Catalyst with Shayle Kann

The mechanics of data center flexibility

57 snips
Aug 28, 2025
Varun Sivaram, Founder and CEO of Emerald AI, discusses the fascinating world of flexible data centers. He explains how adjusting data center loads can significantly alleviate pressure on power grids, potentially unlocking up to 98 gigawatts of unused capacity. Varun highlights the challenges of maintaining high reliability while enabling flexibility in power usage. He shares insights on innovative strategies like demand response and discusses case studies, including Google's latest agreements, to illustrate how adaptability can transform data centers into valuable grid assets.
Ask episode
AI Snips
Chapters
Transcript
Episode notes
INSIGHT

AI Factories Convert Power More Efficiently

  • AI data centers convert electricity into compute outputs with much less non-IT loss than before.
  • Modern AI factories can turn 80–90% of power into computations, increasing grid stress.
INSIGHT

Grids Plan For Worst-Case Nameplate Peaks

  • Grid planners assume a data center will request its full nameplate capacity at worst possible times.
  • Interconnection studies plan for peaks across many years, forcing cautious capacity upgrades.
INSIGHT

Training Vs Inference Load Shapes Differ

  • Training workloads are spiky with ramps and checkpoints, while inference is smoother but still unpredictable.
  • A single data center's profile can change over time as workloads shift between training and inference.
Get the Snipd Podcast app to discover more snips from this episode
Get the app