New chip module slashes AI data center energy use

Researchers at Jiufengshan Lab in Hubei province have unveiled a new gallium nitride power module that promises to dramatically cut the energy consumption and cost of massive AI computing centers.

The module, developed under the leadership of Li Sichao, was officially released on Dec 4. By replacing traditional silicon chips with chips made from gallium nitride — a superior, third-generation semiconductor — the technology reduces electrical power loss by 30 percent and shrinks module volume by an equivalent amount, all while halving production costs.

The potential savings are staggering. Installing these modules in a single, massive 1-gigawatt AI data center could save nearly 300 million kilowatt-hours of electricity annually, cutting power bills by an estimated 240 million yuan ($34 million). This is critical as the AI boom fuels an unprecedented surge in energy use. A 1 GW data center alone can consume as much power as a large nuclear plant, with power conversion for processors accounting for 11 percent of its total load.