forecast of steep demand growth strengthens the case for building the transmission and generation PJM administers. Read it as the operator's own best estimate rather than as an independent audit of that estimate. Neither side disputes the direction; they dispute who should pay for meeting it.