EUVIMEDEUVIMED entdecken
Crossref

Multiscale Grid Intelligence to Fight Artificial Intelligence Data Center Grid Defection: Unlocking a Faster, Cheaper, and Cleaner On-Grid AI Rollout

Thomas Morstyn, Yihong Zhou, Ian Whitfield

IEEE Energy Sustainability Magazine · 2026-02 · 2 (1) · 29-45

DOI: 10.1109/esm.2025.3628276 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Energy Sustainability Magazine, 2026-02. Verfasst von Thomas Morstyn und 2 weiteren Autor:innen. Die Arbeit wurde bisher 2-mal zitiert (Crossref) und nennt 7 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Rapid advances in artificial intelligence (AI) computing capabilities have led to a race to build larger-scale data centers with escalating power demands. AI companies are now planning data center sites with up to 5 GW capacity, and the International Energy Agency (IEA) projects that data center demand could grow to 1,700 TWh by 2035 (6% of current total global electricity demand). Power grid connections are becoming a major bottleneck, and it can now take more than five years to receive a grid connection for a new data center in the United States and Europe. The challenge of getting grid connections has driven a new trend toward grid defection, where data centers are being fully or partially supplied by local microgrids with dedicated generation assets. In some regions this is a voluntary choice, but in most cases, it is the only option available for building a new data center. For most data centers, it is likely that a substantial grid connection would be beneficial from both an economic and a sustainability perspective due to how main power grids unlock generation economies of scale and enable lower-cost renewable integration. However, inefficient AI data center grid defection is being driven by a reliance on preexisting technical approaches and institutional barriers to change. In this article, we propose “multiscale grid intelligence” as a new framework to fight AI data center grid defection and support closer coordination between data centers and power grids. The proposed framework embeds power-grid-focused analytics into how data centers are planned and operated, accounting for cooptimization opportunities across spatial scales (from data center rack-level distribution to national transmission) and time scales (from second-to-second frequency balancing to years-ahead network planning). Across these scales, we identify a range of opportunities for intelligent planning and operation that could offer significant value if implemented together. The article discusses the value multiscale grid intelligence can offer power system and AI stakeholders and proposes key areas for future research.

Zitieren

APA

Morstyn, T., Zhou, Y., & Whitfield, I. (2026). Multiscale Grid Intelligence to Fight Artificial Intelligence Data Center Grid Defection: Unlocking a Faster, Cheaper, and Cleaner On-Grid AI Rollout. IEEE Energy Sustainability Magazine, 2(1), 29-45. https://doi.org/10.1109/esm.2025.3628276

Vancouver

Morstyn T, Zhou Y, Whitfield I. Multiscale Grid Intelligence to Fight Artificial Intelligence Data Center Grid Defection: Unlocking a Faster, Cheaper, and Cleaner On-Grid AI Rollout. IEEE Energy Sustainability Magazine. 2026;2(1):29-45. doi:10.1109/esm.2025.3628276

BibTeX
@article{Morstyn2026,
  author = {Morstyn, Thomas and Zhou, Yihong and Whitfield, Ian},
  title = {Multiscale Grid Intelligence to Fight Artificial Intelligence Data Center Grid Defection: Unlocking a Faster, Cheaper, and Cleaner On-Grid AI Rollout},
  journal = {IEEE Energy Sustainability Magazine},
  year = {2026},
  volume = {2},
  number = {1},
  pages = {29--45},
  doi = {10.1109/esm.2025.3628276}
}
Kostenlos starten →