AI data centers could reduce power draw on demand, study says

Apparently, AI data centers are able to consume less (energy, of course). A recent trial in the UK demonstrated that they can dynamically adjust their energy demands without disrupting critical workloads. This contrasts with the current data center approach of constantly consuming electricity, which can strain networks and drive up prices for everyone.

Over five days in December 2025, more than 200 simulated “network events” tested a London data center’s ability to adjust its energy consumption on the fly. The trial used software from Emerald AI, which participated in the study. Other partners included NVIDIA, National Grid, Nebius and the nonprofit Electric Power Research Institute.

During each simulated network event, the data center successfully adjusted its power consumption to the requested level. It reduced power consumption by up to 40%, while critical workloads continued to run normally throughout the trial.

The data center successfully responded to peaks in demand during football game halftimes. In one case, it reduced its power consumption by 10% for up to 10 hours. It also managed to quickly reduce its demand: during one event, the data center reduced its load by 30% in just 30 seconds.

The study will serve as a model for a 100 MW “flexible power AI factory” that NVIDIA plans to operate in Virginia. “This trial proves that NVIDIA-powered infrastructure can act as a network-enabled asset, modulating demand in real time to support stability,” Josh Paker, NVIDIA’s chief sustainability officer, wrote in a statement. “By making AI workloads responsive, we accelerate deployment while reducing the need for costly network upgrades. »

The organizations involved in the study say they will share their data with the AI ​​industry, regulators and policymakers to try to influence their approach. Fortunately, there is no need to hope that the altruism (ha) of data center operators will lead to their cooperation. Agreeing to limit usage during peak demand could benefit their bottom lines and lead to faster approvals for new data center network connections. “We would like to get to a point where we can attract customers to the grid in two years, and this is part of it,” said Steve Smith, president of National Grid Partners. Bloomberg.