Data Centers

POLISCI 158 · Session 12 · Monday, November 2, 2026 · Simone Paci

Data centers today: three questions

I.

What does AI need from a place?

  • Explain why data centers concentrate costs locally while their benefits are diffuse, and why that produces siting conflict.

II.

When and why do data centers generate political conflict?

  • Read the paper’s event-study figures and explain why announcements drew contention only after ChatGPT’s release.

III.

What does backlash do?

  • Pose the next question and say why it is hard to identify.

Bonus: an AI-native paper

A data center wants to build in a town like yours. What happens?

Predict: does the town end up saying yes or no?

Show of hands

What does AI need from a place?

Land, power and water, somewhere.

How much electricity do data centers use, compared with US manufacturing?

2023: about one-fifth as much as US manufacturing

2028 (projected): approaches three-fifths

LBNL Figure ES-1: total US data center electricity use rising from about 60 TWh in 2014 to 176 TWh in 2023 (4.4% of US total), with a 2028 scenario range of about 325 to 580 TWh (6.7 to 12.0%)

Shehabi et al. (2024), LBNL, 2024 United States Data Center Energy Usage Report, Fig. ES-1 · https://eta.lbl.gov/publications/2024-lbnl-data-center-energy-usage-report · benchmark: EIA, 2022 Manufacturing Energy Consumption Survey, Table 5.3

Where does AI’s computing power go: training or inference?

80–90%

of computing power for AI is used for inference

  • MIT Technology Review estimate

O’Donnell and Crownhart, MIT Technology Review, May 20, 2025 · https://www.technologyreview.com/2025/05/20/1116327/ai-energy-usage-climate-footprint-big-tech/

How big is the boom?

AI-related investment, first three quarters of 2025

0.97 percentage points of real GDP growth · 39% of all growth

2000, dot-com peak: 28%

Excluding data centers: 36% vs 28%

2000 First 9 months of 2025
Information processing equipment 0.58 0.42
Software 0.11 0.35
R&D 0.12 0.13
Data centers — 0.07

Contribution to real GDP growth, percentage points. Rebuilt from the St. Louis Fed table.

Rubinton and Patro, Federal Reserve Bank of St. Louis, January 12, 2026 · https://www.stlouisfed.org/on-the-economy/2026/jan/tracking-ai-contribution-gdp-growth

Why do people hate this new economic boom?

Environment = public good · pollution = externality · environmental politics = collective action problem

Data center power noise water land compute profits GDP growth

Visible local costs → mobilization: Sexton (2020), Journal of Conflict Resolution · NIMBY and YIMBY land-use politics: Brouwer and Trounstine (2024), Annual Review of Political Science

Is local backlash new?

Ontario, 2011: wind turbines

  • Governing party lost about 4 points (proposed turbine) to 10 points (operating turbine) in nearby precincts

Stokes (2016) Figure 2, Trends in the Governing Party's Vote Share: Liberal Party vote share in treated and control precincts, 2003, 2007 and 2011, for proposed (left) and operational (right) wind turbines, with 95% confidence intervals

Stokes (2016), AJPS, Fig. 2 · https://doi.org/10.1111/ajps.12220

Would Americans accept a data center in their own area?

Gallup chart, Seven in 10 Americans oppose local construction of AI data centers: 7% strongly favor, 20% somewhat favor, 23% somewhat oppose, 48% strongly oppose, March 2 to 18, 2026

Gallup, telephone poll of 1,000 US adults, March 2–18, 2026 · https://news.gallup.com/poll/709772/americans-oppose-data-centers-area.aspx

When and why do data centers generate political conflict?

Disclosure: I co-authored the paper in this part, Paci and Russell (2026).

Header of a manuscript title page: Submission to American Political Science Review; title Data Centers and the Emerging Contentious Politics of AI; Anonymised Author(s), Anonymised Institution(s)

Title page, registered report submitted to the American Political Science Review, July 2026

Where did data center announcements, protests and bills surge after 2022?

Announcements: 65 in Q4 2022 → 644 in Q1 2026

Paci and Russell Figure 1, four panels: county maps of data center announcements, community mobilization and state data-center bill sponsorships, and national quarterly dynamics 2018 to 2026 rising sharply after 2023

What would explain the backlash?

H1 · Resource competition

Stronger where resources are already strained

H2 · Environmental concern

Stronger where pro-environment attitudes and mobilization are higher

H3 · AI anxiety

Stronger where AI-related anxiety is higher

Timing test: after ChatGPT (November 2022), an announcement means something new

Paci and Russell (2026)

What would 1,000 Americans do about a hypothetical data center in their community?

Verasight survey, 1,000 US adults, fielded May 13–18, 2026

Four possible actions:

sign a petition

contact an official

speak at a public meeting

attend a protest

Models use 844 complete cases

Utility prices, environment, AI: which concerns predict stated opposition?

Paci and Russell Figure 3: coefficients for concerns about utility prices, environmental harms, AI job displacement, AI changing society and AI as elite project, extensive and intensive margins; utility prices largest, elite project overlaps zero

How do we know an announcement caused anything?

  1. A county’s first announcement starts its clock
  2. Compare it with counties not (yet) announced, over the same quarters
  3. Split by timing: first announced through November 2022 vs. from December 2022
County A County B County C Comparison Nov 2022 quarters

Where are data centers announced after 2022?

After 2022: larger labor markets and existing generation capacity

Paci and Russell Figure 2: predictors of data center announcements before and after ChatGPT, change in announcement probability per standard deviation for power, economy, politics, demography, geography and prior siting variables

Before and after ChatGPT: what happens after an announcement?

Sketch the line, one panel each:

  • First announced through November 2022
  • First announced from December 2022
  1. Alone · 1 min: sketch
  2. Pairs · 2 min: compare, agree on one picture

Written answers welcome

through Nov 2022 from Dec 2022 00 quarters before / after announcement quarters before / after announcement chance of data-center mobilization chance of data-center mobilization

Did announcements spark local mobilization after ChatGPT?

Through November 2022: no detectable increase

From December 2022: +6.4 percentage points in the quarterly probability of recorded contention

Paci and Russell Figure 4: effects of first announcements on DC-specific and general mobilization and news-theme indices, for counties first announced through November 2022 (wide intervals spanning zero) and from December 2022 (DC-specific index about 6.4 points)

Did state legislators switch from accommodating to restricting data centers?

Through 2022: more accommodating bills

From 2023: more restrictive bills, more than threefold the base rate

  • 0.054 vs 0.015 bills per legislator-quarter

Paci and Russell Figure 5: effects on anti and pro data center bill sponsorship per legislator-quarter, earlier cohort through 2022Q4 versus later cohort from 2023Q1; anti-data-center sponsorship rises to about 0.054 in the later cohort

Does strained power and water, green politics or AI exposure change the response?

No: mobilization and bill sponsorship do not vary with local conditions

Reading: a national shock; opponents fold power, water, environmental and AI grievances into one conflict

Paci and Russell Figure 6: interaction effects of AI exposure, environmental concern and resource competition on mobilization and legislative sponsorship; all intervals include zero

Paci and Russell (2026), Fig. 6 · https://papers.ssrn.com/sol3/papers.cfm?abstract_id=7666018 · On frame alignment in movements: Snow, Rochford, Worden and Benford (1986), American Sociological Review

Do ICE detention centers and new power generation draw the data center response?

ICE detention openings: more immigration-motivated mobilization (about 4.6% of its range)

Planned power generation: more pro- and anti-utility bills (about +0.01 each per legislator-quarter)

Neither reproduces the data center response

Data centers are contested as the AI boom’s local address

Less ordinary infrastructure than a physical embodiment of the AI boom

Partisan alignments are not settled:

  • mobilization crosses party lines
  • the legislative shift is weaker where Republicans hold more seats

Roadside yard sign reading Say No to Data Centers in Christian County, beside a dry cornfield and a rural road

“Say No to Data Centers in Christian County” sign, Mount Auburn Township, Christian County, Illinois, September 26, 2026 · User:Visviva, Wikimedia Commons (public domain) · https://commons.wikimedia.org/wiki/File:Say_No_To_Data_Centers_in_Christian_County_sign_2026-09-26_cropped.jpg

What does backlash do?

Work in progress

Where are data centers on the ballot in 2026?

Local data-center ballot measures

Ballot box or buildout: how could backlash matter?

Votes: voters punish officials tied to unwanted projects

Projects: protest changes what gets built, and where

Backlash Votes Projects

Nuclear plants, dams, wind farms: what did earlier siting fights show?

Planners seek out acquiescent, unorganized communities for nuclear plants, dams and airports: Aldrich, Site Fights (2008), close to 500 Japanese municipalities

Ontario wind turbines and incumbent vote losses: Stokes 2016

Anti-nuclear movements in four democracies: Kitschelt 1986

Will data center announcements move candidates and voters in the 2026 midterms?

Paper 2, with Emily Russell

  1. Do candidates in counties with announcements mention data centers?
  2. Do incumbents with data center votes draw challengers or lose?
  3. Does announcement timing, relative to filing deadlines, shape who runs?

Outcomes: county vote and turnout

The midterm paper: three comparisons, pre-registered before polls close

Announcement

Counties with and without announcements, midterms 2018, 2022, 2026

Incumbent accountability

Local ballot measures, by county and by measure

What does backlash do to the buildout, and how much?

Paper 3, with Emily Russell

Outcomes: cancellation, delay, relocation, costs, concessions

Why is it hard to tell whether protest cancels data centers?

In progress: none of these choices is settled.

Selection: projects that draw protest differ from those that don’t

Baselines: other unwanted infrastructure · other infrastructure · data centers before November 2022

Relocation: projects may move to places with less organizing capacity

Selection Protest Cancellation Baseline? Relocation

Midterms or cancellations: what would you want to learn?

Prompt: pick the midterm paper or the cancellations paper.

  1. Think alone.
  2. With a neighbour: one question, one data source, one reason the answer might mislead.
  3. Share: we collect 3–4 questions from the room.

Written cards welcome; sharing is voluntary.

Local costs, ChatGPT timing, backlash: what did we learn?

  1. Data centers put costs on a town and send the benefits elsewhere: a setting for siting conflict.
  2. Announcements drew contention only after ChatGPT’s release, and not more where local conditions predicted it.
  3. What backlash does, to votes and to the buildout, is open and hard to identify.

Back to your town: yes or no?

See you next class

Next: Wednesday, November 4, Computer Vision and Multimodality (Yiqing Xu)

Our first AI-native paper: from idea to submission in a year

December 2025: first idea

March 2026: work begins

July 5, 2026: registered report submitted to the APSR (reject and resubmit as a full paper)

October 2026: full paper submitted

About 4 months of active work; a comparable traditional paper: about 2–3 years

What did we delegate to agents, and what did we keep?

Agents

Delegated: execution

Researchers

Kept: every substantive decision: framing, empirical design, interpretation, final phrasing

Recorded: what we prompted, which model answered, what the agent did, what we changed

Working with AI Agents and With Great Powers: where to read more

When did Americans start hearing about data centers?

Paci and Russell Figure F.4: share of population aware of data centers in eight poll readings from December 2025 to February 2026, ranging from about 37% to 82%

Equipment, software, R&D, data centers: which AI investments drove 2025 growth?

Investments in Information Processing Equipment: Contribution to GDP Growth

St. Louis Fed chart: information processing equipment investment contribution to GDP growth, percentage points, 1947 to 2025

Investments in Software: Contribution to GDP Growth

St. Louis Fed chart: software investment contribution to GDP growth, percentage points

Investments in R&D: Contribution to GDP Growth

St. Louis Fed chart: R&D investment contribution to GDP growth, percentage points

Investments in Data Centers: Contribution to GDP Growth

St. Louis Fed chart: data center investment contribution to GDP growth, percentage points, 2014 to 2025, y-axis capped at 0.15

Rubinton and Patro, Federal Reserve Bank of St. Louis, January 12, 2026, Figs 1–4; data through Q3 2025 · https://www.stlouisfed.org/on-the-economy/2026/jan/tracking-ai-contribution-gdp-growth

Who did Gallup ask in March 2026, and what did it ask?

Telephone, March 2–18, 2026: 1,000 US adults, all 50 states and DC; margin of error ±4 points

“Overall, would you strongly favor, somewhat favor, somewhat oppose, or strongly oppose the construction of a data center in your area to support artificial intelligence, or AI, technology in the U.S.?”

Gallup topline question 51 with results for March 2 to 18, 2026: strongly favor 7, somewhat favor 20, somewhat oppose 23, strongly oppose 48, no opinion 3; total favor 27, total oppose 71

How much does opposition to data centers depend on how pollsters ask?

Paci and Russell Figure F.5: share of population opposing data centers across polls from July 2025 to May 2026, ranging from about 28% to 71%

Who lives next to xAI’s Colossus in South Memphis?

xAI’s Colossus data center in South Memphis was sited next to Boxtown

NAACP federal Clean Air Act suit over unpermitted gas turbines powering xAI’s data centers

Do legislators’ roll-call votes on data centers vary with AI exposure or environmental concern?

Overall, announcements do not change how legislators vote

Cohorts pooled: fewer pro-data-center votes where AI exposure or environmental concern is high

Paci and Russell Figure E.3: interaction effects on roll-call votes; pro-data-center votes lower where AI exposure and environmental concern are high, anti-data-center estimates include zero

Do projects linked to leading AI firms draw a stronger response?

No: the two groups respond in much the same way

Paci and Russell Figure E.8: announcement effects for AI-firm projects versus other projects on DC-specific mobilization, AI-related news concern and sponsorship balance, with event-study panels; the two groups overlap