Overview
- Emerald AI's Conductor modulated flexible GPU jobs on a 256‑GPU cluster, lowering power draw by about a quarter for three hours during grid stress with minimal performance impact, according to a Nature Energy study.
- The demonstration, conducted with NVIDIA, Oracle, Salt River Project and EPRI, moves demand response for AI facilities from simulations into operations and is being prepared for broader deployments with utilities and ISOs.
- MIT experts say machine learning can improve renewable forecasts, real‑time dispatch and grid planning, yet warn algorithms must respect physical constraints to avoid outages.
- New analysis finds AI’s electricity impact is driven by transmission limits, peak availability and highly concentrated loads, with local congestion shaping prices and emissions more than total consumption.
- In California, opinion writers advocate market‑driven capacity expansion to keep AI investment as community opposition grows and climate goals complicate reliance on gas plants.