The Looking Glass Initiative
What has changed, by how much, and what evidence supports that claim?
Data centers have supported banking, telecommunications, healthcare, and the internet for decades. The question is not whether data centers exist — it is whether the current AI-driven expansion materially changes electricity demand, water consumption, infrastructure requirements, or community impacts compared to previous growth cycles.
Society already accepted data centers as part of modern life. The inquiry is: what is different now, and can those differences be measured, documented, and explained in plain language?
This is a transparency project, not an anti-technology project. The Looking Glass Initiative makes the scale visible so that communities, policymakers, and ratepayers can ask informed questions.
What is materially different?
| Dimension | Prior generations | AI buildout |
|---|---|---|
| Purpose | Transactions, storage, communication | Large-scale model training and inference |
| Power demand | Moderate, predictable growth | Rapid, regionally concentrated growth |
| Hardware density | CPUs, standard storage | Massive GPU clusters; higher rack density |
| Grid impact | Localized, utility-scale planning | Regional and national planning concern |
| Water usage | Present but lower density | Often significantly higher cooling demand |
| Public visibility | Mostly invisible | Increasingly subject to permitting scrutiny |
Tier 3 · Survey / Research Institution · McKinsey Quarterly, April 2025
McKinsey Quarterly projects $3.7–$7.9 trillion in total AI infrastructure investment by 2030 across three demand scenarios. Global data center capacity may triple, from 82 GW in 2025 to as much as 219 GW by 2030 — a 3.5× increase. AI workloads are projected to account for approximately 70% of new demand, rising from 44 GW to 156 GW of dedicated AI capacity. Power-sector investment (utilities, energy providers, cooling) is projected at approximately $1.3 trillion to meet this demand, with renewables expected to account for 45–50% of the energy mix by 2030.
Source: McKinsey Quarterly — “The Cost of Compute: A $7 Trillion Race to Scale Data Centers” (Noffsinger, Patel, Sachdeva et al., April 2025) · Tier 3 · These are McKinsey's projections, not BEI conclusions.
Tier 2 · Government-Commissioned · Lawrence Berkeley National Laboratory, December 2024
U.S. data centers consumed 176 TWh in 2023 — approximately 4.4% of total U.S. electricity. DOE research projects 325–580 TWh by 2028.
Source: Lawrence Berkeley National Laboratory / U.S. Dept. of Energy — 2024 United States Data Center Energy Usage Report (Dec 2024) · Tier 2.
Explore the evidence
All figures cited with source, date, and evidence tier. McKinsey figures are that organization's projections — BEI surfaces them as research context, not as BEI conclusions. LBNL figures are from a U.S. government-commissioned report. This page contains no BEI-authored scoring or verdicts. Aggregator, not author (PAT-035).