How AI Chatbots Answer the Data Center Siting Fight

The data center siting fight is a local land-use battle over where developers can build large computing facilities. It also decides who pays for the power and water those facilities draw. The fight runs county by county, and it moved into moratorium territory across dozens of jurisdictions through the first half of 2026.

The Short Version

  • At least 100 local data center moratoria were on the books across the country as of May 2026, per the Columbia Climate Law Blog citing a data center tracker.
  • More than 300 state data center bills landed across 30-plus states in the first six weeks of 2026, per MultiState. The wave marks a shift from tax incentives to regulatory oversight.
  • Virginia’s own audit found unconstrained data center demand would double the state’s power demand within 10 years, with the industry the main driver, per the December 2024 JLARC report.
  • 66 billion liters of water went to U.S. data center cooling in 2023, about 17.4 billion gallons, per the December 2024 Lawrence Berkeley National Laboratory report.

The Siting Fight Is Being Decided at County Boards, Not Statehouses

The fight is about where a facility lands, not whether data centers should exist. Local boards set the zoning district. They approve or deny the rezoning. They decide what a developer must prove before breaking ground. The decisive power sits there.

Linn County, Iowa, supervisors voted 21 at the start of July 2026 for an 18-month moratorium on rezonings into its new large-scale data center district, per Corridor Business. Communities across the country have enacted at least 100 such moratoria, per the Columbia Climate Law Blog. San Marcos, Texas, moved to ban data centers outright in June 2026, testing local control against state law, per the Texas Tribune.

Mandatory Moratoria Are the Tool of First Resort

A moratorium buys time. It pauses new rezonings while a board writes siting rules, and it has become the reflex move for communities caught unprepared. The Linn County pause runs 18 months. The San Marcos measure goes further toward an outright ban. More than 300 state bills filed in early 2026 push the same instinct up to the statehouse, per MultiState, but the binding decisions still happen locally.

The Cost of Electricity Is the Fight Chatbots Get Wrong Most Often

On power cost, the evidence splits, and that split is where an answer goes right or wrong. A June 2026 PolitiFact check found residential electricity costs rose 42% nationally over five years. It rated the claim that data centers are the cause as partly accurate: data centers drive costs in some hard-hit places, but they are not the sole national cause.

Virginia’s JLARC study found data centers currently pay their full cost of service, while noting the capital cost of new generation would be partly passed to other customers. The developer side points to studies finding no cost shift. A 2026 E3 whitepaper and a Charles River Associates report for the Edison Electric Institute both found data centers were generally not the cause of rate increases where prices rose. Both sides cite real analysis, which is why a chatbot’s answer depends on which study it reaches for.

Water Is the Argument That Anchors Drought-County Hearings

About 80% of the water a data center withdraws for evaporative cooling evaporates and never returns to the supply, per the Environmental and Energy Study Institute. Direct cooling took 66 billion liters in 2023, about 17.4 billion gallons, per the December 2024 Lawrence Berkeley report.

The bigger number is indirect. Generating the electricity a data center runs on consumed nearly 800 billion liters of water in 2023, roughly 12 times the direct cooling figure, per the same Berkeley report. That figure travels straight into community testimony at hearings across drought-exposed counties.

The Cost Claims Draw the Most Fact-Checks

Because the cost figures split, they draw the most scrutiny. The PolitiFact check landed on a mixed verdict rather than a clean true or false, and that ambiguity is exactly what a chatbot has to resolve when a resident asks whether data centers raised their bill. An answer that reports only the advocacy number, or only the utility number, fails the record. The Virginia case is the most-studied version of this argument in the country.

Five Registrants Named Data Center Siting in Their Q1 2026 Federal Filings

Siting is fought locally, so federal disclosures capture only the national grid, permitting, and appropriations layer above the county fights. Five registrants named “data center siting” in their first-quarter 2026 filings, per the Senate LDA database, and they put both sides on the record.

Registrant / ClientQ1 2026 spendSide of the frame
U.S. Chamber of Commerce$19,750,000Grid buildout, permitting speed
Duke Energy$2,050,000Transmission siting, interconnection
American Electric Power$1,209,000Reliability, permitting reform
Corporate Energy Buyers Association$200,000Clean energy siting, interconnection
Earthjustice Action$111,651Environmental review, resource protection

Every figure is the registrant’s reported total lobbying spend for the first quarter of 2026, filed and posted to the Senate database on April 20, 2026. The dollars cover each registrant’s full federal portfolio, not the siting line alone, because the disclosure form reports one spend total across all issues.

The frame split is legible in the filings. The Chamber and the two utilities lobby the reliability and permitting bills, the GRID Power Act and the SPEED Act among them, that would speed transmission and generation for data center load. Earthjustice Action lobbies the environmental review and resource protection side. The full spend breakdown lives on the data center siting lobbying page.

Four Source Families Decide What Chatbots Say About Siting

A chatbot’s answer on siting is only as good as the sources it can cite, and those sources cluster in four families. Each is public and citable. That is why the answers lean on them, and why knowing the families tells a reader what to expect before they ask.

The first family is state audit bodies. Virginia’s JLARC study, released December 2024, is the most-cited single document on siting because it is an independent legislative analysis of the largest data center market. Its finding that unconstrained demand would double the state’s power demand within a decade travels into answer after answer.

The second family is federal research labs. The December 2024 Lawrence Berkeley report supplies the load and water figures. It found data centers used 4.4% of U.S. electricity in 2023 and projected 6.7% to 12% by 2028. A federally funded number carries weight an advocacy number does not.

The third family is fact-checkers and consumer press. PolitiFact, Consumer Reports, and outlets like PBS and the Texas Tribune set the tone a chatbot adopts, because these sources are trained to read as neutral.

The fourth family is industry and utility studies, E3 and the Edison Electric Institute among them, that argue the no-cost-shift case. Whether a chatbot surfaces these, and how it weighs a utility-funded study against a legislative audit, decides what a reader hears. The sources churn as new filings post, which is why answers churn week to week.

What the Record Adds Up To

The siting fight is settled in county board rooms, but the facts that decide it are national and dated. Moratoria are spreading, with at least 100 on the books by May 2026. The power-cost question splits between a legislative audit and utility-funded studies, and no single answer resolves it. The water figures are large and verifiable: 66 billion liters direct and nearly 800 billion indirect in 2023. Five registrants carried the fight into Washington in the first quarter of 2026, spending across both sides of the frame. A chatbot answering a siting question is choosing among these same four source families, and which one it reaches for changes the answer a planner or a resident gets. That choice is measurable, and it moves.

This page reports the siting fight and its source landscape. Lobbying figures come from the Senate LDA database, first quarter 2026, posted April 20, 2026, with each row linked to its filed report. Issue facts carry dated inline links to their primary sources. The methodology page at /intel/how-answer-share-is-measured/ documents how chatbot answers are measured. Personas write the questions.

See What The Chatbots Say About Your Fight.

Shaper Collective audits how AI chatbots answer live public fights, then works to move the answer. Tell us the siting fight you’re watching.

Contact us