A groundbreaking trial of flexible electricity usage was recently conducted by National Grid and Nebius at a UK datacentre near London. The trial utilized AI-powered datacentre management software from Emerald AI on a bank of 96 Nvidia Blackwell Ultra GPUs.
During a five-day period in December 2025, over 200 real-time simulated “grid events” were sent to the datacentre to assess the software’s capability to adjust power consumption dynamically. The Emerald AI platform successfully reduced demand by up to 40% while maintaining critical workloads.
The trial showcased the software’s ability to respond to spikes in demand during events like football matches, reduce load during periods of low wind or extreme heat, and swiftly shed load to support grid resilience during stress events. It is estimated that AI datacentres could potentially add over 2GW of capacity back to the grid when required.
By enabling AI datacentres to manage peaks, optimize existing infrastructure, and support the integration of various energy sources into the grid, the partnership aims to shift these datacentres from being a constraint to a controllable asset for the electricity grid.
Steve Smith, president of National Grid Partners, emphasized the importance of leveraging existing capacity to avoid the need for extensive infrastructure upgrades. Varun Sivaram, CEO of Emerald AI, highlighted the flexibility of power consumption achieved through the trial, showcasing the ability to reduce energy consumption rapidly when needed.
Sivaram outlined three methods for achieving flexibility in power consumption for AI workloads, including adjusting workload priorities, moving workloads between datacentres, and intelligently monitoring and managing datacentre operations.
