Policy Instruments for Electricity Supply to AI Data Centers: A Scoping Review

Abstract

Title: Policy Instruments for Electricity Supply to AI Data Centers: A Scoping Review Objective. To map how the electricity supply of AI data centers is governed: which policy instruments are documented, which actors and levels of government hold authority, how problems and distributional burdens are framed, what evidence supports claims, and which contexts are represented. Research questions. Central: How is the electricity supply of AI data centers being governed, through which instruments and by which actors, and on what evidence? SQ1. Which policy instruments are documented, and how do they classify under established instrument typologies? SQ2. Which institutions and levels of government hold decision authority over these instruments? SQ3. How do sources frame problems, and who is said to bear costs and risks? SQ4. What types of evidence support claims, and which outcomes are observed rather than projected? SQ5. Which jurisdictions and power-system contexts are represented, and which are absent? Framework. Arksey & O'Malley (2005); Levac et al. (2010); JBI guidance for scoping reviews; reporting follows PRISMA-ScR. Eligibility. Period and language: published 2022 to the search date; English. Included types: journal articles, reviews, preprints, conference papers, book chapters, reports, and primary government and regulatory documents. Excluded: news; opinion without data; law-firm client alerts (used only to locate primary documents); dissertations; datasets; books; conference abstracts; studies limited to scheduling or optimization algorithms. Topic: policy instruments or arrangements governing electricity supply to AI data centers. Water and emissions are in scope only when linked to a supply instrument. Sources. OpenAlex (primary); Google Scholar; SSRN; targeted searches of regulator, international-organization and national-laboratory websites; forward and backward citation chasing from core papers. Web of Science and Scopus are not available to the review team; this is stated as a limitation. If access is obtained through a collaborator, the string in Appendix B will be run once and the records merged after de-duplication. Search. Boolean string in Appendix A, run on the date recorded at execution; counts recorded. Preprint and published versions of the same work are merged, keeping the published version. Pilot calibration (disclosure). Before registration, pilot searches in OpenAlex (October 2026) were used to calibrate the search string, to set the start of the period to 2022 (almost no relevant records predate 2024 under an AI-specific definition), and to run a known-item recall check (three of four known relevant items were retrieved). No records had been screened at the time of registration. Screening. Title/abstract screening, then full-text screening, against the criteria above; reasons for full-text exclusion recorded; PRISMA-ScR flow diagram. A 10% random sample of decisions will be re-screened after a time interval to check consistency. Data charting. Jurisdiction; instrument; instrument class; actors and level of government; problem framing; evidence type; peer-reviewed or not; observed or projected outcomes; context. Synthesis. Deductive coding with established policy-instrument typologies (nodality, authority, treasure, organization), supplemented by inductive coding. Peer-reviewed and non-peer-reviewed sources are reported separately. No formal quality appraisal is performed (scoping review); evidence type is charted instead. Changes. Any deviation from this protocol will be documented and justified.