Skip to content
01 of 05AI, GDP, Well-being and Energy

AI, GDP, Well-being and Energy

Library

Series and technical notes.

You are in AI, GDP, Well-being and Energy · Introduction.

Series

AI, GDP, Well-being and Energy

5 items

Watch video, summary and related content

Complete General ~40 min 4 chapters
PrerequisitesAI and Generative AI Foundations

This series examines the relationship between electricity, productivity/GDP, well-being, and the role of AI as a new compute-intensive—and therefore energy-intensive—technology.

The goal is to answer two questions:

  • Why does access to reliable, inexpensive electricity so often accompany jumps in well-being?
  • Is AI already affecting GDP/productivity, or is the effect still difficult to see—and why?

Contents

1. Electricity → well-being: the real mechanisms

  • Why reliable, inexpensive electricity enables health, logistics, industry and services.
  • The difference between quantity (kWh) and quality (reliability, stability and outage cost): thresholds and diminishing returns.

2. AI as an electrical technology

  • What AI means in terms of compute, data centers, training vs inference and efficiency—and why demand can keep growing even as hardware improves.
  • The main bottlenecks: energy, chips, data, talent and regulation.

AI as an electrical technology

The scale is already material: data centers consumed 415 TWh in 2024, and the IEA projects between 945 and 1,260 TWh by 2030.

AI load 2024
~62 TWh
Approximately 15% of the total, using accelerated servers as a proxy for AI workload.
AI-specific 2030
554 TWh
Greenpeace estimate for AI-specific workloads. Its methodology differs from the IEA base scenario.
Typical load
100 MW
Order of magnitude for a typical AI data center. The largest facilities under construction are around 2 GW—roughly twenty times more.

Sources: IEA Energy and AI (2025) for 415 / 945 / 1,260 TWh, the 1.5% share, comparison with Japan and the ~100 MW order of magnitude for a typical AI data center; Greenpeace/Öko-Institut (2025) for 554 TWh of AI-specific load in 2030. Caveat: the 2024 AI share uses accelerated servers as a proxy because there is no perfectly clean physical boundary between AI and non-AI workloads.

3. Measurement: GDP vs well-being

  • GDP is not well-being: health, education, safety, access to services, inequality and misleading averages.
  • Subjective well-being and the cases where it diverges from material well-being.

4. AI and GDP today: real impact, lags and early signals

  • Why macro impact takes time: diffusion, reorganization, intangible capital and complementary investments.
  • Where effects appear before GDP: task productivity, quality, time saved, new products/services and value that national accounts measure poorly.

Related series: AI and Generative AI Foundations · Data Centers in Space

View all series