Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Exploding AI Power Use: an Opportunity to Rethink Grid Planning and Management

Liuzixuan Lin, Rajini Wijayawardana, Varsha Rao, Hai Nguyen, Wedan Emmanuel Gnibga, Andrew A. Chien

TL;DR

The paper investigates how the surge in AI-powered datacenter demand interacts with grid resource adequacy, using EirGrid and Dominion as hotspots and CAISO, ERCOT, and SPP as benchmarks. It employs a LOLE-based framework and a datacenter load-growth model to assess whether current grid plans can accommodate 5 years of AI-driven growth, and it explores reliability-relaxation schemes as a potential lever. Key findings show EirGrid may require relaxing datacenter reliability to 0\% for new connections to meet AI demand, while Dominion remains short even with relaxation; CAISO, ERCOT, and SPP appear to have sufficient capacity under the studied scenarios. The work argues for rethinking adequacy assessment and grid planning through coordinated planning, load-flexibility models, and adaptive load abstractions, with broad implications for policy and infrastructure investment in the AI era.

Abstract

The unprecedented rapid growth of computing demand for AI is projected to increase global annual datacenter (DC) growth from 7.2% to 11.3%. We project the 5-year AI DC demand for several power grids and assess whether they will allow desired AI growth (resource adequacy). If not, several "desperate measures" -- grid policies that enable more load growth and maintain grid reliability by sacrificing new DC reliability are considered. We find that two DC hotspots -- EirGrid (Ireland) and Dominion (US) -- will have difficulty accommodating new DCs needed by the AI growth. In EirGrid, relaxing new DC reliability guarantees increases the power available to 1.6x--4.1x while maintaining 99.6% actual power availability for the new DCs, sufficient for the 5-year AI demand. In Dominion, relaxing reliability guarantees increases available DC capacity similarly (1.5x--4.6x) but not enough for the 5-year AI demand. New DCs only receive 89% power availability. Study of other US power grids -- SPP, CAISO, ERCOT -- shows that sufficient capacity exists for the projected AI load growth. Our results suggest the need to rethink adequacy assessment and also grid planning and management. New research opportunities include coordinated planning, reliability models that incorporate load flexibility, and adaptive load abstractions.

Exploding AI Power Use: an Opportunity to Rethink Grid Planning and Management

TL;DR

The paper investigates how the surge in AI-powered datacenter demand interacts with grid resource adequacy, using EirGrid and Dominion as hotspots and CAISO, ERCOT, and SPP as benchmarks. It employs a LOLE-based framework and a datacenter load-growth model to assess whether current grid plans can accommodate 5 years of AI-driven growth, and it explores reliability-relaxation schemes as a potential lever. Key findings show EirGrid may require relaxing datacenter reliability to 0\% for new connections to meet AI demand, while Dominion remains short even with relaxation; CAISO, ERCOT, and SPP appear to have sufficient capacity under the studied scenarios. The work argues for rethinking adequacy assessment and grid planning through coordinated planning, load-flexibility models, and adaptive load abstractions, with broad implications for policy and infrastructure investment in the AI era.

Abstract

The unprecedented rapid growth of computing demand for AI is projected to increase global annual datacenter (DC) growth from 7.2% to 11.3%. We project the 5-year AI DC demand for several power grids and assess whether they will allow desired AI growth (resource adequacy). If not, several "desperate measures" -- grid policies that enable more load growth and maintain grid reliability by sacrificing new DC reliability are considered. We find that two DC hotspots -- EirGrid (Ireland) and Dominion (US) -- will have difficulty accommodating new DCs needed by the AI growth. In EirGrid, relaxing new DC reliability guarantees increases the power available to 1.6x--4.1x while maintaining 99.6% actual power availability for the new DCs, sufficient for the 5-year AI demand. In Dominion, relaxing reliability guarantees increases available DC capacity similarly (1.5x--4.6x) but not enough for the 5-year AI demand. New DCs only receive 89% power availability. Study of other US power grids -- SPP, CAISO, ERCOT -- shows that sufficient capacity exists for the projected AI load growth. Our results suggest the need to rethink adequacy assessment and also grid planning and management. New research opportunities include coordinated planning, reliability models that incorporate load flexibility, and adaptive load abstractions.
Paper Structure (17 sections, 5 equations, 13 figures, 4 tables)

This paper contains 17 sections, 5 equations, 13 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Problem
  3. Study: EirGrid
  4. Methodology
  5. Resource Adequacy Assessment
  6. Datacenter Load Growth Model
  7. New Schemes to Accommodate Datacenter Load
  8. Results
  9. Other Grids
  10. Revisiting AI Power Growth and Grid Readiness
  11. Research Opportunities
  12. Coordinated Planning and Co-investment
  13. New Models for Reliability
  14. New Abstractions for Load and Adaptive Load Management
  15. Grid Resource Adequacy Assessment Details
  16. ...and 2 more sections

Figures (13)

  • Figure 1: EirGrid DC load is growing rapidly, and AI accelerates it. By 2025, Cloud DC growth cannot be supported, and by 2024, Cloud+AI DC growth cannot be supported.
  • Figure 2: EirGrid's Reliable Load Limit (blue line) and Resources, 2022--2028. The load limit does not increase despite a 30% increase in renewable generation capacity.
  • Figure 3: EirGrid Grid Datacenter Capacity Limits with Varied New Datacenter Power QoS vs. Demand.
  • Figure 4: Datacenter Outage Rates (daily fraction), 2028 EirGrid: using 0% reliable DCs to meet 900 MW AI demand. Outages are rare (0.4%, 34.5 hours) and seasonal (winter).
  • Figure 5: Dominion Grid Datacenter Capacity Limits with Varied New Datacenter Power QoS vs. Demand.
  • ...and 8 more figures