Abe Yokell, founder of Emerald AI and former US clean energy diplomat, delivers a TED-style talk arguing that AI data centers — widely seen as a grid liability — could instead become its greatest stabilizing force.
Key insights:
In a May 2025 demonstration at an Oracle data center in Phoenix, Emerald’s Conductor software reduced AI computing power consumption by 25% for three hours during peak grid demand, while maintaining performance above acceptable thresholds.
On average, half of the US power grid’s capacity goes unused — flexible AI data centers could unlock up to 100 gigawatts of that stranded capacity, representing $4 trillion in AI investment without new infrastructure.
Spatiotemporal flexibility allows AI workloads to be paused temporarily or shifted geographically across fiber networks at the speed of light, invisibly to end users.
Data centers are projected to grow from 4% to 12% of US power demand by 2030 — equivalent to adding another Germany to the grid — making grid-cooperative AI a critical near-term priority.