Catalyst with Shayle Kann The rise of flexible data centers
19 snips
Apr 9, 2026 
Varun Sivaram, CEO of Emerald AI and expert in aligning data centers with power grids. He discusses tapping unused grid capacity and why power flexibility is becoming profitable. They explore differences between training and inference flexibility, the role of batteries and turbines in a mini dispatch curve, and plans for a 100 MW power‑flexible AI factory.
AI Snips
Chapters
Transcript
Episode notes
Stranded Grid Capacity Exists For AI Factories
- Data center flexibility can tap into 100+ gigawatts of stranded grid capacity by curtailing briefly during rare peaks.
- Varun cites grid utilization around 50% most of the year and says short, infrequent curtailments unlock huge existing capacity.
Data Centers Driving Most Projected Peak Load Growth
- Data centers are now the dominant driver of projected peak load growth in some regions, e.g., 94% of PJM's projected growth.
- Varun cites NERC and EPRI forecasts showing large future power shares (up to 17% of US power by 2030).
Compute Service Tiers Are Ahead Of Utilities
- There's a major schism: cloud providers now offer differentiated compute tiers while utilities largely still offer one firm electric service.
- Varun notes Google's flex/priority inference and contrasts it with utilities' lack of non‑firm interconnection products.