Democratic gains will delay some US data-center projects

Gina Hinojosa has begun assembling a Texas administration that must reconcile her proposed data-center moratorium with promises of jobs and lower household costs. Her Nov. 5 transition announcement named Ryan Pollock, head of the Texas State Building and Construction Trades Council, and former Republican state representative Ernest Bailes as co-chairs. Their roles put construction jobs alongside family costs in the transition, not an agreement on restrictions. The Associated Press reports that Hinojosa defeated Republican Greg Abbott and Democrats won both chambers of Congress.

Rackline forecasts a modest net drag on new US data-center capacity through 2027, concentrated in projects awaiting permits or power contracts, rather than a national construction freeze. The larger changes should be in location and customer obligations. That forecast rests on enough projects continuing through existing approval routes; the national share of 2027 deliveries already permitted or power-contracted remains unknown.

Texas's divided Legislature limits how far Hinojosa can turn campaign demands into law. AP calls give Democrats the House while Republicans retain the Senate. Her proposals for a moratorium until lawmakers enact protections and an end to data-center tax incentives are not enacted policy. The Legislature convenes Jan. 12, 2027, with a two-year budget to pass. Tax changes require legislation; her campaign demand does not establish authority for a blanket executive ban.

Texas's utility regulator will not automatically change hands with the governor. Its five commissioners serve six-year, governor-appointed, Senate-confirmed terms; the governor selects the chairman. Hinojosa gains executive authority only after installation. ERCOT intends to publish its data-center impact report by Dec. 10 under a process begun in August. That report concerns developers pursuing connections, not committed generating or computing capacity.

Congressional gains provide leverage over agencies without removing the president's veto. New terms begin Jan. 3, 2027. Oversight and appropriations negotiations can press questions about subsidies and consumer costs, but statutory restrictions require legislation. Congressional control does not itself rewrite local permits or state utility tariffs.

Arizona's utility election adds scrutiny without delivering Democratic control. AP called two commission seats for Clara Pratte and Jonathon Hill; Republicans retain three of five seats. Commissioners are elected separately from the governor, who fills vacancies. AP also called Democratic Gov. Katie Hobbs's reelection. She signed a law June 13 restricting new applications for a specific data-center tax-relief program from July 1, 2026, through June 30, 2029, not construction. Her task force's grid-permitting and transmission proposals remain recommendations. Neither measure is a consequence of this vote.

Virginia shows how customer protections can raise funding requirements without banning projects. Dominion's approved large-load class starts Jan. 1, 2027, under a pre-election decision in case PUR-2025-00058. Qualifying new customers contracting from then face 14-year service obligations; those without sufficient credit can also face collateral demands. A September order accommodated statutory opt-outs for six existing customers. These protections can require developers to commit cash before revenue arrives while preserving a route to service.

Pennsylvania preserves a conditional route under a returning Democratic governor. AP reports Josh Shapiro won reelection. His effective Aug. 18 order permits rolling environmental review for qualifying new permit applications when developers sign binding GRID agreements, while permit issuance still requires local approvals. Shapiro also backed Amazon's announced Salem and Falls township investments.

Georgia's two Democratic winners campaigned on the cost of supplying data centers. Shelia Edwards wants data centers to pay their full infrastructure costs; incumbent Peter Hubbard's October analysis argues for solar and storage instead of combined-cycle gas plants. Those individual positions are not a new majority platform or commission orders. Official biographies and unofficial returns point to a switch from 3–2 Republican to 3–2 Democratic control Jan. 1. Georgia Power's Nov. 6 AWS filing in Docket 44280 remains subject to the existing review clock, which can produce deemed approval before then.

Projects still seeking approvals expose lenders to longer carrying periods before operating income begins. Developers able to fund security and minimum payments have an advantage over those that cannot. The utility customer initially bears those obligations; whether an operator passes them to tenants depends on its contracts. CBRE reports increasing minimum-power payment clauses. Such protections can reduce cost-shifting risk for households, but they do not guarantee lower bills.

Relocation reduces the national loss only when another US site can deliver in time. CBRE's 80.4% preleasing figure covers under-construction capacity in eight primary markets, not permits or a national 2027 delivery share. PJM identified a supply-demand imbalance in January. Moving to another constrained region therefore does not create usable power; an offshore opening preserves corporate compute, not US capacity, and a delivery after 2027 misses this forecast window.

AI-model development is less directly exposed than an individual campus: delaying a new site does not remove existing compute, but future capacity still matters. Google reported October model activity alongside capacity constraints, while Amazon says OpenAI's Trainium commitment begins in 2027. Rackline does not expect these policy mechanisms alone to interrupt AI-model development generally; existing activity does not guarantee future capacity. The base case fails if restrictions across major markets or unaffordable power obligations prevent replacement sites from opening and companies revise aggregate US deliveries downward for those reasons. If those cuts reduce capacity needed for model work, they also threaten AI progress.