Shift Key with Robinson Meyer and Jesse Jenkins

A Skeptic’s Take on AI and Energy Growth

58 snips
Apr 3, 2024
Jonathan Koomey discusses the impact of AI on energy consumption and the history of computing's energy use. The episode explores data centers, AI-related computing, and electricity demand growth. Topics include energy efficiency improvements in GPU for AI training, balancing energy demand growth, and AI models in pricing and sustainability.
Ask episode
AI Snips
Chapters
Transcript
Episode notes
INSIGHT

Data Center Electricity Use

  • Data centers use roughly 1% of US electricity, while networks and end-user devices use another 3%.
  • AI-specific data centers represent a smaller fraction within the total data center electricity usage.
INSIGHT

Extrapolation Concerns

  • Extrapolating recent AI growth to predict massive future energy consumption is misleading.
  • Efficiency improvements often offset increased service demand, even if it is substantial.
ANECDOTE

Hyperscale Data Center Efficiency

  • From 2010-2020, global internet traffic grew 17x, while data center electricity demand only grew by 10%.
  • Hyperscale data centers improved efficiency and utilization, enabling service demand growth.
Get the Snipd Podcast app to discover more snips from this episode
Get the app