Big Tech Tests Data Center Flexibility for Local Power Grids

Three data center centers in France and the United States will be equipped with experimental methods to make them more flexible and adapted to their local energy networks. IEEE SICTRUM He learned.

The selected centers will work as a test of a test for solutions for the increasing electricity requirements of artificial intelligence, which is expected to double in the next five years to reach about 3 percent of the total global electricity consumption. Expectations raised concerns about meeting the adequate demand for the data center on electricity without sacrificing reliability to anyone else, and prompted a global effort to innovate about the problem.

The experimental centers stem from pioneering cooperation, called the DCFLEX initiative, which consists of network operators, facilities, and large technology companies including Google, Oracle, NVIDIA, Meta and Microsoft. The group, leads itElectrical Energy Research Institute (EPRI) in Palu Alto, California, announced the first three sites today, and expects to announce up to seven other sites this year.

“This is like the first group,” says Anuja Ratnayaki, the executive of emerging technologies at EPRI. “These three are a happy marriage between what was available from our members to participate and some cases of use that need the longest test periods.”

Work cosmetics from artificial intelligence

You will test the specific centers, which are operational, network centers connected to the network, each different aspect of flexibility. In Lenoir, NC, Google will work directly with the local utility, Duke Energy, to determine and transform the computing center for the data center to meet the network needs.

This approach, which is called dances in the burden of work, includes a center for unloading computing tasks into other facilities or pushing those tasks to a different time frame to reduce the total load on the network at a certain time. Ratnayaki says that the strategy tends to be viable only for Easter customers such as Google who use data centers to train artificial intelligence and work on a schedule.

Contrary to artificial intelligence training, cloud and cloud services that provide broadcasting and banking services online cannot transform their loads through time and geography because these services meet the actual time of a person on the other side of the screen, says Ratnayaki. “If banking services or credit card services are, then there is little flexibility, because this is something that must process transactions with a very high number in a very short window,” she says.

This is where the other two sites come. In the second location, the burdens of a data center work in Phoenix Provide a mixture of artificial intelligence and cloud services from Oracle and NVIDIA. The third -party solution provider, Emerald AI, will coordinate dance design with local facilities, including the Salt River project.

Earlier this year, Phoenix simulated a peak event, where the energy demand on the network is high, to test whether the data centers can respond by reducing its work burden. This test was successful, as the site achieved flexibility between 10 and 40 percent in the work of the designed work, says Ratnayaki. Next, the group will test if the site can respond to the Real Peak Energy event.

UPS Energy Stability Systems

The third site, located in Paris, focuses on maintaining stability for data centers during energy disorders. Most large data centers are equipped with unexpected power supply (UPS), a backup system that keeps things working when there is an energy disable from the network. Data4 Center DATA4 will work with Schneider Electric and RTE, the transport system operator in France, to explore how to use the UPS system to operate it through voltage and frequency problems. Currently, voltage drops and other network problems can operate data centers in a non -connection mode to protect their computers from damage.

There may also be opportunities to use UPS energy storage to provide data centers with additional energy flexibility. This is something that DCFLEX will explore in future sites, as well as cooling the Data Center and Low Carbon’s Curbon for Backups.

The group plans to experience up to ten sites this year. The sites will spread throughout the United States and Europe, with one or two, perhaps in the Middle East or Asia. To date, DCFLEX 45 collected collaborators, a height of only 14 years when it was launched in October 2024. By conducting these field tests, the group aims to create a common framework for flexibility that supports AI’s pregnancy growth.

EPRI expects the results of the first three show sites later this year.