Report
AI InfrastructureNov 9, 2026

The Ballot Box Meets AI Infrastructure

The midterms turned opposition to data centers into votes. The risk sits less in 2027 capacity already being built than in 2028 capacity that has not broken ground.

By LightViewDownload PDF한국어 PDF

Key takeaways

  1. Forty of 41 local ballot measures went against data centers. They were decided on November 3 in ten states including Ohio, and at least 32 of the 34 governors elected the same day back tighter rules, conditional approval or a pause.
  2. The backlash starts with the power bill. PJM’s capacity price (per MW-day) jumped about 9.3x in one year, from $28.92 to $269.92, and has stayed at the cap for three years; Maryland household rates are 49% above 2019.
  3. Most data center capacity due in 2027 is likely to hold. About three-quarters is already under construction. The risk of delay is concentrated in 2028 capacity that has not started, which SemiAnalysis sizes at 5.7 to 30 GW.
  4. Texas is the biggest swing factor. It accounts for about 30% of 2028 US additions, grid connections and permits were frozen before the vote, and a new governor and legislature rewrite the rules in January.
  5. The likely outcome looks closer to “build if you pay” than a ban. Governors of both parties and even OpenAI agree that data centers should pay for the grid, so buildout is more likely to slow and get costlier than to stop. Companies already holding land and power gain the edge.

1Forty of 41 measures went against data centers

Our previous column, on the blue wave, looked at the AI selloff two days after the election and compared valuations with 2000. This one examines the data center issue cited as one trigger of that selloff, because after chips and power, the new bottleneck for AI infrastructure has turned out to be votes.

On November 3, voters in ten states decided 41 local ballot measures on data centers. Ballotpedia counts 27 approved and 14 defeated: all 27 approved measures limited or opposed data centers, and 13 of the 14 defeated ones would have favored them. In all, 40 of the 41 outcomes restricted development. The lone exception was Harwood, North Dakota, a small city that rejected extra review of industrial projects.

Figure 1

The margins were wide. In Ashville, Ohio, 82% voted against a deal for a data center already under construction, and 88% in Morgan County, Illinois, backed restrictions in an advisory vote. Governors’ races moved the same way: in a preliminary count, Data Center Watch finds 12 of 34 winners backing a ban or moratorium and 20 backing regulation or conditional development. With winners in Texas, Ohio and Georgia all promising a pause, this suggests the issue has moved beyond local complaints into state policy.

The measures also went beyond protest votes. Of the 27 approved, five banned data centers outright, 12 imposed partial bans based on size, such as over 25 MW, and three in Maine set six-month moratoriums. Others required a public vote for large projects or financial guarantees for tax breaks. More votes are scheduled in Arkansas and North Dakota in December, which will show whether this is a one-off or a trend.

2The power bill moved first

The main reason voters turned is the electricity bill. In PJM, which runs the grid for 13 states and Washington, D.C., and where data center demand has concentrated, prices in the capacity market, where utilities buy future generating capacity in advance, have soared.

The clearing price rose from $28.92 per MW-day for 2024/25 to $269.92 the next year, about 9.3x, and from 2026/27 it has hit the regulator-approved cap three years in a row. PJM said nearly 5,100 MW of the roughly 5,250 MW increase in forecast peak load for 2027/28 came from data centers. The cost lands on household bills: the 2025/26 increase added about $18 a month in western Maryland and $16 in Ohio, and PJM’s market monitor attributed 63% of that increase to data centers.

Figure 2

High capacity prices are also a sign of shortage. The 2028/29 auction in July fell 6,831 MW short of the reliability target, meaning new generation is not keeping pace with data center growth. In its January report the market monitor estimated data centers add $7.3 billion in capacity charges for that delivery year and $6.5 billion for the next (as cited by FactCheck.org). The sense that data centers use the power while the whole region shares the cost came back as votes against them.

How capacity becomes a bill. PJM runs auctions to secure enough generating capacity three years ahead to meet demand on the hottest day. The clearing price times the capacity procured is the total cost ($16.4 billion for 2027/28), which local utilities spread across household and business bills. The more data centers lift forecast demand, the more capacity is needed and the higher the price.

Actual bills have risen faster in PJM states too. EIA data show residential rates rose 49% in Maryland, 42% in New Jersey and 40% in Pennsylvania from 2019 to 2025, well above the US average of 32%.

Figure 3

Rates kept rising this year: Maryland averaged more than 21 cents per kWh from January to May. Virginia, home to the most data centers, rose only 27%, which suggests data center counts alone do not set the pace. Studies disagree as well. MIT’s energy policy center estimates data center entry raised retail rates 2.7% over 2010–2024, while an EPRI preprint, from a group largely funded by utilities, finds average US rates fell slightly over 2015–2024, and about 6% for customers in areas with average data center growth.

Water and taxes added fuel. In September Texas ordered large water users to report their use, with penalties, and in several states sales-tax breaks for data centers drew criticism as too generous for the jobs they bring. Texas governor-elect Gina Hinojosa promised to end the breaks and tax the electricity data centers use; local reporting puts the exemption at more than $1 billion a year. With governors of both parties offering similar responses, these grievances appear to reach beyond any one camp.

32027 is being built; 2028 has not broken ground

Which part of the buildout does the vote touch? It depends on the stage. Projects under construction with grid agreements will mostly proceed, while unstarted ones must meet whatever conditions new governors and legislatures set.

SemiAnalysis estimates about three-quarters of the 33.8 GW due in 2027 in the top 20 states is under construction. Most of the 62.2 GW of 2028 US net additions has not started, and 43.5 GW (70%) is in 13 states that held November elections. It put 2028 exposure at ~30 GW if pause candidates won everywhere, 5.7 GW if less restrictive ones did.

Figure 4

The result lies between those extremes: pause candidates won in Texas, Ohio and Georgia, but some would allow projects that meet conditions. Exposure means a project could slip, not that it is cancelled; in September SemiAnalysis saw only 1.5 GW of about 20 GW under local rules at risk of delay. Northern Virginia, a leading hub, held no state election this year and is outside this count, though that does not mean it is free of regulatory risk. How far schedules move matters more than the exposed total.

What changes at each stage (LightView summary from SemiAnalysis estimates)
CapacitySizeEffect of new rules
2027, under construction~25 GW (top 20 states)Mostly proceeds; grid-cost conditions may be added
2027, not started~8 GWSome delay possible
2028, 13 states43.5 GW (Texas ~19 GW)Must meet conditions set by new governors and legislatures
2028, other states18.7 GWExisting rules; capacity may shift here

4Texas is the swing factor

Texas is the largest single variable. Wells Fargo estimates 27% of planned US data center capacity is in Texas, and SemiAnalysis puts the state at about 30% of 2028 US additions. Permitting there was frozen before the vote.

Figure 5
Figure 6

On August 3, Governor Greg Abbott (R) ordered ERCOT to audit every data center in the connection queue and pause their progress until the audit is done. In September he enforced water reporting and told agencies to stop issuing data center permits, saying data centers “must pay their own way.”

His successor, Gina Hinojosa (D), pledged to halt approvals until the Legislature passes protections. Democrats provisionally won the state House while Republicans keep the Senate 19 to 12, so Texas is likely to spend early next year in a “pause, then new rules” phase.

The audit has a lot to sort through. Large loads seeking connection total 474 GW, yet connected large loads peaked at 4.1 GW in September (Figure 6). New rules may slow buildout while separating real projects from requests that only hold a place in line.

5Those with sites, and those who must build

When permits become scarce, those already holding land, power and capital gain the edge. The market’s first reaction after the vote showed exactly that split.

Figure 7

On November 4, Microsoft (+6.7%), Amazon (+6.1%) and Alphabet (+3.2%) rose, while power producer Vistra (−13.8%), cooling and power-equipment maker Vertiv (−12.9%) and GPU-rental neocloud CoreWeave (−9.9%) fell sharply. Secured sites became scarce assets; growth needing new sites and power looked riskier. The next day the worry spread across AI: the Philadelphia Semiconductor Index fell 12.3%, Micron and AMD more than 17%, and the three clouds also declined.

Direction of impact by group (LightView view, not a buy or sell call)
GroupWhy it is affectedDirection
HyperscalersSecured sites become scarce assets; capital to meet cost conditions, though spending may slipRelatively favored
NeocloudsRely more on developing or leasing new sites, so they feel permit delays firstUnfavorable
Power and equipmentNew load may arrive more slowly; a longer power shortage would support pricesMixed
Chips and memoryData center build speed sets chip deploymentPressure from 2028
Data center REITsOperating sites gain value; new builds face the same rulesMixed
Korean grid equipmentSolid backlogs, but US delays could slow new ordersA 2028+ variable

The hyperscalers did not make deep cuts in the results just before the vote. Their 2026 capex plans sum to about $730 billion at the midpoints, and Amazon and Alphabet signaled big increases for 2027.

Figure 8

Whether the money can be spent on time is another matter. Microsoft said some third-party GPU capacity slipped into the second half of its fiscal year; that was a supply-timing issue, not permits, but it shows outside delays move the timing of spending before the total. Korean grid-equipment makers are exposed too: on November 5, as the KOSPI fell 5.3%, LS ELECTRIC dropped 13.9% and HD Hyundai Electric 8.4%.

Figure 9

With one to three years of orders in backlog, though, any US delay is more likely to show first in the pace of new orders from 2028 than in existing ones.

6A ban, or build if you pay?

The likeliest outcome looks closer to “build if you pay” than an outright ban, because the demands overlap strikingly across camps.

Figure 10

Republican Abbott said data centers must pay their own way; Democrat Acton’s campaign set conditions including paying all utility costs and no NDAs, and told developers that do not meet them to stay out of Ohio. In industry, OpenAI responded right after the vote with a community compact to pay full power and grid costs and publish water and power use by site. That points to buildout that slows and gets costlier rather than stops, and, as Morgan Stanley warned, a country split into areas open and closed to data centers.

What to watch next
WhenWhatWatch for
Dec 1 · Dec 15Bond votes in Clarksville, Ark.; Mandan, N.D., ban on 25 MW+ projectsWhether the restrictive trend continues
Jan 12 · Jan 19, 2027Texas Legislature opens; Hinojosa takes officeScope and length of the pause, tax breaks, cost rules
Jan–Feb 2027Hyperscaler Q4 results2027 capex figures, comments on delays
Jul 2027One year of New York’s executive orderExtension or conditional lifting (no end date set)

There is a counterview: with so much already under construction, the post-election drop may have overshot, and regulation could favor large players and their pricing power. In the end, how much later and costlier 2028 capacity becomes is likely to decide the next one to two years of AI infrastructure investment.

+Terms and sources

Terms

PJM — Operator of the grid serving 13 eastern and midwestern states and Washington, D.C.
Capacity market — An auction that secures generating capacity years ahead; the cost flows into bills.
ERCOT — Operator of most of the Texas grid.
Moratorium — A halt on permits or approvals for a set time or until conditions are met.
Neocloud — A newer cloud company focused on renting GPU servers (e.g., CoreWeave).
Hyperscaler — A cloud giant that runs its own very large data centers.
GW (gigawatt) — A unit of power; 1 GW is roughly one nuclear reactor’s output.
NDA — Non-disclosure agreement, criticized for hiding project details from residents.
    Elections and ballot measures
  1. Ballotpedia News (Nov 3 and 6, 2026) · State Capital Lobbyist (Nov 5, unofficial results) · Fast Company (Nov 4) · Renewable Energy World (Nov 6) · AP (Nov 6, via OPB and ABC) · Data Center Watch (Nov 5) · CBS Texas (Nov 4) · Fox 4 (Nov 4)
  2. State actions and pledges
  3. Abbott letter (Aug 3, via Akin Gump) · DLA Piper (Sep 25) · Troutman (Sep 22) · TPR and Fox 7 (Sep 21) · Reuters (Nov 4, via GV Wire) · KXAN (Sep 20) · KSAT (Nov 5) · WYSU (Jul 27) · Atlanta Journal-Constitution (Sep) · Jones Day (July, NY Executive Order 62) · NYT (Nov 5, via GV Wire, OpenAI)
  4. Power prices and grids
  5. PJM 2027/28 BRA report and releases (Dec 17 2025, Jul 14 2026) · IEEFA (Jul 30 2025) · FactCheck.org (Oct 29 2026, citing MIT CEEPR, EPRI, LBNL and Brattle) · EIA Electric Power Annual 2025 (Oct 21 2026) and SEDS · Maryland PSC (Jul 30 2026) · ERCOT (Jul 29 Senate testimony, Sep 11 board, Oct 16 operational overview)
  6. Buildout and markets
  7. SemiAnalysis, “The Datacenter Backlash Goes Bipartisan” (Nov 1, free section) · Wells Fargo estimate (cited by Fortune, Nov 5) · company Q3 results (late Oct) and Luminix · AP, CappNotes, TradingKey, Fox Business (Nov 4–5 closes) · Etoday (Nov 5, KOSPI) · News Tomato (Oct 27), Sisa Economy (Oct 30), Mirae Asset (Oct 22) · Herald Business (Apr 27, customs data) · Financial News (Oct 12)
An analytical column based on official statistics and public reporting. It is not an assessment of any party or policy, nor a recommendation to buy or sell any security. SemiAnalysis and Wells Fargo figures are company estimates. It reflects information published through November 8, 2026.

This column's topic and analytical direction are set by the editor, who reviews and revises it before publication. AI tools were used for research, drafting and infographics, and every figure was checked against its source.

Download PDFMore columns
RelatedIndustry mapMarket calendarStock reports

LightView Insight is an analytical column based on public statistics and reporting and does not recommend buying or selling any security. Investment decisions remain the sole responsibility of the investor.

© 2026 LightViewlightview.krLightView Insightcontact@lightview.kr