Powering sustainable AI in the United States

Overview

1/3

The rapid expansion of Artificial Intelligence (AI) is poised to significantly increase U.S. electricity demand, at a time when electricity is becoming a newly scarce resource.

2/3

This surge necessitates strategic planning and infrastructure modernization to avoid potential energy crises and ensure sustainable AI development.

3/3

This report outlines the key drivers of this demand, potential scenarios, and actionable strategies for mitigating risks and enabling sustainable growth.

Key findings

01

Decisions made today regarding AI U.S. infrastructure will determine future AI energy demand

AI could drive 20-50% of U.S. electricity demand growth between 2025-2030, highlighting the need for proactive planning to align AI load growth with a modern energy infrastructure.
02

Sustainable AI is not solely about AI power reduction

Infrastructure deficits and regulatory hurdles can limit a harmonious development of sustainable AI. These challenges may intensify due to insufficient investment in behind-the-meter solutions, microgrids, and flexibility, all of which require strategic reform.
03

Unregulated AI growth risks creating unsustainable infrastructure

Unchecked AI development, potentially reaching 500 TWh by 2030, could lead to oversized and inefficient infrastructure expansions, failing to meet the need for stable, affordable, and accessible electricity to power sustainable AI development.
04

Swift acceleration of AI-ready data center build-out is critical

Achieving 43-92 GW additional data center capacity by 2030 underscores their critical strategic role, requiring streamlined permitting, modular construction, and distributed energy for resilience and scale.
05

Mismanagement of AI growth threatens grid stability

High-end AI growth is forecast to drive power demand to 157 GW by 2029 (29 GW over projections, 173 GW by 2030), potentially overloading the grid and impeding broader electrification efforts.
06

Geographic clustering reveals local vulnerabilities

Driven by data center concentrations, seven ISOs/RTOs, including MISO, PJM, ERCOT, risk reserve shortfalls by 2028, facing critical regional grid constraints. Powering AI sustainably requires addressing these issues.