Study suggests data centers can ease power shortages by adjusting energy use, helping meet AI demand.
In the face of increasing worries about AI-fueled electricity demand, a new study finds that data centers and other high-energy consumers may be able to substantially ease power shortages by making slight changes to their energy use. The study indicates that by cutting back electricity consumption to 90% of peak capacity for a few hours annually, data centers can realize 76 gigawatts of power, higher than the entire world's data centers consume. This possible surplus accounts for around 10% of peak U.S. electricity demand.
Traditionally, industrial consumers such as factories and shopping malls have enrolled in demand-response programs, reducing consumption during high-demand hours for credits. But data centers have avoided such programs largely to preserve uptime for customers. The report contends that data centers are sufficiently flexible to adopt demand response.
Some options would be employed to lower demand without sacrificing service. For example, data centers might move computer work to low-demand times, like reordering AI model training. Spatial flexibility also enables operators to relocate tasks to lower-demand areas. Data centers might aggregate workloads and power down some of their server operations. To support mission-critical applications, batteries might fill in for brief curtailments.
Tech firms have already started looking into such solutions. Google's carbon-aware computing platform and Enel X's activity with uninterruptible power supplies (UPS) are just two instances of how data centers are participating in demand response. PG&E is also providing incentives to bring data centers onto the grid more rapidly if they sign up for these programs.
Though these changes will not completely put an end to the impending crisis of power supply, they are a promising measure to counter shortfalls due to the increasing usage by AI.
.webp)
.webp)
