Emerald AI Simulates Power-Flexible Data Centers to Manage Grid Demand

In December 2025, engineers tested a new type of data center designed to be flexible regarding its electricity needs. [1] They recreated the energy demand on the UK grid during a 2020 Euro tournament soccer match between England and Germany. [1] During the simulated demand spike caused by people turning on electric kettles at halftime, an AI program called Conductor sent instructions to a London data center to slow down some power-hungry chips. [1] This reduction helped ensure supply matched demand, which could stave off potential blackouts or hardware damage. [1]

Conductor was developed by Emerald AI, a Washington, DC-based firm exploring whether data centers can operate within the confines of the existing electric grid. [1] This year, Emerald will deploy Conductor in a new facility in Virginia's Data Center Alley, connected to the live grid. [1] When overall demand spikes, the software will reduce the data center's power usage while ensuring servers continue to perform their most important and timely tasks. [1] Emerald is partnering with Nvidia and Digital Realty on the project, billing it as one of the world's first "power-flexible AI factories." [1]

The ability to dynamically adjust power consumption could alleviate a major bottleneck for the AI industry: the mismatch between the time it takes to build data centers versus new power plants. If successful, Emerald AI's approach, backed by industry giants like Nvidia and Digital Realty, could establish a new standard for how AI infrastructure integrates with local grids, shifting data centers from passive consumers to active participants in grid management. What the sources don't address: How much computing performance is sacrificed when Conductor throttles chips during a prolonged power demand spike.