Uncertain AI: 2030 US Power Sector Futures

Sources and Methodology

Original Article and Analysis Download

1) Core Analysis Framework

Findings synthesize four key data streams:

  1. Announced AI infrastructure projects and investments

  2. Utility capital deployment and infrastructure costs

  3. Regional grid expansion and upgrade requirements

  4. Industry growth patterns and technology trends

2) Primary Source Documents

Grid & Infrastructure

  • LBNL “United States Data Center Energy Usage Report” (2024)

  • EIA "Annual Energy Outlook 2024"

  • EPRI "Powering Intelligence: Analyzing Artificial Intelligence and Data Center Energy Consumption” (2024)

  • IEA "Data Centres and Data Transmission Networks" (2023)

  • DOE "Transmission Needs Study" (2023)

  • NERC "Long-Term Reliability Assessment" (2023)

AI & Chip Technology

  • Nvidia H100/H200 Technical Documentation

  • DeepSeek Model Architecture Papers

  • Microsoft Phi-2 Technical Reports

Data Center Design & Operations

  • Uptime Institute Global Survey (2023)

  • Open Compute Project Standards

  • Green Grid PUE Standards

  • Hyperscale Design Specifications

3) Major Project Examples

Hyperscale Developments Examples

  • Meta's Texas AI Infrastructure (500MW initial capacity, $800M grid infrastructure requirements)

  • Microsoft Virginia Expansion (400MW total capacity, $1.2B including grid infrastructure)

  • Google Ohio Development (350MW planned capacity, $1B infrastructure investment)

Regional Grid Projects Examples

  • Northern Virginia (Current capacity: 3.5GW, Dominion Energy: $4B data center support)

  • Texas ERCOT ($2.8B transmission for new loads, Multiple 345kV upgrade projects)

  • PJM Territory ($3.2B in data center related upgrades, Major transmission corridor expansions)

4) Methodology

Our analysis integrates:

  • Baseline Data Center Demand of 176 TWh, LBNL, 2024

  • AI-specific grid investments estimated at 15-25% of total grid modernization needs

  • Four distinct scenarios to capture the uncertainty in AI's energy trajectory:

    • Conservative estimate (Phantom Datacenters): 10-20% AI workload

    • Moderate scenarios: 40-60% AI workload

    • Aggressive scenario (Gridlock Boom): 70-80% AI workload

  • Percentages derived from

    • Computational intensity

    • Efficiency gains and model architectures

    • Regional concentration patterns

    • Announced AI facility power requirements

    • Grid upgrade requirements for high-density loads

  • Growth Rate Analysis derived from

    • Historical trends

    • Announced expansion plans

    • Regional capacity constraints

    • Technology evolution pathways, generational AI efficiency estimates

Disclaimer

Investment estimates and demand growth derived from meta-analysis of IEA, EIA, and EPRI reports on future grid expansion needs (2024). All projections are subject to significant uncertainty and should be used for strategic planning purposes only.

Original Article and Analysis