KAIST unveils breakthrough that could slash AI data center cooling power

An image shows how a manifold microchannel cooling device directs coolant through microscopic channels inside a semiconductor chip to efficiently remove heat from high-performance processors. Coutesy of Korea Advanced Institute of Science and Technology

By Lee Kyung-min

• Published Jun 16, 2026 1:25 PM KST

As artificial intelligence systems demand ever more computing power, a team of Korean researchers says it has found a way to tackle one of the industry’s most expensive and stubborn problems: heat.

Researchers at the Korea Advanced Institute of Science and Technology said Tuesday they have developed a liquid cooling technology capable of reducing the power needed for semiconductor cooling to roughly one-tenth of current levels while delivering significantly higher efficiency.

The team said the technology integrates microscopic cooling channels, thinner than a human hair, directly into semiconductor chips and combines them with a manifold structure that distributes coolant through multiple pathways.

The design is intended to shorten the distance coolant must travel, reducing energy losses while improving heat removal.

Researchers said previous manifold microchannel cooling systems often suffered from uneven coolant distribution, with some channels receiving more flow than others.

To address that problem, the team optimized the structure so that coolant moves more evenly throughout the system.

After fabricating the design on a silicon wafer, the researchers measured a coefficient of performance of 106,000.

According to the team, that means a single unit of energy used for cooling can remove 106,000 units of heat.

The researchers said the result is more than 10 times higher than the previous global benchmark reported in Nature journal in 2020.

The system operates using room temperature water and does not require boiling-based cooling methods, nanostructured surfaces or expensive materials such as diamond, the researchers said.

The team also said the technology can be integrated into existing semiconductor manufacturing processes without additional production facilities.

Tests applying the design principle to data center cold plates showed cooling performance improvements of more than 30 percent compared with conventional systems.

The findings were published Sunday in the journal Energy Conversion and Management.

This article was published with the assistance of generative AI and edited by The Korea Times.