Real Safety AI Foundation / Research

Policy, Enforcement, and Implementation

No Ceiling, No Denominator, No Exit: Artificial Intelligence and the Architecture of American Energy Law

Publisher: Real Safety AI Foundation

Research article.

Written
July 2026
Version
v1.6
Pages
49
Download PDFPlain language slides (PDF)Export BibTeXExport RIS

Abstract

For half a century the United States has run a federal program to make machines use less energy. It has never once run a program to make the country use less energy. The Energy Policy and Conservation Act regulates efficiency, which is a ratio, and the statute contains no authority to set a ceiling on aggregate consumption. That distinction was tolerable while the ratio and the total moved together. Artificial intelligence has pulled them apart. Google reports that the energy required to answer a median Gemini text prompt fell by a factor of 33 in twelve months, and reports in the same year that its electricity demand rose 37 percent, the largest annual increase in the company’s history. This Article argues that the gap is structural, and that it appears three times. First, coverage. The gate through which a new product enters the consumer program at 42 U.S.C. § 6292(b)(1)(B) is a fraction with households in the denominator, and the standard-setting gate at § 6295(l)(1) is metered the same way, down to an aggregate figure that is expressly an aggregate of household use. When the Department of Energy proposed in 2013 to cover computer servers, it satisfied the statutory test by computing the energy used by servers in houses. Industry objected that the agency lacked authority over the commercial units, the agency withdrew the determination, and no coverage determination for servers exists today. The same agency, reading a 1978 definition, built the regulatory category of computer room air conditioner out of statutory text that never named it, and now sets mandatory efficiency floors on the machine that cools the data center while setting nothing on the machines inside it. Second, disclosure. Google publishes watt-hours per prompt and withholds the prompt count it used as the divisor, and its chief scientist has said on the record that the company is not comfortable revealing that number. OpenAI has released no model-specific energy data since 2020. The aggregate is therefore not merely unregulated; it is uncomputable by anyone outside the firms. Third, allocation. Compute is financed by risk capital that shareholders bear, and the grid that serves it is financed by a rate base that households cannot leave. The Article makes three contributions. It identifies the household denominator as the mechanism of EPCA’s blindness rather than treating the absence of AI standards as agency inattention. It shows that the withheld denominator in corporate environmental disclosure reproduces the statute’s own epistemic structure, so that the regime and the firm converge on publishing the quotient and suppressing the total. And it locates the resulting cost-allocation question in existing doctrine, where the term of art means what the Seventh Circuit says it means and not what the industry hears.

Keywords

  • Energy Policy and Conservation Act
  • energy conservation standards
  • coverage determination
  • data centers
  • artificial intelligence
  • cost causation
  • Illinois Commerce Commission v. FERC
  • Loper Bright Enterprises v. Raimondo
  • Jevons paradox
  • environmental disclosure

Plain language slides

First slide of the plain language summary of No Ceiling, No Denominator, No Exit: Artificial Intelligence and the Architecture of American Energy LawOpen the 26-slide summary (PDF)

Suggested citation

Gilly, Travis. "No Ceiling, No Denominator, No Exit: Artificial Intelligence and the Architecture of American Energy Law." Real Safety AI Foundation Research Article, July 2026. https://realsafetyai.org/research/no-ceiling/

Other versions

This paper is also posted on SSRN.

SSRN version

References (81)

This paper cites its sources in footnotes. Each authority is listed once, where it is first cited, with its footnote number.

  1. Footnote 1.Energy Policy and Conservation Act, Pub. L. No. 94-163, 89 Stat. 871 (1975) (codified as amended at 42 U.S.C. §§ 6291-6317).
  2. Footnote 1.Energy Conservation Program: Procedures, Interpretations, and Policies for Consideration of New or Revised Energy Conservation Standards and Test Procedures for Consumer Products and Certain Commercial/Industrial Equipment, 91 Fed. Reg. 42,034, 42,037 (proposed July 7, 2026) (to be codified at 10 C.F.R. pt. 430) [hereinafter 2026 Process Rule NOPR].
  3. Footnote 3.42 U.S.C. § 6295(o)(2)(A)
  4. Footnote 5.Google, Measuring the Environmental Impact of Delivering AI at Google Scale 1 (2025), https://arxiv.org/abs/2508.15734 (last visited July 15, 2026) [hereinafter Google AI Scale].
  5. Footnote 5.Measuring the Environmental Impact of AI Inference, Google Cloud Blog (Aug. 21, 2025), https://cloud.google.com/blog/products/infrastructure/measuring-the-environmental-impact-of-ai-inference (last visited July 15, 2026) [hereinafter AI Inference Blog].
  6. Footnote 6.Read Google’s 2026 Environmental Report, Google Blog (June 30, 2026), https://blog.google/company-news/outreach-and-initiatives/sustainability/2026-environmental-report/ (last visited July 15, 2026) [hereinafter Google 2026 Report Blog].
  7. Footnote 7.Heather Clancy, Google’s AI Expansion Weighs Heavily on Climate Goals, Trellis (June 30, 2026), https://trellis.net/article/googles-ai-expansion-weighs-heavily-on-climate-goals/ (last visited July 15, 2026).
  8. Footnote 9.Arman Shehabi et al., Data Center Growth in the United States: Decoupling the Demand for Services from Electricity Use, 13 Env’t Rsch. Letters 124030 (2018), https://doi.org/10.1088/1748-9326/aaec9c [hereinafter Shehabi, Decoupling]
  9. Footnote 9.Eric Masanet et al., Recalibrating Global Data Center Energy-Use Estimates, 367 Science 984 (2020), https://doi.org/10.1126/science.aba3758 [hereinafter Masanet, Recalibrating].
  10. Footnote 10.42 U.S.C. § 6292(b)(2)(A).
  11. Footnote 13.Casey Crownhart, Google’s Still Not Giving Us the Full Picture on AI Energy Use, MIT Tech. Rev. (Aug. 28, 2025), https://www.technologyreview.com/2025/08/28/1122685/ai-energy-use-gemini/ (last visited July 15, 2026) (quoting Jeff Dean: “We’re not comfortable revealing that for various reasons,” and reporting his explanation that the total “is an abstract measure that changes over time” and that “the company wants users to be thinking about the energy usage per prompt”).
  12. Footnote 14.W. Stanley Jevons, The Coal Question: An Inquiry Concerning the Progress of the Nation, and the Probable Exhaustion of Our Coal-Mines (1865).
  13. Footnote 15.J. Daniel Khazzoom, Economic Implications of Mandated Efficiency in Standards for Household Appliances, 1 Energy J. 21, 21 (1980) (“[T]he estimates of energy savings predicted to result from these mandated standards are derived mechanically. When mandated standards raise the appliance efficiency by 1 percent, demand is predicted to drop by 1 percent; when they raise efficiency by 2 percent, demand is predicted to drop by 2 percent; and so on.”).
  14. Footnote 16.Harry D. Saunders, The Khazzoom-Brookes Postulate and Neoclassical Growth, 13 Energy J. 131 (1992).
  15. Footnote 16.J. Daniel Khazzoom, Energy Saving Resulting from the Adoption of More Efficient Appliances, 8 Energy J. 85 (1987)
  16. Footnote 16.Amory B. Lovins, Energy Saving Resulting from the Adoption of More Efficient Appliances: Another View, 9 Energy J. 155 (1988)
  17. Footnote 16.J. Daniel Khazzoom, Energy Savings from More Efficient Appliances: A Rejoinder, 10 Energy J. 157 (1989).
  18. Footnote 17.Steve Sorrell, Jevons’ Paradox Revisited: The Evidence for Backfire from Improved Energy Efficiency, 37 Energy Pol’y 1456 (2009)
  19. Footnote 17.Lorna A. Greening, David L. Greene & Carmen Difiglio, Energy Efficiency and Consumption: The Rebound Effect, a Survey, 28 Energy Pol’y 389 (2000).
  20. Footnote 18.Alexandra Sasha Luccioni, Emma Strubell & Kate Crawford, From Efficiency Gains to Rebound Effects: The Problem of Jevons’ Paradox in AI’s Polarized Environmental Debate, in Proceedings of the 2025 ACM Conference on Fairness, Accountability, and Transparency 76 (2025), https://doi.org/10.1145/3715275.3732007.
  21. Footnote 19.Peter Dauvergne, Is Artificial Intelligence Greening Global Supply Chains? Exposing the Political Economy of Environmental Costs, 29 Rev. Int’l Pol. Econ. 696, 696 (2022), https://doi.org/10.1080/09692290.2020.1814381.
  22. Footnote 21.Emma Strubell, Ananya Ganesh & Andrew McCallum, Energy and Policy Considerations for Deep Learning in NLP, in Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics 3645 (2019), https://doi.org/10.18653/v1/P19-1355.
  23. Footnote 22.Eric Masanet et al., Recalibrating Global Data Center Energy-Use Estimates, 367 Science 984 (2020).
  24. Footnote 23.Alexandra Sasha Luccioni, Yacine Jernite & Emma Strubell, Power Hungry Processing: Watts Driving the Cost of AI Deployment?, in Proceedings of the 2024 ACM Conference on Fairness, Accountability, and Transparency 85 (2024), https://doi.org/10.1145/3630106.3658542.
  25. Footnote 24.Sasha Luccioni, Boris Gamazaychikov, Theo Alves da Costa & Emma Strubell, Misinformation by Omission: The Need for More Environmental Transparency in AI (June 18, 2025) (unpublished manuscript), https://arxiv.org/abs/2506.15572.
  26. Footnote 25.Henrik Skaug Sætra, A Framework for Evaluating and Disclosing the ESG Related Impacts of AI with the SDGs, 13 Sustainability 8503 (2021), https://doi.org/10.3390/su13158503
  27. Footnote 25.Henrik Skaug Sætra, The AI ESG Protocol: Evaluating and Disclosing the Environment, Social, and Governance Implications of Artificial Intelligence Capabilities, Assets, and Activities, 31 Sustainable Dev. 1027 (2023), https://doi.org/10.1002/sd.2438.
  28. Footnote 26.Eliza Martin & Ari Peskoe, Extracting Profits from the Public: How Utility Ratepayers Are Paying for Big Tech’s Power (Harv. Elec. L. Initiative, Mar. 2025), https://eelp.law.harvard.edu/wp-content/uploads/2025/03/Harvard-ELI-Extracting-Profits-from-the-Public.pdf (last visited July 15, 2026).
  29. Footnote 28.Irene Vogelaar, Sustainable AI: The Energy-Accuracy Trade-Off in the Artificial Intelligence Act (Mar. 20, 2025), in Environments (Joshua C. Gellers & Henrik Skaug Sætra eds.), in Oxford Intersections: AI in Society (Philipp Hacker ed., 2025), https://doi.org/10.1093/9780198945215.003.0030.
  30. Footnote 28.Joshua C. Gellers, Emerging Technologies and Earth System Governance in the Anthropocene: Editorial, 14 Earth Sys. Governance 100157 (2022), https://doi.org/10.1016/j.esg.2022.100157 (editorial introducing a special issue on artificial intelligence and digitalization).
  31. Footnote 29.Henrik Skaug Sætra, AI for the Sustainable Development Goals 3 (2022).
  32. Footnote 32.Ill. Commerce Comm’n v. FERC, 576 F.3d 470, 476 (7th Cir. 2009) (Posner, J.).
  33. Footnote 32.Ill. Commerce Comm’n v. FERC, 756 F.3d 556 (7th Cir. 2014) (quoting the passage in full and citing it to “576 F.3d at 476”).
  34. Footnote 36.42 U.S.C. § 6291(1).
  35. Footnote 49.42 U.S.C. § 6295(o)(1)
  36. Footnote 52.Energy Conservation Program for Consumer Products and Certain Commercial and Industrial Equipment: Proposed Determination of Computer Servers as a Covered Consumer Product, 78 Fed. Reg. 41,868 (proposed July 12, 2013) (Docket No. EERE-2013-BT-DET-0034, RIN 1904-AD03).
  37. Footnote 53.Energy Conservation Program: Proposed Determination of Computer and Battery Backup Systems as a Covered Consumer Product, 79 Fed. Reg. 11,345, 11,347 (proposed Feb. 28, 2014) (Docket No. EERE-2013-BT-DET-0035, RIN 1904-AD04) [hereinafter Computer Systems Proposal].
  38. Footnote 57.Energy Conservation Program: Proposed Determination of Computer Servers as a Covered Consumer Product, 79 Fed. Reg. 11,350, 11,350 (proposed Feb. 28, 2014) (“Department of Energy (DOE) withdraws for further consideration a proposed determination that computer servers (servers) qualify as a covered product under Part A of Title III of the Energy Policy and Conservation Act (EPCA), as amended”)
  39. Footnote 59.Energy Conservation Program: Energy Conservation Standards for Uninterruptible Power Supplies, 85 Fed. Reg. 1447, 1454 (Jan. 10, 2020) (citing Computer Systems Proposal, 79 Fed. Reg. 11,345, and 79 Fed. Reg. 41,656 (July 17, 2014)).
  40. Footnote 60.42 U.S.C. § 6292(a).
  41. Footnote 62.10 C.F.R. § 431.97(f)(2) (2026) (“Each computer room air conditioner manufactured on or after May 28, 2024, must meet the applicable minimum energy efficiency standard level(s) set forth in this paragraph (f)(2).”)
  42. Footnote 62.Energy Conservation Program: Energy Conservation Standards for Computer Room Air Conditioners, 88 Fed. Reg. 36,392, 36,392 (June 2, 2023) (adopting amended standards effective August 1, 2023 with compliance required May 28, 2024) [hereinafter CRAC Standards Rule].
  43. Footnote 63.Energy Conservation Standards for Computer Room Air Conditioners and Dedicated Outdoor Air Systems, 84 Fed. Reg. 48,006, 48,009 (proposed Sept. 11, 2019).
  44. Footnote 65.42 U.S.C. § 6311(8)(A)-(D)
  45. Footnote 65.10 C.F.R. § 431.92 (2026) (parallel definitions).
  46. Footnote 69.42 U.S.C. § 6311(2)(A)
  47. Footnote 71.42 U.S.C. § 6312(a)-(b).
  48. Footnote 75.42 U.S.C. § 6313(a)(6)(B)(iii)(II)(aa).
  49. Footnote 78.Energy Conservation Program: Review of DOE’s Analytic Methods for Setting Energy Conservation Standards, 91 Fed. Reg. 41,578 (July 7, 2026).
  50. Footnote 79.Exec. Order No. 14,154, 90 Fed. Reg. 8353 (Jan. 29, 2025).
  51. Footnote 81.90 Fed. Reg. 43,371 (Sept. 9, 2025)
  52. Footnote 81.Pub. L. No. 119-8
  53. Footnote 81.Electronic Code of Federal Regulations, 10 C.F.R. pt. 431 editorial note, https://www.ecfr.gov/current/title-10/chapter-II/subchapter-D/part-431 (last visited July 15, 2026).
  54. Footnote 82.Stuart D. Kaplow, Rollback of Federal Appliance Efficiency Standards Happening: What Businesses Need to Know, https://stuartkaplow.com/legal-library/environmental-law/rollback-of-federal-appliance-efficiency-standards-happening-what-businesses-need-to-know/ (last visited July 15, 2026) (quoting Secretary Chris Wright: “In America, you should be able to choose between a drying machine that takes multiple cycles to dry your clothes and one that does it on the first try”).
  55. Footnote 83.David Patterson et al., The Carbon Footprint of Machine Learning Training Will Plateau, Then Shrink, 55 Computer 18 (2022), https://doi.org/10.1109/MC.2022.3148714.
  56. Footnote 83.Quotations from this work follow the authors’ openly available manuscript, https://doi.org/10.36227/techrxiv.19139645.v3 (Mar. 3, 2022).
  57. Footnote 84.Georgia Butler, Google: Median Gemini Prompt Uses 0.24 Watt Hours of Power and Consumes 0.26ml of Water, Data Ctr. Dynamics (Aug. 22, 2025), https://www.datacenterdynamics.com/en/news/google-median-gemini-prompt-uses-024-watt-hours-of-power-and-consumes-026ml-of-water/ (last visited July 15, 2026).
  58. Footnote 91.Google Reveals How Much Energy a Single AI Prompt Uses, EnergySage (Aug. 25, 2025), https://www.energysage.com/news/google-ai-energy-use-electric-bill-impact/ (last visited July 15, 2026)
  59. Footnote 91.AI Efficiency: A Gemini Reality Check, Greenly (Sept. 29, 2025), https://greenly.earth/en-us/leaf-media/data-stories/ai-efficiency-a-gemini-reality-check-with-greenly (last visited July 15, 2026).
  60. Footnote 92.OpenAI’s Lack of Transparency on GPT-5 Energy Needs Sparks Concern, Computing (2025), https://www.computing.co.uk/news/2025/ai/openai-s-gpt-5-sparks-energy-use-debate (last visited July 15, 2026) (“OpenAI has not released energy data for any of its models since GPT-3 in 2020, which had 175 billion parameters.”).
  61. Footnote 93.What’s the Environmental Cost of an AI Text Prompt? Google Says It Has an Answer, CBS News (Aug. 21, 2025), https://www.cbsnews.com/news/ai-environment-impact-study-energy-usage-google-gemini-prompt/ (last visited July 15, 2026).
  62. Footnote 95.OpenAI Flagship: Performance, Cost, Electricity, DigiTimes (Aug. 15, 2025), https://www.digitimes.com/news/a20250815PD238/openai-flagship-performance-cost-electricity.html (last visited July 15, 2026).
  63. Footnote 100.Jonathan Koomey & Eric Masanet, Does Not Compute: Avoiding Pitfalls Assessing the Internet’s Energy and Carbon Impacts, 5 Joule 1625, 1625 (2021), https://doi.org/10.1016/j.joule.2021.05.007.
  64. Footnote 101.Carlota Perez, Technological Revolutions and Financial Capital: The Dynamics of Bubbles and Golden Ages (2002).
  65. Footnote 102.Jim Hayes, The Perils of Irrational Exuberance: The 25th Anniversary of the Dot-Com Boom, ISE Mag. (July 24, 2025), https://www.isemag.com/professional-development-leadership/article/55293922/the-perils-of-irrational-exuberancethe-25th-anniversary-of-the-dot-com-boom (last visited July 15, 2026).
  66. Footnote 102.Logan Kugler, Dark Fiber Is Lighting Up, Comm. ACM (2014), https://cacm.acm.org/news/dark-fiber-is-lighting-p/ (last visited July 15, 2026).
  67. Footnote 105.Old Dominion Elec. Coop. v. FERC, 898 F.3d 1254, 1256 (D.C. Cir. 2018)
  68. Footnote 105.S.C. Pub. Serv. Auth. v. FERC, 762 F.3d 41, 53 (D.C. Cir. 2014) (per curiam)
  69. Footnote 105.Ill. Commerce Comm’n v. FERC, 721 F.3d 764 (7th Cir. 2013).
  70. Footnote 108.Long Island Power Auth. v. FERC, 27 F.4th 705 (D.C. Cir. 2022).
  71. Footnote 110.H.B. 30, 2026 Gen. Assemb., Spec. Sess. I, item 3-5.24 (Va. 2026) (enacted) [hereinafter Item 3-5.24], https://budget.lis.virginia.gov/bill/2026/2/HB30/Chapter/.
  72. Footnote 110.Bradley J. Nowak et al., Virginia Budget Creates New Electricity Consumption Tax for Data Centers, Williams Mullen (July 2026), https://www.williamsmullen.com/insights/news/legal-news/virginia-budget-creates-new-electricity-consumption-tax-data-centers.
  73. Footnote 112.Va. Code Ann. § 58.1-647.
  74. Footnote 113.42 U.S.C. § 6292(b)(1)(B), (b)(2)(A)
  75. Footnote 115.Taylor Giorno, Zara Norman & Samuel Larreal, ‘The Battle Line Has Been Drawn’ Around Virginia’s Data Centers, NOTUS (July 1, 2026), https://www.notus.org/metro/northern-virginia-data-centers-politics (quoting Del. John McAuliff) [hereinafter Giorno et al.].
  76. Footnote 116.2026 Special Session I, Budget Amendments, H.B. 30 (Governor’s Recommendations) amend. 11 (June 26, 2026), https://budget.lis.virginia.gov/amendment/2026/2/HB30/Enrolled/GR.
  77. Footnote 117.Virginia’s New Data Center Electricity Rate Class, Am. Action F. (Apr. 22, 2026), https://www.americanactionforum.org/insight/virginias-new-data-center-electricity-rate-class/.
  78. Footnote 118.S.B. 253, 2026 Gen. Assemb., Reg. Sess. (Va. 2026).
  79. Footnote 118.Virginia Leads the Way on Energy Affordability in 2026, Nat’l Caucus of Env’t Legislators (Apr. 2, 2026), https://www.ncelenviro.org/articles/virginia-leads-the-way-on-energy-affordability-in-2026/.
  80. Footnote 128.Eric Masanet, Nuoa Lei & Jonathan Koomey, To Better Understand AI’s Growing Energy Use, Analysts Need a Data Revolution, 8 Joule 2427 (2024), https://doi.org/10.1016/j.joule.2024.07.018.
  81. Footnote 130.79 Fed. Reg. 11,350, 11,350 (proposed Feb. 28, 2014)

All research