AI Policy Wiki
Dashboard
⚲
AI Policy Wiki › Industries

Energy and Electric Power Sector

medium confidence · updated 2026-09-27

The electric-power industry as the supply constraint on AI data-center growth — electricity demand, grid interconnection, cost-allocation disputes, utility consolidation, and generation procurement (gas, nuclear, geothermal, renewables) tied to AI compute buildout.

Contents

The electric-power sector occupies an unusual position in AI policy: it is less a deployer of AI than the physical input on which AI compute depends. The growth of frontier-model training and inference has turned electricity supply, grid capacity, and generation procurement into binding constraints on data-center expansion, and has made utilities, grid operators, and energy regulators central actors in decisions about where and how fast AI infrastructure is built. The sector's relationship to AI is tracked here as the supply side of AI data centers and data-center siting.

Scale of demand

Estimates of data-center electricity consumption rose sharply through 2025–2026. The International Energy Agency projected that data centers would consume on the order of 1,000 TWh in 2026, comparable to the annual electricity use of a large industrialized country (Source: tech-insider.org). One 2026 U.S. tally counted over 4,500 active data-center facilities consuming roughly 176 TWh annually — about 4.4% of U.S. electricity — with more than 700 additional facilities under construction (Source: electricchoice.com). Analysts have described the resulting load growth as a reversal of roughly two decades of flat U.S. electricity demand. A Belfer Center analysis characterized the rise in consumption growth rates, driven in part by hyperscale data centers, as a turning point for grid planning (Source: belfercenter.org).

Two 2026 projections set the upper end of the U.S. range. The IEA projects that by 2030 U.S. data centers will consume more electricity than all heavy industry combined, spanning cement, steel, chemicals, autos and other industrial sectors (Inside the Dirty, Dystopian World of AI Data Centers — Matteo Wong (The Atlantic, April 2026)). The Department of Energy's Lawrence Berkeley National Laboratory, in analysis reported June 24, 2026, projected that data-center electricity demand could exceed 15 percent of total U.S. consumption by the end of the decade, a figure the department paired with a call for efficiency gains (Source: insideaipolicy.com).

Hyperscaler-level figures published July 16, 2026 place the growth in perspective: Google's electricity consumption rose more than 140% between 2021 and 2025, and Google, Microsoft, Amazon and Meta added combined new power demand between 2022 and 2025 roughly twice New York City's annual consumption (Source: axios.com).

Demand is regionally concentrated rather than evenly spread. Loudoun County, Virginia hosts 199 operating data centers with roughly 30 more in the pipeline, representing 13% of global data-center capacity within 520 square miles; Dominion Energy forecasts 5.5% annual electricity-demand growth in Virginia, doubling by 2039, which it has called the largest growth in power demand since the years following the Second World War (Inside the Dirty, Dystopian World of AI Data Centers — Matteo Wong (The Atlantic, April 2026)).

Grid constraints and cost allocation

The central policy disputes concern interconnection and who pays. Large data-center loads can take years to connect to the grid, and their concentration in particular regions strains local transmission and generation. A recurring question is whether AI data centers should bear the cost of the grid upgrades their demand requires, or whether those costs are spread to other ratepayers. Texas, which courted data centers with cheap power, abundant land, and tax incentives, became a test case for whether such facilities should pay for the grid capacity they consume (Source: cryptorank.io). Cost-allocation proceedings before state public utility commissions and the Federal Energy Regulatory Commission, the federal regulator of interstate transmission and wholesale power, are where these questions are increasingly litigated. Financing for the buildout has also reached into new channels: Morgan Stanley began pitching data-center developers on raising money in the leveraged-loan market as construction costs pressed them toward new capital sources (Source: theinformation.com).

Capacity prices supply the clearest market signal. PJM's 2025/2026 capacity auction cleared at roughly $269.92/MW-day, about ten times the prior year, an increase widely attributed to AI and data-center load colliding with coal retirements, and one that mobilized ratepayer advocates across Ohio, Maryland and Illinois. In ERCOT, an isolated grid combined with light permitting and the SB 6 (2025) load-interconnection reforms make Texas the highest-growth AI power market, with peak-demand forecasts having doubled since 2023.

How much of the requested load is real is itself contested. Documents from ERCOT and Texas Governor Greg Abbott, described in a review published September 1, 2026, put requests from data centers and other large energy users to connect to the Texas grid at more than 474 gigawatts, up from about 48 gigawatts in 2023. Across portions of the Midwest, Mid-Atlantic and South, electricity requests from very large power users, mostly data centers, exceed 700 gigawatts — more than ten times industry estimates of current U.S. data-center power use — and the account states that much of that demand may be an illusion, with developers submitting the same project to several utilities or filing speculative requests. The cornerstone of Abbott's responding order is an audit requiring proposals to power up data centers to disclose who ultimately owns them, replacing earlier rules under which only an affiliate company had to be named (Source: reuters.com).

The distributional question reached the federal regulator directly on June 24, 2026, when the Federal Energy Regulatory Commission pressed grid operators to justify or revise their market rules for connecting large power users such as data centers, with the stated goal of preventing the costs of serving that load from being shifted onto other customers (Source: insideaipolicy.com).

Congress took up the same question a month later. On July 21, 2026 the House Energy and Commerce Committee approved a data-center energy package including the Ratepayer Protection Act (H.R. 9340, passed 52–0), which would have states consider making data centers with 100 MW of demand pay full grid-upgrade costs (Source: broadbandbreakfast.com; thehill.com; energycommerce.house.gov). The bill adds a standard to section 111(d) of the Public Utility Regulatory Policies Act of 1978 under which a utility's rates for a large-load customer are designed to recover the full incremental cost of generation, transmission or distribution upgrades, including if the customer terminates its contract, and the utility must obtain financial assurances before building. The covered class is defined as non-residential consumers whose facilities take electricity primarily for information-technology infrastructure with 100 MW or more of aggregate peak demand at a single site or campus; the text does not use the words "artificial intelligence" or "data center." State regulatory authorities and nonregulated utilities must consider the standard within one year and make a determination within two, may decline to adopt it, and are exempt where the state has already implemented, held a proceeding on, or legislatively voted on a comparable standard (Ratepayer Protection Act, H.R. 9340 — text as reported in the House (September 10, 2026)). The House passed the bill 417–3 on September 16, 2026 (Source: clerk.house.gov). A Senate companion, S. 5028, introduced by Sen. Jon Husted on July 16, 2026, is pending in the Energy and Natural Resources Committee (Source: congress.gov). President Trump separately expanded a voluntary ratepayer-protection pledge to 23 governors and 187 companies (Source: apnews.com). Sen. Mark Warner introduced legislation, made public July 23, 2026, conditioning a tax exemption for AI data-center developers on efficiency standards plus energy- and water-use reporting (Source: insideaipolicy.com).

California legislated tariff requirements directly. Of the seven data-center bills Governor Gavin Newsom signed on September 21, 2026, SB 886 and AB 2383 together create Public Utilities Code Article 14.7, the California Technology Innovation and Ratepayer Protection Act. It directs the Public Utilities Commission to establish data-center tariffs by January 1, 2028 that prevent stranded costs for, or cost shifts to, other customers; assigns all transmission upgrade costs of transmission-level data centers to the data center, caps refunds of its contributions at 75 percent of the annual net transmission revenue it produces, and imposes early termination fees for departure within 10 years; requires data centers to pay a reasonable share of wildfire, electrification and other societal costs; and requires generation tariffs of at least 10 years, with collateral, minimum payments based on projected load, and a demand threshold no higher than 25 MW, from investor-owned utilities and, by the same date, from community choice aggregators and electric service providers. Applicants must disclose every other utility territory where they have filed for the same data center. A companion bill, SB 1168, directs the Commission to assess data-center rate structures that relieve cost pressure on residential and low-income customers (California data-center laws (2026); California data-center bill package — chaptered texts (Chapters 434–440, Statutes of 2026)).

Public opinion on the cost question was measured in an AP-NORC Center and Energy Policy Institute at the University of Chicago survey of 3,424 adults conducted July 6–24, 2026 and released September 16. Sixty-three percent said they were extremely or very concerned about data centers' effect on electricity prices in the communities where they are built, and 53% about power outages; 73% supported requiring companies that build data centers to pay for the electric grid upgrades needed to support them, and 41% supported restricting construction of new natural gas plants to supply data centers. Support for the grid-upgrade requirement was 79% among Democrats and 76% among Republicans (AP-NORC/EPIC 2026 Energy Survey: AI, Data Centers, Climate and Energy).

Demand flexibility

One industry response to interconnection delay treats data-center load as adjustable. On September 16, 2026 Emerald AI, Google and Nvidia launched the AI Energy Management Alliance, whose 18 launch partners include the utilities National Grid and AES and the power producers Constellation, NRG and RWE, to press regulators, grid operators and utilities to connect data centers faster when they can reduce their draw during peak grid stress through colocated generation, storage or computational flexibility (Source: businesswire.com). The coalition is a relaunch of the Advanced Energy Management Alliance, founded in 2014 to advocate for demand-response companies. Google said it has committed 1 gigawatt of reducible power demand through utility agreements nationwide, and Axios reported that federal regulators had in June directed regional grid operators to examine new options for connecting large, flexible power users (Source: axios.com). Emerald AI's chief executive Varun Sivaram has written that Silicon Valley Power, a California municipal utility, launched the first flexible-load interconnection program in the country and that ERCOT was finalizing rules letting controllable data centers connect sooner (Source: fortune.com). Nvidia and Emerald AI had earlier announced, on March 23, 2026, a collaboration with AES, Constellation, Invenergy, NextEra Energy, Nscale Energy & Power and Vistra on data centers designed to operate as flexible grid assets (Source: nvidianews.nvidia.com).

Generation procurement

The demand surge has reshaped procurement decisions by AI developers and the energy companies that serve them. Independent power producers and utilities including Constellation Energy, Dominion Energy, and NextEra have featured in data-center power agreements, and AI firms have contracted for dedicated generation including natural gas, restarted or uprated nuclear capacity, and renewable portfolios. The intersection of energy policy and AI regulation — permitting, emissions, reliability, and siting — was the subject of background analysis prepared for the Forum for Cooperation on AI (Source: brookings.edu). Generation choices carry emissions and reliability implications that connect the AI buildout to broader energy and climate policy.

Nuclear

Direct power-purchase agreements between hyperscalers and nuclear operators have been signed at a pace described as unseen since the 1970s, treated in detail at Nuclear PPAs for AI. The reference transactions are Constellation's Three Mile Island Unit 1 restart for Microsoft (announced September 2024, roughly 835 MW); Talen Energy's Susquehanna co-location for Amazon, whose initial behind-the-meter structure FERC blocked in late 2024 before it was restructured; Vistra's Comanche Peak PPA expansion; and small-modular-reactor term sheets between Oklo, X-energy and Kairos and Google, Amazon and Microsoft. Whether premium-priced 24/7 firm nuclear power crowds out residential and industrial allocation, or instead finances new build that expands supply, is unresolved.

The licensing pathway drew a legal challenge in mid-2026: Democratic attorneys general from nine states urged the Nuclear Regulatory Commission, in June 15, 2026 written comments, to withdraw its proposed streamlined licensing of microreactors intended to power data-center growth, arguing the plan violates federal statutes including the National Environmental Policy Act (Source: insideaipolicy.com).

Renewables, gas and frontier generation

During the week of July 13, 2026 Google announced its largest solar-and-battery purchase to date, the Steel River Energy Center in Arkansas, adding 1 gigawatt of solar and 1.9 gigawatt-hours of battery storage in its first two phases, backed by $3.5 billion in secured financing and scaling to about 1.8 gigawatts by 2029, which would make it the largest solar facility in the United States. The site sits 40 miles north of xAI's unpermitted gas-turbine power plant serving Colossus 2 (Source: techcrunch.com).

Accounts of the power supply behind Anthropic's first Australian lease differ. For that lease, at Zerra DC's proposed Western Downs Digital Park in Queensland, two people familiar with the deal told Reuters that Zerra would source renewable power purchase agreements and bear grid connection costs without cost to existing customers (Source: reuters.com), while the Australian Financial Review reported that the development application allows the first stage to be powered by gas or electricity (Source: afr.com) and the site's landowner told Capital Brief he expected it to draw on existing coal plants (Source: capitalbrief.com).

On-site gas generation has drawn a separate regulatory line. An Environmental Protection Agency interpretation that became public by July 24, 2026 exempts "islanded" data-center power generation — on-site generation not connected to the grid — from Clean Air Act acid-rain provisions (Source: insideaipolicy.com).

Data-center demand has been large enough to underwrite capital formation in generation technologies ahead of commercial deployment. Fervo Energy ran a roughly $2 billion IPO at a valuation above $10 billion to supply geothermal power for AI data centers, the first major IPO of an AI-data-center power supplier separate from a utility (Source: theinformation.com). Fusion startup Helion Energy said on June 4, 2026 that it had raised $465 million in a Thrive Capital-led round, nearly tripling its valuation to $15.5 billion on demand for fusion power for data centers (Source: theinformation.com).

Utility consolidation

Ownership of regulated generation and transmission began to consolidate on AI-driven demand. The Financial Times reported on May 15, 2026 that NextEra and Dominion Energy had entered talks to combine, explicitly driven by electricity demand for AI data centers (Source: ft.com). The companies announced the deal on May 18, 2026 as an all-stock merger valued at approximately $67 billion that would create the world's largest regulated electric utility by market capitalization, serving roughly 10 million customer accounts across Florida, Virginia and the Carolinas (Source: pbs.org). Dominion shareholders receive 0.8138 NextEra shares per Dominion share plus a $360 million one-time cash payment at closing; NextEra holders own 74.5% of the combined entity and Dominion shareholders 25.5%; NextEra chief executive John Ketchum serves as chairman and chief executive; and closing is expected in 12 to 18 months, pending Nuclear Regulatory Commission and other approvals. Both companies framed the combination as a response to AI-driven electricity demand.

Reliability and security

Concentrated, digitally controlled data-center load has entered the grid-security discussion as a reliability question distinct from the ratepayer one. In September 2, 2026 testimony to the House Permanent Select Committee on Intelligence, Hudson Institute senior fellow David Feith cited a routine transmission fault in Northern Virginia on July 22, 2026 that moved roughly 3,800 MW of data-center load from the grid to backup generation within seconds, producing voltage and frequency swings from Virginia to Chicago without a blackout and without any cyber component, and argued that the event demonstrates the physics an attacker with synchronized access to inverters, batteries or uninterruptible power systems would exploit. He argued that large data centers should be treated as critical grid infrastructure (A New Counterintelligence Map for the AI Age (Feith, HPSCI, 2026-09-02)).

Open questions

  • Whether the interconnection-queue figures reported for ERCOT and the Midwest and Mid-Atlantic reflect real load or duplicate and speculative filings is unresolved; the Texas ownership-disclosure audit is the first mechanism directed at measuring it.
  • Whether hyperscaler nuclear procurement crowds out other allocation or finances net-new capacity is not settled by the available record.

Relationships

Categories: Industries | energy | deployer | compute | us | global
What links here (34)