With Great Power Can AI help save the grid?
9 snips
Mar 17, 2026 
Varun Sivaram, CEO and founder of Emerald AI and former energy advisor with a physics and energy policy background, discusses making AI data centers grid-friendly. He shares demos that cut power use, how software creates temporal and spatial flexibility, and plans for a 100 MW “AI factory” in Virginia. Conversations cover partnerships with utilities, hyperscalers, and an all-of-the-above approach to scaling AI without stressing the grid.
AI Snips
Chapters
Transcript
Episode notes
AI Data Centers As Flexible Grid Resources
- AI data centers can act as flexible grid resources by slowing, moving, or using on-site assets to reduce load during peak moments.
- Emerald's OS forecasts grid constraints and orchestrates temporal, spatial, and on-site resource flexibility across computations and batteries.
Phoenix Demo Cut AI Load 25% For Three Hours
- Emerald's demonstration with Oracle, NVIDIA, EPRI, and Salt River Project reduced a data center's AI power use by 25% for three hours.
- That result was peer-reviewed and published in Nature Energy as the first rigorous demo of AI flexibility.
Use Flexibility To Avoid Costly Grid Expansion
- Utilities should use Emerald to dispatch flexible data centers instead of expanding grid capacity for every new customer.
- Data centers run Emerald Conductor so they can receive utility signals and reduce load during peak moments without harming mission-critical jobs.