Ryu Jae-cheol, chief executive of LG Electronics (066570), has singled out cooling systems as the key technology that will determine the competitiveness of artificial intelligence data centers. Cutting the power consumed by cooling equipment, he said, frees up additional electricity for AI computing.
In a contributed article published in the U.S. business magazine Forbes on the 19th, Ryu wrote that reducing the energy used for cooling means more than lower operating costs, LG Electronics said on the 20th. Where power supply is constrained, he said, the electricity saved can be directed to graphics processing units and AI computation.
Ryu's focus on cooling reflects the sharp rise in heat generated by data centers as AI chips become more powerful. According to the Uptime Institute, a global data center research organization, the share of operators running high-density server racks that draw 30 kilowatts or more rose to 24% this year from 19% last year. One in seven operators now runs ultra-high-density racks of 50 kW or more.
Ryu also argued that the standards for assessing data center efficiency need to change. The main metric has long been power usage effectiveness, or PUE, which divides a data center's total electricity consumption by the consumption of its information technology equipment such as servers. Going forward, he said, operators must also weigh how flexibly cooling equipment responds to the cooling demands of rapidly evolving AI chips, and how reliably it performs under local conditions including power supply, water resources and climate.
He noted that the ability to connect and optimize an entire facility now matters more than the performance of individual cooling units. Everything from the cooling devices that absorb heat from GPUs to coolant distribution units, piping, pumps, heat exchangers and chillers must be controlled as a single integrated system, he said.
Shortages of power and water are also raising the stakes for cooling technology. The International Energy Agency projects that annual electricity consumption by data centers worldwide will more than double to about 945 terawatt-hours in 2030 from 415 TWh in 2024. Because AI chips are replaced more frequently than data center infrastructure is renewed, Ryu said, cooling systems must be able to adapt flexibly to new generations of chips.
"The competitiveness of next-generation AI infrastructure depends on how much computing performance it can deliver continuously within limited power, water and space," Ryu said.
LG Electronics is building up its liquid-cooling-based thermal management business as a future growth engine as the AI data center market expands. The company recently signed a long-term contract with Air Control Concepts, a North American infrastructure platform company, to supply cooling solutions for AI data centers totaling 5 gigawatts, part of a broader push to expand the business.
