owlfox.studioBig claims, checked.YouTube

Sources · Who Pays? · Episode 1

Who Really Pays for the AI Boom?

Every figure shown on screen in this episode, with the document, page and date it comes from, and a label for how strong the evidence is.

A data center can bring its county serious tax revenue, about $1.2 billion in Loudoun County in fiscal year 2026, while drawing on a shared grid, water and land. Who carries the other costs depends on where it is built, how it operates and what its agreements require; in one PJM capacity auction, a market monitor’s model put the extra cost to other customers at roughly $6.5 billion.

Published
17 September 2026
Length
16:21
Figures checked
26 September 2026
Documents
21
Watch the episodeJump to the documents

On screen in this episode

  • DOCUMENTED37The document shows it, and you can check it yourself.
  • CLAIMED12Someone says so, but nothing independent backs it up.
  • NOT SHOWN10Something that matters is missing from every source we checked.
  • OUR CALCULATION1Our own arithmetic from the sourced numbers. The method is below.

How we label evidence

NOT SHOWNWhat isn’t shown

  • 1:53WHAT THE STUDY DOES — AND DOES NOT — SHOW · ~15% · THIS SUPPORT WORKPLACE · EVERY AI PRODUCT · THE WHOLE ECONOMY · Not established by this result · Average issues resolved per hour in the studied workplace
  • 2:39REVENUE IS NOT THE SAME AS NET BENEFIT · REPORTED REVENUE ~$1.2bn · FY2026 property taxes · OVERALL COSTS ? · Not quantified by this figure · No net-benefit total can be inferred here
  • 6:09WHOLESALE MODEL · AUCTION ESTIMATE 2027–2028 ≠ HOUSEHOLD BILL · NOT MEASURED HERE
  • 7:07WHO CARRIES THE RISK? · PROTECTION ≠ NO RISK
  • 9:06PHOTO ≠ PROJECT PROOF · LOCATION MUST BE VERIFIED · LOCATION · DEVELOPMENT HISTORY · LINK TO PROJECT · EVIDENCE REQUIRED
  • 10:52WATER USE ≠ LAKE-LEVEL ATTRIBUTION
  • 11:02“AI DRIED THE LAKE” / NOT ESTABLISHED
  • 11:39DESIGN ≠ ENTIRE OPERATING FLEET · ELECTRICITY + MANUFACTURING STILL MATTER
  • 12:45DIFFERENT ACCOUNTING BOUNDARIES · COMPANY-WIDE ≠ AI-ONLY
  • 13:40OPERATING HEAT PLANT ≠ SERVER HEAT NOW · CURRENT OPERATION / FUTURE CONNECTION

On screen, in order

  1. 0:27

    $1.2bn · Loudoun · FY2026 reported property-tax revenue

    DOCUMENTED

    County-reported real and personal property tax revenue from data centers, fiscal year 2026.

    • [1] Loudoun County, The Loudoun Story, Loudoun’s Data Centers by the Numbers
  2. 0:50

    AI · cloud · mixed workloads · A data center is not automatically an AI-only facility

    DOCUMENTED

    The IEA reports all data centres and, separately, AI-focused data centres.

    • [2] IEA, Key Questions on Energy and AI (2026), Executive summary: data centre demand +17% and AI-focused data centres +50% in 2025
    • [3] JLARC Report 598, Data Centers in Virginia (2024), p. i, About data centers
  3. 1:00

    485 TWh · 2025 estimate · electricity · worldwide · ALL DATA CENTERS · IEA · Energy and AI, 2026

    DOCUMENTED

    IEA estimate for all data centres worldwide. The on-screen credit shortens the title of the 2026 report.

    • [2] IEA, Key Questions on Energy and AI (2026), Executive summary: 485 TWh in 2025 to 950 TWh in 2030
  4. 1:09

    485 → 950 TWh · 2025 estimate → 2030 central projection · ALL DATA CENTERS · IEA · forecast, not a meter reading (3D blocks for the two values)

    CLAIMED

    950 TWh is the IEA’s central projection for 2030, not a measurement.

    • [2] IEA, Key Questions on Energy and AI (2026), Executive summary: 485 TWh in 2025 to 950 TWh in 2030
  5. 1:17

    More than chatbots · Scope: all data centers

    DOCUMENTED

    The IEA figures cover all data centres, not only AI.

    • [2] IEA, Key Questions on Energy and AI (2026), Executive summary: data centre demand +17% and AI-focused data centres +50% in 2025
  6. 1:37

    RESEARCH / CUSTOMER SUPPORT · 5,172 · customer-support agents · One workplace setting · Brynjolfsson, Li & Raymond · Generative AI at Work

    DOCUMENTED

    Chat support agents serving one Fortune 500 software company, directly or through outsourcing firms.

    • [4] Brynjolfsson, Li & Raymond, Generative AI at Work (v2), abstract; p. 3; p. 10, Data
  7. 1:43

    STUDY RESULT / AVERAGE · Issues resolved per hour · Baseline (indexed) 100 · With AI assistance 115 · Indexed baseline = 100 · not actual issue counts

    OUR CALCULATION
    • [4] Brynjolfsson, Li & Raymond, Generative AI at Work (v2), abstract

    How we worked this out

  8. 1:45

    ~15% higher on average · Benefits varied across workers · One application. One workplace.

    DOCUMENTED

    Average rise in issues resolved per hour, with large differences between workers.

    • [4] Brynjolfsson, Li & Raymond, Generative AI at Work (v2), abstract; p. 3
  9. 1:53

    WHAT THE STUDY DOES — AND DOES NOT — SHOW · ~15% · THIS SUPPORT WORKPLACE · EVERY AI PRODUCT · THE WHOLE ECONOMY · Not established by this result · Average issues resolved per hour in the studied workplace

    NOT SHOWN

    The authors describe medium-run effects within a single firm and say the paper does not address aggregate employment or wage effects.

    • Checked[4] Brynjolfsson, Li & Raymond, Generative AI at Work (v2), p. 3
  10. 2:14

    Loudoun County · Virginia, United States · County outline: US Census Bureau · campus is illustrative

    DOCUMENTED

    Locator map. The campus model is illustrative.

    • [5] U.S. Census Bureau county boundaries (2018), Loudoun County, GEOID 51107
  11. 2:21

    LOUDOUN COUNTY / VIRGINIA · ~$1.2 billion · FISCAL YEAR 2026 · Real + personal property tax revenue · County-reported figure · Loudoun County

    DOCUMENTED

    The county’s own figure, from its web page.

    • [1] Loudoun County, The Loudoun Story, Loudoun’s Data Centers by the Numbers
  12. 2:39

    REVENUE IS NOT THE SAME AS NET BENEFIT · REPORTED REVENUE ~$1.2bn · FY2026 property taxes · OVERALL COSTS ? · Not quantified by this figure · No net-benefit total can be inferred here

    NOT SHOWN

    The county reports gross revenue; no net figure after costs is given.

    • Checked[1] Loudoun County, The Loudoun Story
  13. 2:44

    RESULTS DEPEND ON THE LOCAL TAX RULES · LOUDOUN ~$1.2bn · FY2026 property taxes · ANOTHER COUNTY ? · Different rules · unknown result · Loudoun’s result cannot simply be copied elsewhere

    DOCUMENTED

    JLARC: local data center revenue depends on factors such as the size of the local market and local tax rates.

    • [3] JLARC Report 598, Data Centers in Virginia (2024), p. i
  14. 2:48

    WHY A COUNTY MIGHT WELCOME THE CAMPUS · A valuable tax base · Buildings · Equipment · Reported revenue is substantial; total costs still matter

    DOCUMENTED

    JLARC: local revenue comes primarily from business personal property and real property taxes.

    • [3] JLARC Report 598, Data Centers in Virginia (2024), p. i
    • [1] Loudoun County, The Loudoun Story, Loudoun’s Data Centers by the Numbers
  15. 3:18

    Source: JLARC · Data Centers in Virginia (2024) · CONSTRUCTION WORK (sign on the model crane)

    DOCUMENTED

    Narration: much of the industry’s economic benefit came from construction rather than ongoing operations.

    • [3] JLARC Report 598, Data Centers in Virginia (2024), p. i; p. 15
  16. 4:01

    VIRGINIA / JANUARY 2026 · MODELED FISCAL RETURN · Positive · State + local combined · REPORT SUMMARY / NOT A FACSIMILE

    DOCUMENTED

    Five-year model for investments in fiscal years 2024–2025: estimated state and local tax revenue exceeds the exemption’s value by $2.1 billion.

    • [6] Virginia Tax and VEDP, data center exemption report, p. 4; Table 2, pp. 4–5
  17. 4:10

    WITH INCENTIVE · ILLUSTRATIVE SCENARIO · WITHOUT INCENTIVE · ILLUSTRATIVE SCENARIO

    DOCUMENTED

    The scenes are illustrations. The report assumes, after JLARC, that 90% of this investment would not have come to Virginia without the exemption.

    • [6] Virginia Tax and VEDP, data center exemption report, pp. 5–6; Table 3
  18. 4:15

    MODEL BOUNDARY · LOCAL INCENTIVES · EXCLUDED

    DOCUMENTED

    Local discretionary incentives are not included in the model.

    • [6] Virginia Tax and VEDP, data center exemption report, p. 7, Assumptions; p. 9, Limitations
  19. 5:36

    Source: Monitoring Analytics · PJM market monitor · PJM · CAPACITY AUCTION · 2027–2028 DELIVERY YEAR

    DOCUMENTED

    Base Residual Auction for the 2027/2028 delivery year, held 4–10 December 2025.

    • [7] Market monitor, 2027/2028 PJM auction analysis, p. 1
  20. 5:57

    MODELED DIFFERENCE · DATA CENTER DEMAND · COSTS FOR OTHER CUSTOMERS · ~$6.5 BILLION · vs demand excluded · 2027–2028 / WHOLESALE

    DOCUMENTED

    $6,497,653,512 (Scenario 4), shown rounded: the auction cleared again without existing and forecast data center load.

    • [7] Market monitor, 2027/2028 PJM auction analysis, p. 3, Table 1; p. 5
  21. 6:09

    WHOLESALE MODEL · AUCTION ESTIMATE 2027–2028 ≠ HOUSEHOLD BILL · NOT MEASURED HERE

    NOT SHOWN

    The analysis models one capacity auction; it does not measure household bills.

    • Checked[7] Market monitor, 2027/2028 PJM auction analysis
  22. 6:29

    OHIO · 2025 · AEP OHIO / TARIFF · SUMMARY / NOT A FACSIMILE · Source: PUCO · AEP Ohio tariff approval (2025)

    DOCUMENTED

    On 9 July 2025 PUCO adopted a settlement and ordered AEP Ohio to file data center tariffs.

    • [8] PUCO announcement, 9 July 2025
  23. 6:45

    MINIMUM PAYMENTS · FINANCIAL PROTECTIONS

    CLAIMED

    The terms (minimum monthly demand charges and collateral, in effect from 23 July 2025) are as AEP Ohio describes its tariff. PUCO's announcement confirms the tariff order but does not list the terms, and the order itself could not be opened.

    • [9] AEP Ohio, Data Center Tariff page, Minimum Demand Charges; Collateral Requirement
    • [8] PUCO announcement, 9 July 2025
  24. 7:07

    WHO CARRIES THE RISK? · PROTECTION ≠ NO RISK

    NOT SHOWN

    The documents describe protections; they do not show that every future cost is covered.

    • Checked[8] PUCO announcement, 9 July 2025
    • [9] AEP Ohio, Data Center Tariff page
  25. 7:20

    Source: IEA · Energy and AI · SPARE SUPPLY · DIFFERENT STARTING CONDITION

    DOCUMENTED

    IEA: with excess supply, load growth can lower prices; in tighter systems it can raise them.

    • [2] IEA, Key Questions on Energy and AI (2026), Executive summary, section on data centres and electricity prices
  26. 7:49

    AIR POLLUTION · GREENHOUSE GASES · WATER USE · WASTE

    DOCUMENTED
    • [10] EPA, electricity system and the environment, Environmental Impacts of the Electricity System
  27. 7:56

    GENERATION · ACCESS ROAD · FUEL ROUTE · POWER LINES

    DOCUMENTED

    Power plants may need access roads, fuel pipelines or railroads, and transmission lines.

    • [11] EIA, electricity and the environment, The effect of power plants on the landscape
    • [10] EPA, electricity system and the environment, Environmental Impacts of the Electricity System
  28. 8:09

    Source: IEA · United States · GROWING DEMAND · Includes data centers · RETIREMENT TIMING · NOT THE ONLY FACTOR

    DOCUMENTED

    IEA: the U.S. government has moved to delay coal plant retirements in response to strong demand growth, including from data centres.

    • [12] IEA, Electricity 2026: Supply, United States
  29. 8:39

    HABITAT CROSSING · COLLISION RISK · ELECTROCUTION RISK · SOME CONFIGURATIONS · Illustrative risk · no actual incident depicted

    DOCUMENTED

    General risks; no actual incident is shown.

    • [13] USFWS, power line measures for birds, How do power lines impact migratory birds?
    • [11] EIA, electricity and the environment, Electric power lines and other distribution infrastructure
  30. 8:47

    Source: U.S. Fish & Wildlife Service · avian protection · RISK REDUCTION · SITING + DESIGN

    DOCUMENTED

    Siting lines outside high-risk areas and avian-safe designs.

    • [13] USFWS, power line measures for birds, What can I do to avoid and minimize impacts to birds?
  31. 9:06

    PHOTO ≠ PROJECT PROOF · LOCATION MUST BE VERIFIED · LOCATION · DEVELOPMENT HISTORY · LINK TO PROJECT · EVIDENCE REQUIRED

    NOT SHOWN

    No habitat loss from a specific AI project is documented in this episode; the card lists the evidence such a claim would need.

  32. 9:29

    SERVER HEAT · EVAPORATION · LEAVES LOCAL SUPPLY

    DOCUMENTED

    Evaporation in cooling towers raises concerns about local water consumption and availability.

    • [14] Berkeley Lab, 2024 U.S. data center energy report, pp. 40–41, Table 4.2; p. 56
  33. 9:39

    WITHDRAWAL · CONSUMPTION · RETURNED

    DOCUMENTED

    Consumption is withdrawn water permanently removed from the immediate water cycle.

    • [14] Berkeley Lab, 2024 U.S. data center energy report, p. 56, definition of water consumption
  34. 9:49

    ELECTRICITY GENERATION · WATER CONSUMED ELSEWHERE

    DOCUMENTED

    Indirect water use occurs at the power plants that supply the electricity.

    • [14] Berkeley Lab, 2024 U.S. data center energy report, pp. 56–57
  35. 9:54

    US DATA CENTERS / 2023 · BERKELEY LAB ESTIMATES / ALL DATA CENTERS · DIRECT · 66 bn litres

    DOCUMENTED

    Direct water consumption, all U.S. data centers, 2023.

    • [14] Berkeley Lab, 2024 U.S. data center energy report, p. 55; Figure 5.9, p. 56
  36. 10:01

    ELECTRICITY SUPPLY · ~800 bn litres · MODELED INDIRECT USE

    DOCUMENTED

    Nearly 800 billion liters, modeled from the regional grid mix; purchase agreements and on-site generation are not included.

    • [14] Berkeley Lab, 2024 U.S. data center energy report, pp. 57–58
  37. 10:24

    GREAT SALT LAKE / UTAH · NASA GODDARD / LANDSAT / 1984–2023 (satellite time series of the lake)

    DOCUMENTED
    • [15] NASA SVS 15020, The Shrinking Great Salt Lake
  38. 10:38

    UPSTREAM DIVERSIONS + CHANGING CLIMATE

    DOCUMENTED

    NASA names upstream water diversions and a shifting climate as the drivers of the decline.

    • [15] NASA SVS 15020, The Shrinking Great Salt Lake
  39. 10:43

    WATER-USE RECORDS · 7 / 22 facilities · 2024 PUBLIC-RECORDS INVESTIGATION · Source: Aspen Tech Policy Hub · Utah (2024) (7 of 22 tiles highlighted)

    DOCUMENTED

    Records obtained for 7 of the 22 data centers in the basin; about 15 others have unknown water use.

    • [16] Aspen Policy Academy, Great Salt Lake fact sheet, p. 1, Overview; p. 3
  40. 10:52

    WATER USE ≠ LAKE-LEVEL ATTRIBUTION

    NOT SHOWN

    The fact sheet reports water use; it does not estimate any effect on the lake’s level.

    • Checked[16] Aspen Policy Academy, Great Salt Lake fact sheet
    • [15] NASA SVS 15020, The Shrinking Great Salt Lake
  41. 10:57

    SAMPLE ≠ GENERATIVE AI ONLY

    DOCUMENTED

    The sample includes the NSA’s Utah Data Center.

    • [16] Aspen Policy Academy, Great Salt Lake fact sheet, pp. 2–3
  42. 11:02

    “AI DRIED THE LAKE” / NOT ESTABLISHED

    NOT SHOWN

    Neither source attributes the lake’s decline to AI or data centers.

    • Checked[15] NASA SVS 15020, The Shrinking Great Salt Lake
    • [16] Aspen Policy Academy, Great Salt Lake fact sheet
  43. 11:24

    COLD PLATE / CHIP · MICROSOFT / NEW DESIGNS · INTENDED TO AVOID COOLING EVAPORATION

    CLAIMED

    Microsoft describes a new AI datacenter design that cools chips in a closed loop without water evaporation.

    • [17] Microsoft blog, water intensity and cooling, Early water stewardship by design; Datacenter cooling methods, explained
  44. 11:34

    HEAT TO OUTDOOR AIR · COMPANY DESCRIPTION / ILLUSTRATIVE MECHANISM

    CLAIMED

    Microsoft: air-cooled chillers remove heat from closed coolant loops using outside air.

    • [17] Microsoft blog, water intensity and cooling, Datacenter cooling methods, explained
  45. 11:39

    DESIGN ≠ ENTIRE OPERATING FLEET · ELECTRICITY + MANUFACTURING STILL MATTER

    NOT SHOWN

    The zero-water design is for new AI datacenters; the post does not address water used to generate the electricity or make the equipment.

    • Checked[17] Microsoft blog, water intensity and cooling
  46. 11:47

    LIQUID AT THE CHIP · ONE STAGE IN THE SYSTEM · FINAL HEAT RELEASE

    CLAIMED

    Closed-loop, direct-to-chip cooling as Microsoft describes it.

    • [17] Microsoft blog, water intensity and cooling, Datacenter cooling methods, explained
  47. 12:10

    RESOURCES PER TASK · TASK VOLUME + COMPLEXITY

    DOCUMENTED

    IEA: energy use per AI query has fallen sharply while more energy-intensive uses spread.

    • [2] IEA, Key Questions on Energy and AI (2026), Executive summary: energy per AI query and energy-intensive use cases
  48. 12:21

    GOOGLE / 2026 ENVIRONMENTAL REPORT · COMPANY-REPORTED 2025 DATA / YEAR-ON-YEAR · ELECTRICITY USE +37% · TOTAL COMPANY ELECTRICITY · 2024 = 100 / EACH OWN BASELINE (paired bars)

    CLAIMED

    Google’s total electricity consumption grew 37% in 2025.

    • [18] Google 2026 Environmental Report, p. 19

    How we worked this out

  49. 12:32

    OPERATIONAL EMISSIONS −2% · SCOPE 1 + MARKET-BASED SCOPE 2

    CLAIMED
    • [18] Google 2026 Environmental Report, pp. 74–75

    How we worked this out

  50. 12:40

    AMBITION-BASED EMISSIONS +18% · BROADER COMPANY MEASURE

    CLAIMED

    Google attributes the growth mainly to supply-chain activities; its supply-chain emissions rose 25% in 2025.

    • [18] Google 2026 Environmental Report, pp. 74–75

    How we worked this out

  51. 12:45

    DIFFERENT ACCOUNTING BOUNDARIES · COMPANY-WIDE ≠ AI-ONLY

    NOT SHOWN

    The electricity and emissions figures shown are company-wide; they are not an AI-only footprint.

    • Checked[18] Google 2026 Environmental Report
  52. 12:57

    MANUFACTURE · USE · REPLACEMENT · DISPOSAL · Source: Wang et al. · Nature Computational Science (2024) · FUTURE E-WASTE · DEPENDS ON GROWTH + EQUIPMENT LIFE

    DOCUMENTED

    Scenario study: projected e-waste varies with AI growth settings and server turnover.

    • [19] Wang et al., Nature Computational Science (2024), abstract
  53. 13:20

    FINLAND / FORTUM + MICROSOFT · ESPOO + KIRKKONUMMI / DISTRICT HEATING

    CLAIMED

    Heat pump plants at Kolabacken (Kirkkonummi) and Hepokorpi (Espoo), the sites of two Microsoft data centres.

    • [20] Fortum news, 6 May 2026
  54. 13:27

    FORTUM UPDATE / MAY 2026 · HEAT PRODUCTION STARTED · USING OTHER SOURCES

    CLAIMED

    On 6 May 2026 the plants ran on ambient-air heat pumps and electric boilers.

    • [20] Fortum news, 6 May 2026
  55. 13:34

    DATA CENTER WASTE HEAT · PLANNED IN STAGES / FROM 2027

    CLAIMED

    The 6 May 2026 release puts the start of waste-heat recovery in the following year, in stages.

    • [20] Fortum news, 6 May 2026
  56. 13:40

    OPERATING HEAT PLANT ≠ SERVER HEAT NOW · CURRENT OPERATION / FUTURE CONNECTION

    NOT SHOWN

    On 6 May 2026 the plants ran on other sources; recovery of data centre heat was to start the following year.

    • Checked[20] Fortum news, 6 May 2026
  57. 14:22

    PJM / LARGE-LOAD FORECAST SCREENING · ILLUSTRATIVE PLANNING / NOT A FORECAST · Source: PJM · 2026 forecast update · ANNOUNCED · OPERATING

    CLAIMED

    PJM says it vets large-load requests for how likely they are and for double counting, and derates non-firm projects.

    • [21] PJM Inside Lines, 2026 load forecast, Improving the Forecast for Large Load Additions
  58. 15:07

    WHO PAYS FOR THE AI BOOM? · INVESTORS + CUSTOMERS · HOST COMMUNITIES CAN BENEFIT

    DOCUMENTED

    Recap. The film does not measure how total costs divide between these groups.

    • [2] IEA, Key Questions on Energy and AI (2026), Executive summary: capital expenditure and capital-market funding
    • [1] Loudoun County, The Loudoun Story
    • [3] JLARC Report 598, Data Centers in Virginia (2024), p. i
  59. 15:24

    OTHER ELECTRICITY USERS

    DOCUMENTED

    Recap of the modeled auction result at 5:57.

    • [7] Market monitor, 2027/2028 PJM auction analysis, p. 3, Table 1
  60. 15:27

    SHARED WATER + LANDSCAPES · LOCAL PRESSURES / ILLUSTRATION

    DOCUMENTED

    Recap; the scene is an illustration.

    • [14] Berkeley Lab, 2024 U.S. data center energy report, pp. 55–57
    • [10] EPA, electricity system and the environment, Environmental Impacts of the Electricity System

Also said in the narration

  1. 2:14

    Loudoun County, in Virginia, is a major data center hub.

    DOCUMENTED

    JLARC: Northern Virginia is the largest data center market in the world.

    • [3] JLARC Report 598, Data Centers in Virginia (2024), p. i
    • [1] Loudoun County, The Loudoun Story
  2. 3:56

    Virginia offers qualifying data centers tax exemptions.

    DOCUMENTED

    Sales and use tax exemption for computer equipment in data centers whose operators sign a memorandum of understanding with VEDP on minimum investment and job creation.

    • [6] Virginia Tax and VEDP, data center exemption report, p. 2

Our calculations

  1. OUR CALCULATIONIndexed study result

    The study reports a 15% average increase in issues resolved per hour. On an index where the comparison group is 100, that is 100 × 1.15 = 115. The bars show these index values, not counts of issues.

    • [4] Brynjolfsson, Li & Raymond, Generative AI at Work (v2), abstract

  2. OUR CALCULATIONGoogle bars on a 2024 = 100 scale

    Each bar pair sets that measure’s own 2024 value to 100. Google’s reported changes give 2025 levels of 137 for electricity use (+37%), 98 for operational emissions (−2%) and 118 for ambition-based emissions (+18%). The three measures have different units and boundaries, so the pairs are not comparable with each other.

    • [18] Google 2026 Environmental Report, p. 19; pp. 74–75

Documents

  • The video’s figures are as of 16 September 2026. Every figure on this page was checked against its source on 26 September 2026.
  • Page numbers are the documents’ own printed page numbers.
  • The campus, grid and water scenes are illustrative 3D models, not specific sites. The video carries no evidence labels on screen; this page gives each figure its label.
  1. [1]

    Loudoun County, Virginia

    Data Centers: The Loudoun Story

    Retrieved 26 September 2026

    The county updates this page. Checked on 26 September 2026.

    Used at 0:27 · 2:14 · 2:21 · 2:39 · 2:48 · 15:07

  2. [2]

    International Energy Agency (IEA)

    Key Questions on Energy and AI — Executive summary

    Published 16 April 2026 · Retrieved 26 September 2026

    Licence: CC BY 4.0

    Used at 0:50 · 1:00 · 1:09 · 1:17 · 7:20 · 12:10 · 15:07

  3. [3]

    Joint Legislative Audit and Review Commission (JLARC), Virginia

    Data Centers in Virginia (JLARC Report 598)

    Published 9 December 2024 · Retrieved 26 September 2026

    Cover date. The PDF file was created on 7 March 2025.

    Page numbers refer to the full report, not the separate summary PDF.

    Used at 0:50 · 2:14 · 2:44 · 2:48 · 3:18 · 15:07

  4. [4]

    arXiv

    Generative AI at Work (Erik Brynjolfsson, Danielle Li and Lindsey Raymond · arXiv:2304.11771v2)

    Published 6 November 2024 · Retrieved 26 September 2026

    Version 2, posted on 6 November 2024; the PDF is dated 7 November 2024.

    The journal version reports the same 5,172 agents and 15% average.

    Used at 1:37 · 1:43 · 1:45 · 1:53

  5. [5]

    U.S. Census Bureau

    Cartographic Boundary Files: 2018 counties, 1:20,000,000 (cb_2018_us_county_20m)

    Retrieved 26 September 2026

    Source of the Loudoun County outline in the video.

    Used at 2:14

  6. [6]

    Virginia Department of Taxation and Virginia Economic Development Partnership (VEDP)

    Virginia Tax Exemptions for Data Centers: Biennial Data Center Retail Sales and Use Tax Exemption Report (RD40 (2026))

    Published 2 January 2026 · Retrieved 26 September 2026

    Covers fiscal years 2024 and 2025. Jobs and investment were reported by the operators and not independently validated (p. 3).

    Used at 3:56 · 4:01 · 4:10 · 4:15

  7. [7]

    Monitoring Analytics, Independent Market Monitor for PJM

    Analysis of the 2027/2028 RPM Base Residual Auction – Part A

    Published 5 January 2026 · Retrieved 26 September 2026

    Cover date. The PDF file was created on 7 April 2026.

    Sensitivity analysis: the auction cleared again without data center load. Results are modeled, not measured bills.

    Used at 5:36 · 5:57 · 6:09 · 15:24

  8. [8]

    Public Utilities Commission of Ohio (PUCO)

    PUCO orders AEP Ohio to create data center specific tariff (Case 24-508-EL-ATA)

    Published 9 July 2025 · Retrieved 26 September 2026

    Used at 6:29 · 6:45 · 7:07

  9. [9]

    AEP Ohio

    Data Center Tariff

    Retrieved 15 September 2026

    The page refused connections on 26 September 2026; the terms were checked against the archived copy of 15 September 2026.

    Used at 6:45 · 7:07

  10. [10]

    U.S. Environmental Protection Agency (EPA)

    About the U.S. Electricity System and its Impact on the Environment

    Published 18 February 2026 · Retrieved 26 September 2026

    The page shows this as its last-updated date.

    Used at 7:49 · 7:56 · 15:27

  11. [11]

    U.S. Energy Information Administration (EIA)

    Electricity explained: Electricity and the environment

    Published 16 April 2024 · Retrieved 26 September 2026

    The page shows this as its last-updated date.

    Used at 7:56 · 8:39

  12. [12]

    International Energy Agency (IEA)

    Electricity 2026 — Supply

    Published 6 February 2026 · Retrieved 26 September 2026

    Licence: CC BY 4.0

    Used at 8:09

  13. [13]

    U.S. Fish & Wildlife Service, Migratory Bird Program

    Avoidance and Minimization Measures: Power Lines

    Retrieved 26 September 2026

    Used at 8:39 · 8:47

  14. [14]

    Lawrence Berkeley National Laboratory

    2024 United States Data Center Energy Usage Report (LBNL-2001637 · Shehabi et al.)

    Published December 2024 · Retrieved 26 September 2026

    Estimates for all U.S. data centers.

    Used at 9:29 · 9:39 · 9:49 · 9:54 · 10:01 · 15:27

  15. [15]

    NASA Scientific Visualization Studio

    The Shrinking Great Salt Lake (SVS ID 15020)

    Published 6 May 2026 · Retrieved 26 September 2026

    Natural-color Landsat time series, 1984–2023.

    Licence: Credit: NASA’s Goddard Space Flight Center

    Used at 10:24 · 10:38 · 10:52 · 11:02

  16. [16]

    Aspen Policy Academy

    Fact sheet: Great Salt Lake Basin Data Center Water Use (Jonathan Gilmour, Rebecca Kilberg and Mary-Clare Bosco)

    Published 2024 · Retrieved 26 September 2026

    The fact sheet carries no date; the file name says 2024 and the PDF was created on 10 December 2024.

    Named on screen as the Aspen Tech Policy Hub, the Academy’s predecessor. Water-use data gathered through public records requests.

    Used at 10:43 · 10:52 · 10:57 · 11:02

  17. [17]

    Microsoft (The Official Microsoft Blog)

    Inside Microsoft’s two-decade push to cut water intensity while scaling for growth

    Published 24 June 2026 · Retrieved 26 September 2026

    Used at 11:24 · 11:34 · 11:39 · 11:47

  18. [18]

    Google

    2026 Environmental Report

    Published 2026 · Retrieved 26 September 2026

    The report page gives no date; the PDF was created on 24 June 2026. Data for fiscal year 2025.

    Company-reported figures.

    Used at 12:21 · 12:32 · 12:40 · 12:45

  19. [19]

    Nature Computational Science

    E-waste challenges of generative artificial intelligence (Peng Wang, Ling-Yu Zhang, Asaf Tzachor and Wei-Qiang Chen · vol. 4, p. 818 · doi:10.1038/s43588-024-00712-6)

    Published 28 October 2024 · Retrieved 26 September 2026

    Published online 28 October 2024, in the November 2024 issue.

    Used at 12:57

  20. [20]

    Fortum

    Fortum has started heat production at two large data centre sites in Finland

    Published 6 May 2026 · Retrieved 26 September 2026

    Used at 13:20 · 13:27 · 13:34 · 13:40

  21. [21]

    PJM Interconnection (PJM Inside Lines)

    PJM’s Updated 20-Year Forecast Continues To See Significant Long-Term Load Growth

    Published 14 January 2026 · Retrieved 26 September 2026

    Used at 14:22

Corrections

  1. 26 September 2026 · 1:00

    Was: The on-screen source lines at 1:00 and 7:20 credit the IEA figures to “Energy and AI, 2026”.

    Now: The report is the IEA’s “Key Questions on Energy and AI” (2026). “Energy and AI” is a separate IEA report from 2025.

    Corrected in: The episode’s source page and the video description

  2. 26 September 2026 · 3:18

    Was: The narration says the 2024 report on data centers came from “Virginia’s Legislative Research Commission”.

    Now: The report is by Virginia’s Joint Legislative Audit and Review Commission (JLARC). The on-screen source line names it correctly.

    Corrected in: The episode’s source page and the video description

  3. 26 September 2026 · 10:43

    Was: The source line credits the Utah water-use fact sheet to the “Aspen Tech Policy Hub”.

    Now: The fact sheet is published by the Aspen Policy Academy.

    Corrected in: The episode’s source page and the video description

Found an error? Email [email protected] with the time stamp and your source, or leave a comment under the video.

Data and licences

  • IEA: the 3D chart of data centre electricity use (2025 estimate and 2030 projection) is a work derived by owlfox.studio from IEA material and owlfox.studio is solely liable and responsible for this derived work. The derived work is not endorsed by the IEA or its Member countries in any manner. Source: IEA 2026, Key Questions on Energy and AI, iea.org/reports/key-questions-on-energy-and-ai, Licence: CC BY 4.0 (creativecommons.org/licenses/by/4.0).
  • Great Salt Lake time series: NASA’s Goddard Space Flight Center, Landsat 1984–2023 (NASA Scientific Visualization Studio, svs.gsfc.nasa.gov/15020).
  • Loudoun County outline: U.S. Census Bureau, cartographic boundary file cb_2018_us_county_20m.