The Data Center Debate

Report center

Every report, and what it is used to prove

78 reports, studies, datasets and official documents. Each one is tagged with the arguments it supports or refutes, who paid for it, and the caveat the other side will raise. Filter by argument to see the evidence on each question, from both directions.

Used by the case for Used by the case against Cited by both sides 78 shown
Cited by both sides industry report October 12, 2026

North America Data Center Report, Midyear 2026

JLL research team

JLL's midyear report describes a market at absolute capacity, with multi-year grid queues pushing tenants and developers toward secondary hubs and new financing arrangements. It notes growing local opposition as a structural factor rather than a short-term headwind.

  • Multi-year interconnection queues cited as a binding constraint across North America
Capacity prices and grid costs: frames interconnection queues as the main scarce resource Local opposition and consent: lists local opposition as a durable constraint on growth

Who paid: Produced by JLL, a commercial real estate firm with data center brokerage, consulting and advisory lines of business.

Caveat: JLL is a commercial broker that benefits from active data center transactions, so its growth outlook may frame constraints as surmountable; its own siting data is not public.

Added October 13, 2026 by the daily update.

Used by the case against government report October 9, 2026

Power and Profits: How the AI Data Center Boom Costs Households and Communities

Senators Elizabeth Warren, Richard Blumenthal, Chris Van Hollen, U.S. Senate

The report compiles responses from Amazon, Google, Meta, Microsoft, CoreWeave, Digital Realty and Equinix to a December 2025 Senate inquiry. The senators conclude the companies are not paying the full cost of grid upgrades their new AI data centers trigger and will continue to pursue state tax exemptions and nondisclosure agreements.

  • 7 companies surveyed
  • billions in state sales tax exemptions claimed by named firms
Who pays for the grid buildout: says companies pay less than full grid cost State tax breaks: flags state sales tax exemptions as costing public budgets Secret contracts and disclosure: documents continuing use of NDAs

Who paid: Prepared by three Democratic senators' office staffs on official U.S. Senate time; based on written responses from seven named companies.

Caveat: This is a Democratic senators' report, not a bipartisan committee product; several companies dispute the characterization and point to their own grid and community spending.

Added October 11, 2026 by the daily update.

Used by the case against government official document October 9, 2026

Power and Profits: How the AI Data Center Boom Costs Households and Communities

Office of Senator Elizabeth Warren with Senators Chris Van Hollen and Richard Blumenthal, United States Senate

The report, drawn from nearly a year of document requests to Amazon, Alphabet, Meta, Microsoft, CoreWeave, Digital Realty and Equinix, concludes that hyperscalers reject but-for cost allocation for shared grid upgrades, continue to use nondisclosure agreements with utilities and local governments, and are still pursuing state and local tax breaks. It frames the findings as evidence that households are picking up part of the AI buildout's cost.

  • Virginia tax exemption cost 1.6 billion dollars in fiscal 2025, 16 times the original estimate
  • Meta's federal tax bill fell from 9.6 billion to 2.8 billion dollars under 2025 bonus depreciation changes
Who pays for the grid buildout: finds hyperscalers decline to pay but-for shares of shared grid work Secret contracts and disclosure: documents continued use of NDAs with utilities and local officials State tax breaks: cites state and federal depreciation benefits in dollar terms

Who paid: Produced by Senate Democratic offices as part of a congressional oversight investigation; no outside funding disclosed.

Caveat: The report is produced by three Democratic senators and reflects their reading of company responses; the companies named dispute the characterization and distinguish between full energy cost and but-for cost allocation.

Added October 13, 2026 by the daily update.

Used by the case against academic study October 8, 2026

UMBC Maryland Statewide Poll on Data Centers

UMBC Institute of Politics, University of Maryland, Baltimore County

The statewide survey found 78 percent of Maryland likely voters oppose new data centers in the state, with 13 percent in support. Top concerns listed were household electricity costs, water quality and quality of life.

  • 78 percent oppose
  • 13 percent support
Local opposition and consent: state level evidence of broad voter opposition

Who paid: UMBC Institute of Politics

Caveat: Media summary did not include sample size and margin of error; look for UMBC's full methodology before citing in detail.

Added October 9, 2026 by the daily update.

Used by the case against academic study October 8, 2026

Data Center Policy in Ohio: A Cost-Benefit Analysis

Scioto Analysis (Rob Moore)

The firm's cost-benefit analysis estimates that under current Ohio policy, data centers would produce only about $2.4 billion in net benefits between 2027 and 2034, barely more than their costs. A construction ban scored a benefit-cost ratio of 2.7 and a requirement that new centers use 90 percent clean power scored 2.9, both higher than the current regime's 1.04.

  • Net social benefit 2013 to 2026: $62 billion, mostly from 96,000 jobs and $75 billion in wages
  • Net benefit 2027 to 2034 under current policy: $2.4 billion
  • Benefit-cost ratios: current policy 1.04, ban 2.7, 90 percent clean power requirement 2.9
  • Estimated electricity cost to Ohio ratepayers 2027 to 2034: $16 billion
  • Estimated air pollution costs 2027 to 2034: $41 billion
State tax breaks: Ohio has spent about $2.3 billion in subsidies for data centers while the forward-looking policy barely breaks even. Air quality and on-site power: Air pollution is modeled as the largest single cost of future data center growth in Ohio. Who pays for the grid buildout: Study attributes about $16 billion of future ratepayer cost to higher electricity prices.

Who paid: Columbus-based policy research firm; funding not disclosed in the news coverage reviewed.

Caveat: The results are modeled projections built on specific damage-cost assumptions for electricity, air pollution and jobs. A different set of assumptions, especially about whether jobs would have gone elsewhere, could change the ratios.

Added October 10, 2026 by the daily update.

Used by the case against advocacy report October 8, 2026

The AI Gas Rush: Risks and Implications of Bring Your Own Power

Earthjustice and Better Data Center Project

The report tracks announced bring your own power projects for AI data centers and estimates that gas plants will supply about 80 percent of their generation. It warns that current permitting rules let these plants move outside standard utility review.

  • 177 GW of planned gas capacity tied to BYOP projects
  • roughly 300 GW of AI data center capacity planned or under construction in the US
  • 80 percent of BYOP generation from gas over the next 5 years
Air quality and on-site power: frames BYOP buildout as a new gas plant wave outside normal permitting Siting, zoning and standards: argues BYOP sits outside existing regulatory tools

Who paid: Earthjustice is a non-profit environmental law organization funded by foundations and individual donors; the report was co-produced with the Better Data Center Project.

Caveat: Earthjustice is an advocacy group that litigates against gas and coal plants, so the framing is adversarial; the 177 GW figure includes announced but not permitted or financed projects.

Added October 11, 2026 by the daily update.

Used by the case against academic study October 8, 2026

With Americans in a Sour Mood, Democratic Voters Are More Engaged in Midterms

Pew Research Center staff

In a survey conducted September 28 through October 4, Pew asked whether voters liked or disliked local elected officials who support building data centers in their area. 61 percent said they disliked such officials and 8 percent said they liked them, with majorities disliking across party, geography and age groups.

  • 61 percent dislike officials who support local data centers
  • 8 percent like officials who support local data centers
  • 75 percent of Democrats dislike, 44 percent of Republicans dislike
Local opposition and consent: measures political cost of visible local support Laws, bills and regulation: frames the issue as electorally salient in 2026

Who paid: Pew Research Center is a nonprofit subsidiary of The Pew Charitable Trusts.

Caveat: The question asks about a hypothetical nearby data center rather than any specific project; feelings may shift when voters see actual terms, benefits agreements or local jobs.

Added October 13, 2026 by the daily update.

Used by the case against news tracker October 7, 2026

More Than $260 Billion in Data Centers Canceled After Local Backlash in 2026

Heatmap Pro, Heatmap News

Heatmap Pro says it tracked at least $260 billion of data center investment canceled in 2026 after local opposition, with roughly $130 billion of that in the third quarter alone. The outlet also says about $1 trillion of planned investment now faces meaningful local resistance.

  • $260B canceled in 2026
  • $130B canceled Q3 2026
  • ~$1T facing opposition
Local opposition and consent: quantifies local backlash as material to the buildout Efficiency and forecasts: many announcements will not become built capacity

Who paid: Heatmap News, a for profit climate news outlet; the data is produced by its paid Heatmap Pro subscription product.

Caveat: Heatmap Pro's methodology is proprietary and its totals are larger than other trackers; readers cannot audit the project list.

Added October 9, 2026 by the daily update.

Used by the case against government official document October 7, 2026

Minutes of the Federal Open Market Committee, September 15 to 16, 2026

Federal Open Market Committee, Federal Reserve Board

The minutes record FOMC participants' view that core goods price pressure from AI buildout investment was growing even as tariff effects faded. Several warned the pace could push aggregate demand beyond supply over the medium term.

  • Inflation projected to stay above 2 percent target through 2029
Capacity prices and grid costs: data center capex now identified as a macro inflation factor How much electricity data centers use: supports claims that data center demand is reshaping the broader economy

Who paid: Federal Reserve System

Caveat: FOMC minutes report what participants said, not an official Fed finding; dissenting views are included.

Added October 9, 2026 by the daily update.

Used by the case against government legal filing October 2, 2026

Minnesota Attorney General initial comments on Xcel Energy's special contract with Google for the Pine Island data center

Minnesota Office of the Attorney General, Residential Utilities Division, Minnesota Public Utility Commission docket

Filing to the Minnesota PUC arguing that the Xcel-Google special contract does not fully assign very large customer costs to Google, as required by state law, and that redactions in the record prevent regulators and the public from verifying who pays for what.

  • Potential cost shift to other ratepayers described as millions and possibly more than a billion dollars
  • 1,900 megawatts of new clean energy resources tied to the agreement
  • Most dollar amounts in filings are redacted
Who pays for the grid buildout: State AG argues the contract leaves other customers exposed Secret contracts and disclosure: AG cites widespread redactions in the filing

Who paid: State of Minnesota, Office of the Attorney General, representing residential and small business ratepayers before the PUC

Caveat: This is an AG filing, not a PUC ruling. The commission will weigh Xcel and Google's responses before deciding.

Added October 5, 2026 by the daily update.

Cited by both sides academic study October 1, 2026

Leverage Earth observation data to support the clean energy transition

Ana Andries et al, Nature Reviews Clean Technology

Peer reviewed article argues governments and regulators should formally require satellite and Earth observation checks before approving AI data centers, combining renewable potential, groundwater, drought, biodiversity, land use and climate risk layers. Authors cover siting criteria rather than attempting a national electricity consumption estimate.

  • Combines at least eight geospatial layers for siting review including renewables, land surface temperature, groundwater and climate risk
Siting, zoning and standards: Advocates mandatory satellite based siting review Water use: Flags drought and groundwater exposure as decision inputs

Who paid: University of Surrey research team; funding sources not characterized in available summaries.

Caveat: The paper proposes a framework; it does not evaluate any individual US data center application and has no legal effect by itself.

Added October 3, 2026 by the daily update.

Used by the case for market research report October 1, 2026

October 2026 Data Center Report: Spending Reaches $98.5 Billion YTD, Construction Costs Rise

ConstructConnect research team

ConstructConnect reports that August 2026 US data center project spending totaled $14.4 billion across 17 projects, lifting year to date spending to $98.5 billion across 153 projects. The report also finds the average data center cost rising to $1.18 billion at 707,000 square feet, up from $507 million and 493,000 square feet in 2025.

  • $14.4 billion in August 2026 data center starts across 17 projects
  • $98.5 billion year to date through August 2026 across 153 projects
  • Average project cost $1.18 billion, up from $507 million in 2025
  • AVAIO Little Rock, Arkansas campus listed at $6 billion as the largest August start
How much electricity data centers use: Spending surge implies continued load growth even as some projects stall. Jobs: Rising construction spend supports industry claims of construction job demand.

Who paid: Published by a construction data vendor whose customers are contractors, suppliers and lenders.

Caveat: The tracker counts announced projects and starts, not completions, so the dollar totals include capacity that may be delayed, scaled back or canceled later.

Added October 4, 2026 by the daily update.

Used by the case against government legal filing October 1, 2026

Attorney General initial comments to the Minnesota PUC on the Xcel-Google Pine Island electric service agreement

Minnesota Office of the Attorney General, Residential Utilities Division, Minnesota Office of the Attorney General Keith Ellison

Initial comments to the Minnesota PUC argue Xcel has not shown that its electric service agreement with Google for the proposed Pine Island data center is just, reasonable or compliant with state law assigning very-large-customer costs to those customers. The office warns of exposure to about $1 billion in net costs for other ratepayers and criticizes Xcel's redaction of key dollar figures.

  • Potential net cost to other ratepayers around $1 billion, versus Xcel's claim of about $1.1 billion in benefits
  • Billions of dollars in incremental resource costs if Google reduces or terminates purchases early
Who pays for the grid buildout: concrete state-level filing arguing data center deal could shift costs to households Secret contracts and disclosure: flags Xcel's redactions of financial impact figures Who holds the risk: stranded-cost risk if Google exits

Who paid: Taxpayer-funded state consumer advocacy office.

Caveat: These are initial AG comments, not a PUC ruling; Xcel disputes the numbers and the commission has not yet acted.

Added October 6, 2026 by the daily update.

Cited by both sides academic report October 1, 2026

Data Centers in Illinois: Economic, Fiscal and Environmental Impacts

Frank Manzo IV and Robert Bruno, Illinois Economic Policy Institute and the Project for Middle Class Renewal at the University of Illinois Urbana Champaign

Models the economic and fiscal effects of $57 billion in planned Illinois data center investment through 2035, concluding the sector would support 121,000 total jobs but only 2,800 direct permanent positions, add about $300 million per year in local property taxes once operational, and raise an average household's electric bill by about $12 per month.

  • $57 billion planned investment through 2035
  • 121,000 total jobs modeled
  • 2,800 direct permanent jobs
  • ~$12 per month average residential bill impact
  • $300 million per year projected property tax at full operation
Jobs: only 2.3 percent of modeled jobs are permanent Local property tax revenue: sizable property tax base at full operation Who pays for the grid buildout: models a $12 monthly household bill rise

Who paid: ILEPI is a labor aligned nonprofit research institute. The report lists no specific external funder.

Caveat: Jobs and bill impact figures are modeled, not measured. ILEPI is a union aligned research group, and the policy recommendations largely track priorities a labor coalition has already supported.

Added October 8, 2026 by the daily update.

Cited by both sides academic study September 24, 2026

Financing the AI Buildout

Stijn Van Nieuwerburgh, Brookings Institution

The paper estimates US AI infrastructure spending at 10.3 trillion dollars from 2025 through 2032, roughly 3.6 percent of annual GDP, and warns that increasingly complex debt structures combined with unproven AI revenue could create systemic risk. It projects US data-center capacity rising from about 57 GW today to about 183 GW by 2032.

  • 10.3 trillion dollars projected US AI infrastructure spending 2025-2032
  • 3.6 percent of annual GDP on average
  • US data-center capacity projected to rise from about 57 GW to about 183 GW by 2032
How much electricity data centers use: capacity roughly tripling by 2032 raises the demand side of the debate Who holds the risk: warns of systemic financial risk from opaque, complex financing AI race and national security: frames scale as a national economic bet

Who paid: Brookings Institution publication by a Columbia Business School finance and real-estate professor.

Caveat: This is a projection, not built capacity, and the author explicitly frames the financing arrangements as potentially unstable.

Added September 29, 2026 by the daily update.

Used by the case against academic study September 22, 2026

Pennsylvania Data Center Opposition Ethnographic Study

Livia Garofalo, Maia Woluchem and colleagues, Data and Society Research Institute

A qualitative study drawing on 18 months of fieldwork and 44 interviews with Pennsylvania residents that describes how industrial history, secrecy through NDAs and material concerns over rates and property values shape local opposition to data centers.

  • 44 residents interviewed
  • 18 months of fieldwork 2024 to 2026
  • 66 possible Pennsylvania data centers, 5 under construction and 5 approved as of the study
  • $19,000 in Data Center Coalition lobbying reported in Pennsylvania the prior year
Local opposition and consent: Documents grassroots and multi partisan roots of opposition Secret contracts and disclosure: Highlights role of NDAs and industry campaigns in eroding trust Jobs: Authors say residents see thin evidence of promised job growth

Who paid: Part of the Trustworthy Infrastructures program supported by a Mellon Foundation grant made in 2023.

Caveat: Qualitative interview based work in one state. Findings are not statistically generalizable and reflect a specific sample rather than representative polling.

Added September 23, 2026 by the daily update.

Used by the case against academic study September 22, 2026

More Americans now say data centers are negative for the environment, energy costs and more

Pew Research Center

A survey of 10,548 US adults conducted July 20 to August 9 found Americans are markedly more negative about data centers than in January, with 54 percent now saying they are mostly bad for the environment, up from 39 percent, and 50 percent saying they are mostly bad for home energy costs, up from 38 percent. About 6 in 10 said they would be uncomfortable with a new data center in their area.

  • 54 percent say data centers are mostly bad for the environment, up from 39 percent in January
  • 50 percent say mostly bad for home energy costs, up from 38 percent
  • 49 percent say mostly bad for quality of life nearby, up from 30 percent
  • 60 percent uncomfortable with a data center in their area
  • 88 percent have heard at least something about data centers, up from 75 percent
Local opposition and consent: Discomfort is majority across parties and community types Who pays for the grid buildout: Half of adults now link data centers to higher home energy costs Jobs: Only 24 percent say bad for local jobs, so a majority still see jobs as neutral or positive

Who paid: Pew Research Center, funded primarily by The Pew Charitable Trusts

Caveat: The survey measures perceptions, not measured environmental or price impacts.

Added September 28, 2026 by the daily update.

Used by the case against advocacy tracker September 21, 2026

Data Center Watch Q2 2026 Report on Local Opposition

Data Center Watch research team

The quarterly tally finds 45 US data center projects worth about $68 billion were blocked or delayed between April and June 2026, brings the tracked count of local opposition groups to 843 across 49 states, and logs opposition activity outside the US for the first time.

  • 45 projects blocked or delayed
  • $68 billion combined value
  • 843 opposition groups in 49 states
Local opposition and consent: quantifies growing local resistance Siting, zoning and standards: opposition tied to zoning and disclosure fights

Who paid: Run by 10a Labs, a for-profit firm whose paying clients are AI and technology companies. Its author says this tracker is unfunded and kept separate from that paid work, which is the firm's own statement rather than an audited fact.

Caveat: The report bundles blocked and delayed projects together, and its methodology relies on public records and news reports rather than a project-level regulatory database.

Added September 21, 2026 by the daily update.

Used by the case against market research study September 17, 2026

Colorado 2026 AI and Data Center Opinion Survey

Magellan Strategies

A survey of 1,389 registered Colorado voters conducted August 26 to September 2 finds broad opposition to nearby data centers and majority support for a construction ban. Opposition holds across party lines and rises among younger voters.

  • 82 percent oppose a data center in their city or town
  • 86 percent oppose one near their neighborhood
  • 55 percent favor a ban on all new data center construction
  • margin of error plus or minus 2.63 percentage points
Local opposition and consent: state-level polling reinforces national opposition trend

Who paid: The pollster states the survey was designed and self-funded by Magellan Strategies and was not developed with, shared with, or sponsored by any outside organization before fieldwork.

Caveat: The pollster identifies as a Republican firm, and the survey was conducted online via MMS text-to-web, which can differ from live-caller results.

Added September 21, 2026 by the daily update.

Used by the case against government study September 16, 2026

Impacts of Data Center Growth on Pennsylvania Electric Reliability and Affordability

Synapse Energy Economics for the Pennsylvania Public Utility Commission

The study models PJM reliability from 2027 through 2030 and finds that under current forecasts the region falls short of its planning standard; under a high data center growth scenario without new generation, expected loss-of-load hours climb to more than 13 per year by 2030.

  • baseline ~5.9 loss-of-load events per decade by 2030
  • high-demand case: 13+ hours per year by 2030
  • 48-fold increase in shortage risk 2029-2030 under high case
  • PA PUC floats a model tariff for customers of 50 MW and above
Capacity prices and grid costs: Independent state agency now says data-center-driven growth threatens reliability without new build. How much electricity data centers use: Confirms data centers as primary driver of PJM shortfall. Who holds the risk: PUC uses findings to justify large-load tariff for 50 MW-plus customers.

Who paid: Prepared for the Pennsylvania Public Utility Commission, which engaged Synapse Energy Economics and Mondre Energy to do the analysis.

Caveat: The 13-hour figure is a modeled high-growth scenario that assumes little new generation and limited demand response.

Added September 17, 2026 by the daily update.

Used by the case against academic dataset September 16, 2026

BGSU Poll September 2026: Ohio voters on data centers

BGSU Democracy and Public Policy Research Network with YouGov, Bowling Green State University

Survey of 1,000 likely Ohio voters conducted September 1 to 10 finds broad support for pausing new data centers and skepticism about their local benefits. The poll also asks about candidates and general election priorities.

  • 78 percent support a moratorium until more information is gathered
  • 75 percent oppose data center construction in their own community
  • 63 percent oppose construction elsewhere in Ohio
  • 47 percent say a candidate's stance on data centers is very or extremely important
  • 9 percent list data centers as their top issue
  • Margin of error plus or minus 3.96 points
Local opposition and consent: supermajority backs pause Laws, bills and regulation: voter salience for statewide bills

Who paid: BGSU academic program, fielded by YouGov.

Caveat: A stated preference poll is not a vote, and support for a pause is not the same as support for a permanent ban.

Added September 20, 2026 by the daily update.

Cited by both sides industry report September 16, 2026

2026 Annual Outage Analysis

Uptime Institute Intelligence

The annual review reports that data center outages are becoming less frequent overall but that resilience gains are slowing under pressure from AI workloads, aging power infrastructure and grid dependence. Power availability, interconnection delays and external grid reliance are called out as growing systemic risks.

  • Power and grid factors cited as top emerging outage risks
How much electricity data centers use: AI workloads raise reliability pressure Capacity prices and grid costs: grid capacity shortfalls are reshaping outage risk

Who paid: Uptime Institute's own annual survey. The institute is funded by the data center operators it certifies, and the full methodology, including who was surveyed, is behind a customer paywall.

Caveat: Uptime is funded by industry members, and its outage data comes from voluntary reporting and press-tracked incidents rather than a comprehensive regulator dataset.

Added September 21, 2026 by the daily update.

Cited by both sides market research report September 15, 2026

September 2026 Data Center Report: Year-to-Date Spending Nearly Three Times a Year Ago

ConstructConnect research team

Aggregates project-start data from public and private sources for US data center construction. Finds year-to-date starts through July 2026 reached $84.1 billion, close to 25 percent of all US nonresidential building starts.

  • $84.1 billion YTD data center starts through July 2026
  • roughly 24-25% of US nonresidential construction starts
  • cost per square foot rising faster than square footage
  • activity shifting from coastal metros toward Midwest and Southeast
How much electricity data centers use: Construction volume implies large downstream load additions. Jobs: Construction spending supports temporary employment in host regions. Local property tax revenue: Higher investment lifts property values but concentrates revenue impact in a few counties.

Who paid: ConstructConnect's own market intelligence product; the company is a business unit of Roper Technologies. Its paying customers are contractors and suppliers who benefit from a growing construction market.

Caveat: Starts count announced project value at groundbreaking, not built capacity or completed spending.

Added September 17, 2026 by the daily update.

Cited by both sides industry report September 14, 2026

US AI data centers will need $110 billion in new power plants by 2030

Moody's Ratings

Moody's estimates 45 GW of new generation, mostly natural gas, is needed by 2030 to meet U.S. data center demand, costing about $110 billion and adding an estimated $25 to $30 billion a year to national electricity costs.

  • 45 GW new generation by 2030
  • $110 billion buildout
  • Over 30 GW from natural gas
  • Under 5 percent from nuclear restarts
  • $25-30 billion added to annual electricity costs
  • $15 billion paid directly by data centers via behind-the-meter projects
How much electricity data centers use: large new fossil buildout attributed mainly to data centers Who pays for the grid buildout: $25-30B annual cost increase includes portions not paid by data centers Air quality and on-site power: over 30 GW of new gas capacity implies rising emissions

Who paid: Moody's own research, not commissioned. Worth knowing: Moody's rates the debt of the utilities and data center operators this analysis covers, so it is assessing borrowers it also grades.

Caveat: Moody's numbers are analyst estimates, not utility integrated resource plans, and it treats behind-the-meter data center power as a full offset to grid costs.

Added September 18, 2026 by the daily update.

Used by the case against government study September 14, 2026

PJM reliability projection for Pennsylvania under rising data center demand

Synapse Energy Economics, Pennsylvania Public Utility Commission

Independent modeling for the Pennsylvania PUC projects that PJM will breach its own one-in-ten-year reliability planning standard from 2027 through 2030 as data center load grows, with a base-case loss-of-load expectation of about 5.9 events per decade by 2030 and a worst case of 13.2 events in a single year.

  • Base case 5.9 loss-of-load events per 10 years by 2030
  • Worst case 13.2 loss-of-load expectation in 2030
  • 89 data centers planned, under construction or operating in Pennsylvania as of mid-August 2026
How much electricity data centers use: attributes near-doubling of state load by 2040 to data centers Capacity prices and grid costs: regional grid falls below reliability standard 2027 to 2030 Efficiency and forecasts: modeled scenarios include high-load and low-supply extremes

Who paid: Commissioned and funded by the Pennsylvania Public Utility Commission.

Caveat: The reliability shortfalls are modeled, not observed, and assume limited new generation additions relative to PJM's current interconnection queue.

Added September 19, 2026 by the daily update.

Cited by both sides government official document September 9, 2026

September 2026 Short-Term Energy Outlook

U.S. Energy Information Administration

EIA projects record US electricity generation and sales in 2026 and 2027, driven by commercial-sector data center growth and manufacturing. Commercial sales are forecast to rise 3.3 percent in 2026 and 2.7 percent in 2027; commercial and industrial together account for the bulk of growth.

  • 4,135 BkWh US electricity sales projected in 2026 (up ~2%)
  • 4,211 BkWh projected in 2027 (up ~2%)
  • commercial sales +3.3% in 2026, +2.7% in 2027
  • West South Central remains largest contributor to sales growth
How much electricity data centers use: Official federal forecast attributes bulk of near-term growth to data centers and manufacturing. Efficiency and forecasts: EIA growth path is well below some hyperscaler-driven scenarios, reflecting continuing forecast uncertainty.

Who paid: Federal agency of the U.S. Department of Energy; no external funding.

Caveat: Short-term outlooks change with each monthly issue; the West South Central number reflects existing projects and does not fully account for the recent Texas interconnection pause.

Added September 17, 2026 by the daily update.

Cited by both sides government official document September 9, 2026

September 2026 Short-Term Energy Outlook press release

U.S. Energy Information Administration

EIA forecasts record U.S. electricity generation of 4,368 billion kWh in 2026 and another 1.7 percent increase in 2027, with commercial sector sales growth of 3.3 percent in 2026 and 2.7 percent in 2027 driven mainly by data centers and manufacturing.

  • 4,368 billion kWh generation in 2026
  • 2.2 percent generation growth in 2026
  • 1.7 percent generation growth in 2027
  • Commercial sales up 3.3 percent in 2026
  • Commercial sales up 2.7 percent in 2027
How much electricity data centers use: record generation forecast driven by data centers Efficiency and forecasts: EIA has revised past STEO figures multiple times

Who paid: U.S. federal statistical agency in the Department of Energy.

Caveat: STEO forecasts change monthly; commercial sector electricity sales lump data centers together with many other business uses.

Added September 18, 2026 by the daily update.

Cited by both sides government official document September 1, 2026

Data Centers and the Electricity Grid: Frequently Asked Questions

Congressional Research Service

The report reviews data center electricity use, federal policies encouraging AI infrastructure, and the recent downward revision of NERC electricity demand projections. It notes that only about half to sixty percent of near-term announced data center capacity is expected to come online on schedule.

  • 50 to 60 percent of near-term announced capacity expected to come online on time
  • NERC revised demand projections downward in May 2026
How much electricity data centers use: provides federal baseline for grid demand claims Efficiency and forecasts: highlights slippage between announced and delivered capacity

Who paid: Prepared by CRS for members of Congress; funded by federal appropriations.

Caveat: CRS reports summarize existing sources and do not conduct new modeling. Their forecasts inherit the uncertainty of the underlying utility and NERC projections.

Added September 21, 2026 by the daily update.

Cited by both sides industry tracker September 1, 2026

Large Load Projects and Tariffs September 2026

Edison Electric Institute staff

EEI's September 2026 tracker lists approved and pending large load and data center tariffs by state, showing 23 states with at least one approved tariff and another seven with pending proposals as of mid 2026. The compilation identifies which utilities have filed rate class carveouts, minimum bill provisions and cost causation terms.

  • 23 states with approved large load tariffs
  • 7 states with pending large load tariffs
  • dozens more state proposals under review
Who pays for the grid buildout: shows how many states now separate large load costs Laws, bills and regulation: map of state regulatory activity

Who paid: Compiled by the Edison Electric Institute, the trade association for investor-owned utilities, funded by member dues. The list is the industry's own self-reported account of its projects and tariffs.

Caveat: EEI represents utilities and the compilation is not audited by regulators; state definitions of large load vary.

Added September 24, 2026 by the daily update.

Cited by both sides government official document August 8, 2026

2026 Draft National Transmission Needs Study

Office of Electricity, U.S. Department of Energy

The draft study identifies Virginia and Texas as having the highest current data center electricity demand and projects that Virginia, Texas, Arizona and Oregon will see the largest increases through 2030. It concludes the U.S. transmission system requires substantial new investment to accommodate rising data center, manufacturing and industrial loads.

  • Virginia and Texas cited as top current data center loads
  • Comments due September 8, 2026
How much electricity data centers use: official federal view on load growth drivers Capacity prices and grid costs: identifies transmission expansion needs

Who paid: Issued by the Department of Energy and paid for by federal appropriations, with much of the underlying analysis performed by national laboratories that are themselves run by outside contractors.

Caveat: This is a draft for public comment, not a final study.

Added September 22, 2026 by the daily update.

Cited by both sides news news article August 1, 2026

Most Americans Loathe Data Centers. Construction Workers Love Them.

Boston Globe staff

News report contrasting broad public opposition to data centers with strong support for the industry among the construction workers who build them.

Local opposition and consent: Documents that most Americans dislike data centers, cited as evidence opposition is broad. Jobs: Documents construction workers' support for the industry, underscoring the jobs argument.

Who paid: Regional daily newspaper; independent news reporting.

Caveat: A single news snapshot of public opinion and worker sentiment, not a rigorous survey instrument in its own right.

Handbook source 7. Full citation and biography

Cited by both sides industry dataset July 2026

Statistical Review of World Energy 2026 (data centre electricity table)

Energy Institute with Ember, KPMG and Kearney; data center figures from S&P Global Energy

The 75th edition of the former BP Statistical Review, and the first to publish a data center electricity table. It puts US data center use at 312.6 TWh in 2025, up 25.5% from 249.0 TWh in 2024, out of a world total of 787.8 TWh.

  • US 312.6 TWh in 2025 (39.7% of world)
  • US 249.0 TWh in 2024
  • World 787.8 TWh in 2025, up from 658.2 TWh
How much electricity data centers use: the highest of the three 2025 estimates; used by the case against to show scale and by the case for to show the US lead AI race and national security: US at nearly 40% of world data center electricity

Who paid: Produced by the Energy Institute, a UK professional body, with industry partners; the data center series comes from S&P Global Energy, a commercial market-intelligence firm.

Caveat: Its 2024 US figure (249 TWh) is far above Berkeley Lab's (192 TWh), so the series are not interchangeable. The full table sits behind a registration wall; the figures here are as reported by Forbes on August 23, 2026.

Added September 16, 2026 by the daily update.

Used by the case for news news article June 19, 2026

A State Study Found That Data Center Tax Break Was Virginia's 2nd Best Incentive, Producing $6.10 of Labor Income for Every $1 in Exemptions

Cardinal News staff

Reports a 2026 state incentive evaluation finding Virginia's data center sales tax exemption was the state's second-best incentive, producing $6.10 of labor income for every dollar exempted.

  • $6.10 of labor income per exempted dollar
  • Ranked Virginia's 2nd-best incentive
State tax breaks: Strongest pro-side framing of Virginia's tax exemption math, contrasting with JLARC's 48-cent state-revenue finding.

Who paid: Nonprofit, nonpartisan newsroom covering Southwest and Southside Virginia; reader and foundation funded.

Caveat: Labor income measures wages generated, mostly during construction, a different question from the 48-cents-on-the-dollar figure measuring actual state treasury return; citing only one framing is cherry-picking.

Handbook source 28. Full citation and biography

Cited by both sides government report June 2026

United States Data Center Energy Usage Report: 2025 Update

Sarah J. Smith, Alex Hubbard, Alex Newkirk, Mohan Ganeshalingam, Billie Holecek, Dale Sartor, Michael Mills, and Arman Shehabi, Lawrence Berkeley National Laboratory

The federal government's official accounting of how much electricity US data centers use, with a range of projections for 2030. Found data centers used 192 TWh in 2024, about 4.7 percent of US electricity, with a 2030 range of 578 to 782 TWh.

  • 192 TWh in 2024 (4.7% of US electricity)
  • 649 TWh central estimate for 2030 (11.8%)
  • 578-782 TWh across the 2030 sensitivity scenarios (10.5%-14.2%); wider compounded uncertainty range 521-843 TWh (9.5%-15.3%)
How much electricity data centers use: Source of the baseline 2024 usage figure and the 2030 range that both sides accept. Efficiency and forecasts: The same authors' current projections show steep future growth, used to counter the efficiency-fixed-this-before argument.

Who paid: Funded by the US Department of Energy's Industrial Technologies Office; the lab is operated for the DOE by the University of California under contract DE-AC02-05CH11231. Input data was purchased from Omdia Research, a commercial market research firm, and industry equipment makers and data center owners commented on the analysis assumptions.

Caveat: The 2030 estimate is a range, 578 to 782 TWh across the sensitivity scenarios and 521 to 843 TWh on the report's wider compounded uncertainty band, precisely because a single confident number would not be honest; quoting only the midpoint as a prediction overstates the report's certainty.

Handbook source 1. Full citation and biography

Cited by both sides government official document May 19, 2026

Data center server energy use grows across the commercial building stock

US Energy Information Administration, EIA, Today in Energy

Explains how EIA's Annual Energy Outlook 2026 treats data centers. Estimates servers alone were about 7% of commercial-sector electricity in 2025 and projects server use of 446 to 818 billion kWh by 2050.

  • Servers about 7% of commercial-sector electricity in 2025
  • 446 to 818 billion kWh of server use by 2050
  • 22% to 33% of commercial building electricity by 2050
How much electricity data centers use: the federal statistical agency's own accounting of data center load

Who paid: Federal statistical agency.

Caveat: Counts servers only, not cooling or the rest of the building, and the 2025 number is an estimate.

Added September 16, 2026 by the daily update.

Used by the case for news news article May 2, 2026

Building Trades Unions Emerge as a Key Ally of Tech Giants in Push for AI Data Centers

Boston Globe staff

News report on construction trade unions partnering with technology companies on data center buildout, including union hours and apprenticeship figures.

  • Local 26 electricians' union: 28 million hours worked last year, double a decade earlier
  • 600 new apprentices in a single year
Jobs: Source of construction-phase union labor hours and apprenticeship figures, the pro side's strongest jobs evidence.

Who paid: Regional daily newspaper; independent news reporting.

Caveat: Reflects union and industry framing of a genuinely large construction workforce; construction jobs are temporary and end when a given wave of building finishes, a point the anti side presses.

Handbook source 7. Full citation and biography

Used by the case for industry study May 2026

Understanding the Drivers of Rising Electricity Rates and the Role of Data Centers

Energy and Environmental Economics, Inc. (E3)

Energy consulting firm study finding no measurable evidence to date that other customers are subsidizing data centers, noting the states with the fastest demand growth, Texas and Virginia, had among the smallest price increases, and cataloging at least 38 new regulatory rules for large customers since 2018.

  • No measurable evidence to date of ratepayer subsidy of data centers
  • At least 38 new rules for large customers 2018-2026, 30 since 2025
  • Texas and Virginia: fastest demand growth, among smallest price increases
Who pays for the grid buildout: Central pro-side study finding no measurable subsidy of data centers by other ratepayers. Who holds the risk: Documents 38 new rules creating separate cost responsibility for large customers.

Who paid: The report's own acknowledgements: funded by the Data Center Coalition, the industry trade association, which reviewed the whitepaper before publication. E3 states the conclusions are its own.

Caveat: Paid for and reviewed before publication by the Data Center Coalition; the report's own fine print admits only a few studies exist, most show correlation rather than causation, and the risk is not absent.

Handbook source 8. Full citation and biography

Used by the case for advocacy report April 7, 2026

Four Reasons New AI Data Centers Won't Overwhelm the Electricity Grid

Robin Gaster, Information Technology and Innovation Foundation

Think tank analysis arguing announced data center projects consistently overstate what actually gets built, citing a more than 40 percent drop in deals in late 2025 and only about a third of announced projects under construction.

  • Deals fell more than 40% in late 2025
  • Only about a third of announced projects actually under construction
Efficiency and forecasts: Central evidence that announced buildout figures overstate what is actually built, used to temper scary growth forecasts.

Who paid: No funder is stated on the report. Its publisher discloses supporters including Amazon, Alphabet, Meta, Microsoft and NVIDIA, the companies that benefit if its argument holds.

Caveat: Written by an analyst at a pro-buildout think tank that does not hide its starting position; most useful narrowly for its hard numbers on the gap between announced and built projects.

Handbook source 10. Full citation and biography

Used by the case against government press release March 25, 2026

Sanders, Ocasio-Cortez Announce AI Data Center Moratorium Act

Office of Senator Bernie Sanders

Press release announcing legislation from Senator Bernie Sanders and Representative Alexandria Ocasio-Cortez proposing a national moratorium on AI data center construction.

Laws, bills and regulation: Represents the most restrictive proposed policy position, a national construction pause.

Who paid: Official congressional press release.

Caveat: A political announcement, not research; the bill is not expected to pass, and the handbook itself warns the anti side not to adopt a total ban as its position since it is easy for opponents to dismiss.

Handbook source 26. Full citation and biography

Used by the case for government official document February 2026

SCC Data Center Initiatives: Ensuring Data Centers Pay Their Own Costs

Virginia State Corporation Commission

Fact sheet summarizing the SCC's data center cost-allocation initiatives, framed as ensuring large customers pay their own way.

  • Restates the 14-year minimum-take contract framework from the November 2025 order
Who holds the risk: Restates the minimum-take contract framework as evidence data centers now bear project risk.

Who paid: Official fact sheet from Virginia's utility regulator.

Caveat: Self-congratulatory framing from the regulator about its own ruling; consumer groups argue it left the transmission-cost question unresolved.

Handbook source 9. Full citation and biography

Used by the case against government official document January 2026

Ordinance No. 2026-03, amending the Subdivision and Land Development Ordinance to add Section 713, Data Centers

Susquehanna County Board of Commissioners, Susquehanna County, Pennsylvania

A rural Pennsylvania county answering the question almost no one asks during a boom: who takes the building down afterward. Following a public hearing on January 5, 2026, the county added Section 713 to its land development ordinance. Section 713.18 requires the owner to complete decommissioning at its own expense within twelve months after the end of the facility's useful life, treats a facility as finished if it sits unused for twelve continuous months, and requires the county to be told within 48 hours of any cessation of use. The owner must post and maintain decommissioning funds equal to the net decommissioning cost plus ten percent, held by a bonding company or chartered lending institution, with cost estimates refiled after the first year of operation and every fifth year after that. The ordinance also requires Tier 4 or better generators and lets the county demand financial security for infrastructure and utility work.

  • Decommissioning funds equal to net decommissioning cost plus 10 percent
  • Decommissioning completed within 12 months of end of useful life
  • Unused for 12 continuous months is presumed end of useful life
  • Cessation of use reported to the county within 48 hours
  • Cost estimates refiled after year one and every five years
Siting, zoning and standards: Used by communities arguing that approval should carry an enforceable exit plan, borrowing the bonding mechanism long standard for wind and solar. Who holds the risk: Puts the cost of removal on the owner in advance rather than on the county if the operator leaves or fails.

Who paid: A county's own adopted ordinance. No funder; this is the binding local law itself.

Caveat: This is a small rural county with no large data center built under the rule yet, so it is a template rather than a tested outcome. A bond is only as good as the cost estimate behind it, and the estimate is prepared for the owner.

Used by the case for government official document 2026

Data Centers: The Loudoun Story

Loudoun County, Virginia

Official Loudoun County materials describing how data center tax revenue has shaped the county budget, including a decade of property tax rate cuts even as home values rose.

  • About $1.2 billion in property tax revenue, FY2026
  • About 39% of the county budget
  • Property tax rate cut every year, $1.145 to $0.805 per $100 assessed value, 2016-2026
Local property tax revenue: Documents ten straight years of county property tax rate cuts funded by data center revenue.

Who paid: Official county government communications, published partly to defend a locally controversial policy.

Caveat: The tax-rate figures are public budget documents and not seriously disputed, but the framing is a government advertising its own policy; critics focus on the risk of 39 percent of a budget depending on one industry rather than disputing the numbers.

Handbook source 4. Full citation and biography

Cited by both sides academic study 2026

New Evidence on Data Center Employment Effects

Dany Bahar and Greg Wright, Brookings Institution

Compares roughly 1,500 built data center facilities to 52 announced-then-canceled projects, finding computing employment rose 56 percent and telecom employment 43 percent in host counties, translating to roughly 100 to 200 permanent jobs, with home values up 2 to 5 percent.

  • Computing employment up 56%, telecom up 43% in host counties
  • Roughly 100-200 permanent jobs per county
  • Home values up 2-5% over a decade
  • Tax incentives = 2% of investment in company-owned facilities vs 62% in rental facilities
Jobs: The one jobs study both sides cite: modest but real permanent employment gains, far below the industry's modeled multiplier claims. State tax breaks: Finds subsidies flow disproportionately, 62%, to lower-job rental facilities versus only 2% for large company-owned ones. Local property tax revenue: Documents a 2-5% rise in home values in host counties.

Who paid: No funder is named for this study. Brookings discloses on the article that Amazon, Google, Meta and Microsoft are general, unrestricted donors to the institution, and states the conclusions were not influenced by any donation. Built on public data.

Caveat: Neither author specializes in data centers, part of why it is trusted as neutral; it measures modest gains, which limits the pro side's jobs claims while giving the anti side its best evidence that subsidies target the wrong buildings.

Handbook source 6. Full citation and biography

Used by the case for news news article 2026

Microsoft Says Its Data Centers Use 90% Less Water Than Its Earliest Facilities as Public Concern Grows

GeekWire staff

News report on Microsoft's claim that it cut data center water use roughly 90 percent from its earliest facilities, put back more freshwater than it took in during 2025, and designs its newest facilities to use no water for cooling.

  • ~90% water use reduction vs earliest facilities
  • Put back more freshwater than consumed in 2025
  • Newest designs use no water for cooling in normal operation
Water use: Source of Microsoft's water-efficiency improvement claims, used to counter water-footprint criticism.

Who paid: Technology news outlet reporting on Microsoft's self-reported sustainability claims.

Caveat: These are the company's own, self-reported and unaudited figures; the peer-reviewed water research this responds to exists precisely because company disclosures often count only the water used at the building, not the water used generating the electricity.

Handbook source 14. Full citation and biography

Used by the case for company report 2026

Operating Sustainably

Google

Google's own sustainability page describing its data center environmental commitments, including a pledge to match the no-water-cooling design goal by 2030.

  • Committed to a no-water-cooling design goal by 2030
Water use: Source of Google's stated water-reduction commitments.

Who paid: Google reporting on Google. Self-published, self-measured, and written to be persuasive about its own performance.

Caveat: A company reporting on itself with a direct financial interest in how it looks; treat the specific figures as self-reported and unaudited.

Handbook source 14. Full citation and biography

Used by the case against news news article 2026

Data Centers Drove $6.3B in PJM Capacity Auction Costs: Market Monitor

Utility Dive staff, reporting Monitoring Analytics findings

Reports the PJM grid's independent market monitor finding data centers responsible for $6.3 billion, 38 percent, of $16.4 billion in charges from one capacity auction, and nearly half of $63.6 billion across the last four auctions.

  • $6.3 billion, 38% of $16.4 billion in one auction
  • Nearly half of $63.6 billion across the last four auctions
Capacity prices and grid costs: Core source for the claim that data centers are the primary driver of the PJM capacity price spike.

Who paid: Industry news publication reporting findings from Monitoring Analytics, PJM's independent market monitor.

Caveat: The watchdog measures who caused the cost, not how the bill gets split among customer classes; new rules and Virginia's data center customer class govern the latter.

Handbook source 17. Full citation and biography

Used by the case against advocacy report 2026

Why Data Centers Will Be Economic Development Duds

Sean O'Leary, Ohio River Valley Institute

Argues data centers are capital-heavy and labor-light by design, drawing on the fracking boom in Appalachian Ohio, Pennsylvania, and West Virginia, where the biggest gas-producing counties posted strong growth statistics while still losing jobs and population.

  • Thirty biggest Appalachian gas-producing counties: above-average growth statistics, still lost jobs and population
Jobs: Argues data center investment figures do not translate into local jobs, by analogy to the fracking boom.

Who paid: Progressive regional research group founded in 2020, openly opposed to fossil-fuel and heavy-industry-style development; does not claim neutrality.

Caveat: An openly non-neutral advocacy group; its evidence is a historical comparison to natural gas counties, not a direct measurement of data center jobs, which the pro side will press.

Handbook source 19. Full citation and biography

Used by the case against advocacy report 2026

Data Center Subsidies Surge as States Lose Billions

Kasia Tarczynska, Good Jobs First

Documents that Georgia, Texas, Virginia, and Ohio each have data center tax exemptions capable of costing over $1 billion a year, with Ohio's cost jumping from $555 million in 2024 to $1.6 billion in 2025.

  • GA, TX, VA, OH exemptions each capable of costing over $1B/year
  • Ohio: $555M (2024) to $1.6B (2025), about $330 per household
  • Indiana: $655M, over 83% to one company, Amazon
  • Texas: $1B in one year
  • Oregon: $616M to Amazon, Apple, Alphabet, and Meta
State tax breaks: Core source for state-by-state tax exemption cost figures.

Who paid: Nonprofit founded in 1998, openly opposed to corporate subsidies across both parties and every industry, not funded by the data center industry.

Caveat: The revenue-loss numbers come from states' own legally required financial disclosures, but the analysis counts money given away without fully counting money later collected back.

Handbook source 20. Full citation and biography

Used by the case against advocacy report 2026

Inside Memphis' Fight Against xAI

Southern Environmental Law Center

Account of the fight over xAI's gas turbines in South Memphis, near Boxtown, a mostly Black neighborhood in a county that already fails the federal smog standard, where hundreds of residents opposed the permit and it was issued anyway.

  • County already fails federal smog standard; American Lung Association 2025 grade F for ozone
  • Hundreds attended the public hearing; every speaker opposed the permit
Air quality and on-site power: Core account of gas turbine pollution sited next to a majority-Black neighborhood already failing federal air standards. Local opposition and consent: Documents unanimous public opposition at the hearing, with the permit issued anyway.

Who paid: Nonprofit public interest environmental law firm representing the Memphis NAACP chapter in an active air permit dispute.

Caveat: The law firm is a party to an active dispute representing one side, so its characterizations argue that side; the underlying facts, permit terms, turbine count, air-quality grade, are independently checkable.

Handbook source 24. Full citation and biography

Used by the case against advocacy press release 2026

NAACP Asks Court for Emergency Action to Stop Illegal Air Pollution from xAI's Data Center Power Plant

Earthjustice

Press release announcing the NAACP's request for emergency court action against alleged illegal air pollution from xAI's Memphis data center power plant, reporting 59 turbines running without permits at a second nearby site.

  • 59 turbines reported running without permits at a second site
Air quality and on-site power: Source of the allegation of unpermitted turbines and the request for emergency court action.

Who paid: The largest nonprofit environmental law firm in the country, founded in 1971; represents only the side doing the suing.

Caveat: A legal filing and press release by one party in an active case, not a neutral finding or a court ruling; cite it for what is formally alleged and requested.

Handbook source 24. Full citation and biography

Used by the case against market research tracker 2026

Q1 2026 Report

Data Center Watch (10a Labs)

Tracker report finding about 75 projects worth roughly $130 billion blocked or delayed by local opposition in the first three months of 2026 alone, more than matching all of 2025, with opposition groups more than doubling across 49 states.

  • About 75 projects, about $130 billion blocked or delayed in Q1 2026
  • Opposition groups more than doubled across 49 states
  • 14 state legislatures took up pause proposals in three months
Local opposition and consent: Core tracker evidence of the scale and speed of local opposition to data center projects.

Who paid: Run by 10a Labs, a for-profit firm whose paying clients are AI and technology companies. Its author says this tracker is unfunded and kept separate from that paid work, which is the firm's own statement rather than an audited fact.

Caveat: Run by a firm whose clients are AI companies, giving it every incentive to undercount opposition, which makes its own large numbers harder for the industry to dismiss; it does not disclose funding for this specific research.

Handbook source 25. Full citation and biography

Cited by both sides market research report 2026

Actual, Planned Capital Investments in U.S. Data Centers Soar

Industrial Info Resources

Reports record actual data center construction spending of $44.7 billion in Q1 2026, with a full-year 2026 investment forecast near $700 billion.

  • $44.7 billion actual construction spending, Q1 2026
  • Full-year 2026 forecast near $700 billion
Local opposition and consent: Used by the pro side to show the buildout is not being stopped despite opposition.

Who paid: A subscription data business, paid for by the industrial companies, suppliers and investors who buy its project data. Its figures come from its own proprietary project database.

Caveat: A commercial tracker with no stated side in the debate; both sides use its numbers, the pro side for the denominator showing construction records, the anti side alongside its companion market-risk estimate.

Handbook source 29. Full citation and biography

Used by the case against market research report 2026

Data Center Market Update (Q1 2026)

New Market Pitch

Market analysis estimating that 30 to 50 percent of the 2026 data center project pipeline is on track to be delayed or canceled.

  • 30-50% of the 2026 pipeline estimated at risk of delay or cancellation
Local opposition and consent: Source of the market-risk estimate showing local opposition and other factors now threaten a large share of the pipeline.

Who paid: A commercial outfit selling market research decks. It names no owner, staff or funder anywhere on its site, which is itself the disclosure: this one cannot be checked.

Caveat: A market analysis firm without a dedicated biography in the handbook; its risk estimate is presented alongside Industrial Info Resources' construction-spending figures as the fuller picture.

Handbook source 29. Full citation and biography

Used by the case for market research tracker 2026

United States vs China AI Data Centers (2026)

AI Data Center Index

Tracking site estimating the United States holds roughly three-quarters of the world's AI computing capacity against roughly 15 percent for China.

  • US holds about 75% of world AI computing capacity vs about 15% for China
AI race and national security: Source of the headline US-versus-China AI compute capacity comparison underpinning the national security argument.

Who paid: Operated by SheetGenius, Inc., trading as Bitter Research, a small AI software company. It discloses no funders, and facilities can be submitted by users, so the industry can feed the dataset directly.

Caveat: A tracking site tallying a fast-moving target; treat exact figures as estimates, though the rough shape is consistent with academic and government estimates.

Handbook source 30. Full citation and biography

Used by the case for advocacy report 2026

China Is Coming for America's AI Edge

American Edge Project

Advocacy piece framing compute capacity as the one pillar of AI strength, alongside energy, talent, and adoption, where America still clearly leads.

  • Frames compute as the pillar America still clearly leads, of four: energy, compute, talent, adoption
AI race and national security: Provides the industry's own framing of the AI race and America's compute advantage.

Who paid: The group names no funder. Tax filings reported by CNBC, Gizmodo and the Tech Transparency Project show Meta gave it $4 million and then $34 million in consecutive years, effectively its sole funder in that period.

Caveat: An openly pro-buildout, tech-industry-funded advocacy group; its framing that compute is the pillar America leads also functions as an admission that everything else, especially the electric grid, is not leading, which cuts both ways.

Handbook source 30. Full citation and biography

Used by the case for academic report 2026

The National Security Implications of Building Frontier AI Data Centers Overseas

Brookings Institution

Analysis arguing frontier AI data center facilities need to be built on US soil rather than overseas, for national security reasons.

AI race and national security: Academic argument for keeping frontier AI infrastructure on US soil.

Who paid: Brookings states on the article that this project is supported by funding from Google.org. Google has a direct commercial interest in US policy on overseas AI data centers.

Caveat: No dedicated biography for this specific piece appears in the handbook, unlike Brookings' employment research by a different team; Brookings is generally a well-established, nonpartisan research organization.

Handbook source 30. Full citation and biography

Cited by both sides intergovernmental report 2026

Key Questions on Energy and AI

International Energy Agency, IEA

Update to the IEA's 2025 Energy and AI report. Finds world data center electricity use rose more than 15% to 485 TWh in 2025, with AI-focused facilities up about 50%, and keeps its 2030 base case near 950 TWh.

  • 485 TWh world data center electricity in 2025 (about 1.5% of world demand)
  • About 415 TWh in 2024
  • Base case about 950 TWh by 2030
  • US is about half of world installed capacity
How much electricity data centers use: the international scorekeeper's 2025 figure and 2030 base case Efficiency and forecasts: keeps its forecast rather than revising up, despite booming pipelines

Who paid: Intergovernmental agency funded by its member governments.

Caveat: Its world total is far below the Energy Institute's for the same year because the two define and model data centers differently.

Added September 16, 2026 by the daily update.

Used by the case against news news article December 17, 2025

PJM Capacity Prices Hit Record High as Grid Operator Falls Short of Reliability Target

Utility Dive staff

Reports the standby (capacity) price hitting a record $329.17, later its ceiling of $333.44, with PJM projecting 32 gigawatts of peak demand growth by 2030, all but 2 gigawatts of it from data centers.

  • Price rose from $28.92 to $329.17, hit ceiling of $333.44
  • PJM expects 32 GW peak demand growth by 2030, all but 2 GW from data centers
Capacity prices and grid costs: Source of the tenfold capacity price spike figures and PJM's own demand-growth attribution to data centers.

Who paid: Industry news publication.

Caveat: A single auction cycle's price data; PJM has also cited other contributing factors such as retiring power plants and market design.

Handbook source 18. Full citation and biography

Used by the case for government press release November 2025

SCC Issues Order on DEV Biennial Review 2025

Virginia State Corporation Commission

News release announcing the Virginia SCC's order creating a separate customer class for large electricity users over 25 megawatts, requiring fourteen-year minimum-take contracts starting January 1, 2027.

  • New customer class for users over 25 MW
  • 14-year contracts
  • Minimum 85% of wires capacity, 60% of generating capacity paid regardless of use
  • Effective January 1, 2027
Who holds the risk: Establishes binding minimum-take contract terms shifting project-failure risk onto large customers. Who pays for the grid buildout: Creates a dedicated cost-responsibility class for data centers, the concrete fix cited against cost-shifting claims.

Who paid: Official regulatory order and news release from Virginia's utility regulator.

Caveat: The referee describing its own ruling; industry says it proves big customers now pay their own way, while consumer and land-conservation groups say the commission ducked who pays for transmission lines, and the protections do not start until January 1, 2027.

Handbook source 9. Full citation and biography

Used by the case against advocacy report November 2025

Cloudy Data, Costly Deals: How Poorly States Disclose Data Center Subsidies

Good Jobs First

Report finding that states measuring the return on data center tax breaks lose between 52 and 70 cents on every dollar given away, and that fourteen states with these exemptions publish no annual total at all.

  • States that measured lose 52-70 cents per dollar given away
  • 14 states publish no annual subsidy total
State tax breaks: Source of the finding that measured states lose money on data center tax breaks, and that disclosure is poor. Secret contracts and disclosure: Documents that fourteen states publish no annual total of what they give away.

Who paid: Nonprofit founded in 1998, openly opposed to corporate subsidies; no industry funding.

Caveat: Same advocacy funding and disclosure-based methodology as the companion Good Jobs First article; counts giveaways without fully netting later collections.

Handbook source 20. Full citation and biography

Used by the case for government official document July 23, 2025

Accelerating Federal Permitting of Data Center Infrastructure

Executive Office of the President, The White House

Executive order directing federal agencies to exempt qualifying data center projects from standard federal environmental review, put them on an accelerated permitting track, and open federal land for construction.

  • Exempts qualifying projects from standard federal environmental review
Laws, bills and regulation: Formal federal action accelerating permitting for data centers. Siting, zoning and standards: The environmental-review exemption removes the step where local objections would normally get heard.

Who paid: Official federal executive order; not an independent study.

Caveat: Anti-side critics target the environmental-review exemption specifically, since that review is where local objections would otherwise be heard.

Handbook source 3. Full citation and biography

Used by the case for government official document July 2025

Winning the Race: America's AI Action Plan

The White House

Federal policy document treating data center and AI infrastructure buildout as central to the economy and national security, laying out more than ninety actions to speed construction.

  • More than 90 actions proposed to speed data center construction
Laws, bills and regulation: States federal policy favoring accelerated data center buildout. AI race and national security: Frames data center capacity as central to national security and the AI race.

Who paid: Official federal policy document issued by the executive branch; not an independent study.

Caveat: Cite as an authoritative statement of current US policy, not as independent research; its economic assumptions are contested.

Handbook source 2. Full citation and biography

Used by the case for industry report April 1, 2025

Reports and Publications

Data Center Coalition; Josh Levi (written testimony to Congress)

The data center industry's trade group's hub of reports, plus its president's April 2025 congressional testimony, presenting the industry's preferred jobs and economic-contribution figures and policy positions.

Jobs: Source of the industry's national jobs and economic-contribution figures, used alongside the PwC study. Laws, bills and regulation: Testimony to Congress advocating for the industry's policy positions.

Who paid: The industry's own trade association, funded by member dues; the studies it publishes here were commissioned by it from PwC.

Caveat: This is advocacy, not independent research; its large national figures use a counting method that includes every downstream effect, inflating totals compared with measured studies like Brookings.

Handbook source 5. Full citation and biography

Used by the case against academic study March 2025

Extracting Profits from the Public: How Utility Ratepayers Are Paying for Big Tech's Power

Eliza Martin and Ari Peskoe, Environmental and Energy Law Program, Harvard Law School

Review of nearly fifty utility regulatory cases concluding that confidentiality rules meant to share the cost of the grid are being used to shield large customer contracts from public scrutiny, allowing costs to be shifted onto ordinary ratepayers.

  • Nearly 50 regulatory cases reviewed
Secret contracts and disclosure: Central source for the argument that sealed utility contracts prevent the public from checking whether data centers pay their fair share. Who pays for the grid buildout: Argues rules built to share grid costs are being used to shift costs onto ratepayers for equipment serving a handful of large companies.

Who paid: The paper names no funder; it thanks four individual reviewers. The program says it is supported by an endowed gift, the MacIntyre Environmental Law Fund. Built on public regulatory dockets rather than purchased data.

Caveat: A process and fairness argument based on legal filings, not a dollar-figure measurement of cost-shifting; the study that did try to measure actual subsidy, E3, found no evidence of it.

Handbook source 16. Full citation and biography

Used by the case against government press release February 12, 2025

NJBPU Announces Conclusion of New Jersey's Annual Electricity Supply Auction

New Jersey Board of Public Utilities

New Jersey's state utility regulator announcing its own auction results, translating the PJM capacity price spike into roughly 17 to 20 percent residential electricity increases for the year.

  • Residential increases of about 17-20% for the year, roughly $25/month on a typical bill
Capacity prices and grid costs: Translates the wholesale capacity price spike into a concrete household bill impact, the clearest dollar figure either side has.

Who paid: Official state regulator news release.

Caveat: New Jersey was among the hardest-hit states, so its 17-20% is the high end of the range; PJM's own gentler estimate for some states is 1.5-5%.

Handbook source 27. Full citation and biography

Used by the case for news news article February 2025

Hemstreet Presents $4.7B FY 2026 Budget with 'Unprecedented' Data Center Revenues

Loudoun Now staff

Local news report on Loudoun County's proposed $4.7 billion FY2026 budget, describing data center tax revenue as unprecedented.

  • $4.7 billion FY2026 county budget
Local property tax revenue: Local press confirmation of the scale of data center revenue in the county budget.

Who paid: Independent local newsroom reporting on the county budget.

Caveat: A local news account of the county's own budget presentation; corroborates but does not independently audit the county's figures.

Handbook source 4. Full citation and biography

Used by the case for academic study February 2025

Rethinking Load Growth: Assessing the Potential for Integration of Large Flexible Loads in US Power Systems

Tyler H. Norris, Tim Profeta, Dalia Patino-Echeverri, and Adam Cowie-Haskell, Nicholas Institute for Energy, Environment and Sustainability, Duke University

Duke University study finding roughly 100 gigawatts of new demand could be added across grid regions serving about 95 percent of the country using existing infrastructure, if large loads like data centers agree to periodic curtailment at peak hours.

  • ~100 GW of new demand addable using existing infrastructure
  • Covers grid regions serving about 95% of the US
Flexible load: Source of the headline flexible-load capacity finding, the pro side's strongest argument.

Who paid: The report names no funder anywhere in its 43 pages; its acknowledgements thank two research assistants. Published by a university research center.

Caveat: The headline number depends entirely on data centers agreeing to curtail at peak times, which most contracts do not currently require; that gap is where critics push.

Handbook source 11. Full citation and biography

Used by the case against news news article 2025

Data Centers 'Primary Reason' for High PJM Capacity Prices: Market Monitor

Utility Dive staff, reporting Monitoring Analytics findings

Earlier report of the same market monitor identifying data centers as the primary reason for high PJM capacity auction prices.

Capacity prices and grid costs: Earlier confirmation from the same independent watchdog that data centers are the primary driver of capacity price increases.

Who paid: Industry news publication reporting findings from Monitoring Analytics, PJM's independent market monitor.

Caveat: Same source and limitation as the companion 2026 article: identifies cause, not bill allocation.

Handbook source 17. Full citation and biography

Used by the case against academic study 2025

Making AI Less 'Thirsty': Uncovering and Addressing the Secret Water Footprint of AI Models

Pengfei Li, Jianyi Yang, Mohammad A. Islam, and Shaolei Ren, Communications of the ACM

Peer-reviewed paper estimating that training one AI model in Microsoft's US facilities could directly evaporate on the order of 700,000 liters of clean freshwater, and projecting global AI water use at 4.2 to 6.6 billion cubic meters by 2027.

  • About 700,000 liters (about 185,000 gallons) evaporated training one model
  • Global AI water use projected at 4.2-6.6 billion cubic meters by 2027
Water use: Foundational academic source that put AI water use into public discussion, and the basis for the claim that company water disclosures undercount by excluding electricity-generation water use.

Who paid: Peer-reviewed academic research; Ren is a University of California, Riverside professor appointed to a United Nations panel on computing's environmental cost.

Caveat: The authors say plainly that estimates rely on efficiency figures the companies themselves report and on regional averages, so there is real uncertainty in the numbers.

Handbook source 23. Full citation and biography

Used by the case against news news article 2025

NAACP, Others Appeal xAI Turbine Permits for Memphis Data Center

Tennessee Lookout staff

News brief on the NAACP and allied groups formally appealing the air permits issued for xAI's Memphis gas turbines.

Air quality and on-site power: Independent news confirmation of the formal appeal against the xAI turbine permits.

Who paid: Nonprofit news outlet; independent reporting.

Caveat: A brief news report on a legal appeal; it documents the filing, not the outcome.

Handbook source 24. Full citation and biography

Used by the case against market research tracker 2025

$64 Billion of Data Center Projects Have Been Blocked or Delayed Amid Local Opposition

Data Center Watch (10a Labs)

Earlier tracker report finding $64 billion in data center projects blocked or delayed by local opposition through 2025.

  • $64 billion in projects blocked or delayed through 2025
Local opposition and consent: Baseline 2025 opposition figure against which the record Q1 2026 quarter is measured.

Who paid: Run by 10a Labs, a for-profit firm whose paying clients are AI and technology companies. Its author says this tracker is unfunded and kept separate from that paid work, which is the firm's own statement rather than an audited fact.

Caveat: Same funding structure and limitation as the Q1 2026 report; a firm serving AI industry clients has an incentive to undercount, making its own figures notable.

Handbook source 25. Full citation and biography

Used by the case against news news article 2025

PJM Capacity Prices Set Another Record with 22% Jump

Utility Dive staff

News report on a further 22 percent jump in PJM capacity auction prices, part of the multi-year run-up in standby power costs.

  • 22% jump in capacity auction prices
Capacity prices and grid costs: Documents the continued escalation of PJM capacity prices across successive auctions.

Who paid: Industry news publication.

Caveat: A single auction-cycle news report; part of a broader multi-year price trend covered across several Utility Dive articles cited in this handbook.

Handbook source 27. Full citation and biography

Used by the case for industry study 2025

Economic Contributions of Data Centers in the United States 2017-2023

PwC, PwC, prepared for the Data Center Coalition

Industry-commissioned study using input-output modeling (IMPLAN) to estimate that the data center industry supports 4.7 million American jobs, with each direct job supporting 6.5 more.

  • 4.7 million total supported jobs nationally in 2023, up from 2.9 million in 2017
  • Employment multiplier of 7.5: each direct job supported an average of 6.5 additional jobs
Jobs: Source of the industry's national jobs multiplier claim, its most-cited pro-side jobs figure.

Who paid: Prepared for the Data Center Coalition, which engaged PwC to quantify the industry's economic contribution. The jobs figure is modeled with an input-output model, not counted.

Caveat: Commissioned by the Data Center Coalition; uses input-output economic modeling that estimates ripple effects rather than counting actual jobs, and reports gross activity rather than net new jobs, unlike the measured Brookings county-level study.

Handbook source 31. Full citation and biography

Cited by both sides government report December 9, 2024

Data Centers in Virginia

Joint Legislative Audit and Review Commission, Joint Legislative Audit and Review Commission (JLARC)

The most comprehensive government study of data centers to date, from the state with more of them than anywhere on earth. Found 74,000 jobs, $5.5 billion in wages, and $9.1 billion in annual economic activity, mostly during construction, alongside costs likely to raise electricity prices for all customers and potentially double the state's electricity needs within a decade.

  • 74,000 jobs, $5.5 billion in wages, $9.1 billion in annual economic activity
  • Benefits mostly during initial construction
  • Electricity needs could double within 10 years if demand grows unchecked
  • 48 cents returned per dollar of tax exemption given up, 2014-2023
Jobs: Source of the headline jobs and wage figures both sides accept. State tax breaks: Found only 48 cents returned per dollar of tax exemption given up over a full decade. How much electricity data centers use: Found electricity needs could double within ten years if demand grows unchecked, with data centers the main driver. Capacity prices and grid costs: Found power demand will likely increase costs for all customers. Local property tax revenue: Credits real local economic contribution and jobs, mostly during construction.

Who paid: Paid for by the Virginia legislature, which ordered the study. JLARC contracted two consultants: the University of Virginia's Weldon Cooper Center for the economic analysis, and Energy and Environmental Economics (E3) for the grid and ratepayer modeling. E3 separately produced a study funded by the Data Center Coalition.

Caveat: Widely regarded as the most authoritative document in the debate; it genuinely supports both sides, so quote the whole finding, benefits and costs, rather than only the favorable half.

Handbook source 22. Full citation and biography

Cited by both sides government official document November 2024

Order Rejecting Amendments to Interconnection Service Agreement (Susquehanna co-located load), 189 FERC para 61,078

Commissioners Willie L. Phillips (Chairman), Mark C. Christie, Lindsay S. See, Federal Energy Regulatory Commission, Docket Nos. ER24-2172-000 and ER24-2172-001

The landmark ruling on plugging a data center straight into an existing nuclear plant. PJM asked federal regulators to approve changes to the interconnection agreement at the Susquehanna nuclear station so the co-located load feeding an Amazon data center campus could rise from 300 megawatts to 480. On November 1, 2024 the Commission rejected the amendment, finding PJM had not shown the non-standard provisions were necessary deviations justified by specific reliability concerns, novel legal issues or other unique factors. Read the actual order and two things stand out that the coverage usually loses. It was a request to enlarge an arrangement that already existed at 300 megawatts, not a first-time approval, and the grounds were narrow and procedural rather than a finding about who pays for the grid. Chairman Phillips dissented, Commissioner Christie concurred separately, and two commissioners did not participate. What happened next is not in this order and comes from the company: Talen announced on June 11, 2025, in a release filed with the SEC, that the co-located arrangement would move to a front-of-the-meter structure after transmission reconfigurations expected in spring 2026, with the plant selling into the PJM grid, Talen acting as retail supplier to Amazon and PPL handling delivery, under an agreement for up to 1,920 megawatts through 2042.

  • Co-located load would have risen from 300 MW to 480 MW
  • Order issued November 1, 2024
  • Chairman Phillips dissenting, Commissioner Christie concurring, two commissioners not participating
  • Restructured June 2025 as a grid-delivered agreement for up to 1,920 MW through 2042 (Talen, SEC filing)
Capacity prices and grid costs: Cited by opponents as proof regulators will not wave through behind-the-meter deals, and by supporters because the rejection turned on paperwork standards rather than a finding that the arrangement shifted costs. Who holds the risk: The central document on whether a large customer can take power directly from a plant without paying toward the shared grid. Laws, bills and regulation: Led to a separate proceeding directing PJM to write actual rules for co-located load.

Who paid: Federal regulator's own order. No funding interest; this is the decision document itself, not commentary about it.

Caveat: This order rejects one amended agreement on narrow grounds. It is not a ruling that co-location is unlawful, and it is not a finding about cost shifting. Anyone citing it for either of those claims is going further than the document does.

Used by the case for industry report 2024

DCFlex: Data Center Flexible Load Initiative

Electric Power Research Institute

Live, multi-site initiative testing whether data centers can actually reduce power use on command, expanded in 2026 to nine demonstration sites across the US and Europe with participants including major cloud, chip, and utility companies.

  • 49 participants
  • 9 demonstration sites across the US and Europe
  • Participants include Google, Meta, Microsoft, Oracle, NVIDIA, utilities, and grid operators
Flexible load: Real-world pilot evidence that flexible load is moving from theory to practice.

Who paid: Run by EPRI, a nonprofit funded by electric utility members, with named participants including Google, Meta, NVIDIA and Constellation. The demonstration sites are the participants' own data centers, so the companies being studied are also funding and designing the study.

Caveat: Participants include the same companies being tested, so this is a joint engineering effort among interested parties rather than an independent audit.

Handbook source 12. Full citation and biography

Cited by both sides government official document December 2022

Ordinance No. 5033, amending Articles II and XXII of Chapter 35 of the City Code (data centers)

Chandler City Council, City of Chandler, Arizona

The oldest tested US example of a city writing data center noise rules into binding law, adopted December 8, 2022. Its design is unusual and often misreported: the ordinance sets no decibel limit at all. Instead the applicant must hire a third-party acoustic engineer to measure the neighborhood's existing sound level before the first neighborhood meeting, and the finished building must then be designed and built so it does not exceed that baseline. Compliance is retested after occupancy and again every year for five years, with results filed with the City. The operator must notify residents within half a mile, hold two neighborhood meetings, and staff an on-site liaison from 8am to 10pm daily to take noise complaints. If backup generators are used, the operator must run a public website announcing when they will run, give 24 hours notice of any routine test, and confine testing to weekdays between 9am and 5pm.

  • No decibel cap; the standard is the measured pre-construction ambient baseline
  • Annual noise retesting for five years after occupancy
  • Neighbor notification within a half-mile radius, two neighborhood meetings
  • Generator testing limited to 9am to 5pm weekdays, announced 24 hours ahead on a public website
Noise, appearance and rural character: Supporters point to it as proof noise is a solvable permitting problem; opponents note a baseline standard only holds a site to the quiet it found, and depends on the operator's own consultants measuring it. Siting, zoning and standards: A concrete template other localities have followed, with enforceable monitoring rather than a promise.

Who paid: A city's own adopted ordinance. No funder; this is the binding local law itself.

Caveat: A relative standard is not the same as a quiet one: it prevents a facility from making an area louder than it was, measured by studies the applicant commissions. The ordinance also applies only to data centers built in a Planned Area Development district.

Cited by both sides academic study February 28, 2020

Recalibrating Global Data Center Energy-Use Estimates

Eric Masanet, Arman Shehabi, Nuoa Lei, Sarah Smith, and Jonathan Koomey, Science

Peer-reviewed study showing that widely repeated predictions of data center energy use doubling or quadrupling from roughly 2010 to 2018 were wrong, because efficiency gains held global use near flat while computing output grew enormously.

  • Global data center electricity use held near 1% of world electricity, 2010-2018, despite enormous growth in computing output
Efficiency and forecasts: Central evidence that past scary energy forecasts were wrong, due to efficiency gains. How much electricity data centers use: Anti side notes the same authors now publish projections showing steep growth, meaning the flat period has ended.

Who paid: The funding statement could not be retrieved; the journal is paywalled and the lab mirrors blocked automated access. Two authors are federal scientists at Lawrence Berkeley National Laboratory, so DOE funding is likely but unconfirmed, and we will not state it as fact until someone reads the acknowledgements.

Caveat: Covers roughly 2010 to 2018; those one-time efficiency gains from consolidating server rooms have largely been spent, and the same authors now publish projections showing steep growth, so it should be used for humility about forecasts, not as a claim that energy use is flat today.

Handbook source 13. Full citation and biography