CNBC examines data centers' water and electricity demands and introduces DeepX's low-heat technology
Interview with CEO Kim Nok-won and butter experiment video compare semiconductor heat
As the construction of artificial intelligence (AI) data centers became a community issue during the US midterm elections, a South Korean low-heat AI semiconductor has drawn attention after appearing on US television as a potential technological alternative for reducing electricity and cooling-water demands.
US business broadcaster CNBC highlighted South Korean AI semiconductor company DeepX's ultra-low-power, low-heat technology in an in-depth article and video about data center cooling on November 5, local time.
CNBC examined resource consumption associated with data center expansion, reporting that a series of ballot measures restricting AI data center construction and operation had passed in the US elections.
The broadcaster said cooling equipment is essential for removing heat generated by semiconductors performing AI computations. It identified reducing both water and electricity demands as a major challenge, noting that cooling methods using less water can consume more electricity.
It also reported on the current state and limitations of cooling technology through examples involving AI servers from US semiconductor companies Nvidia and AMD, data center operator Equinix, and liquid-cooling specialist CoolIT.
Alongside improvements to cooling equipment, CNBC introduced an approach that reduces heat generated by the semiconductors themselves, presenting South Korean AI chipmaker DeepX as an example.
The broadcast included an interview with DeepX CEO Kim Nok-won in San Francisco, along with footage of an experiment comparing heat by placing butter on semiconductors.
Although chip heat comparisons alone cannot establish water and electricity savings across an entire data center, the footage has been described as a visual demonstration of DeepX's low-heat characteristics because the butter on its AI semiconductor did not melt.
The broadcaster reported: "South Korean startup DeepX has developed an AI processor that operates at temperatures tens of degrees lower than competing products."
DeepX has been mass-producing its DX-M1 ultra-low-power AI semiconductor, based on Samsung Electronics' 5-nanometer process, since August 2025.
According to DeepX, as of October 14, it had also secured 88 commercial purchase orders worth about $13.3 million across more than 10 countries.
The company is also developing its next-generation DX-M2 AI semiconductor using Samsung Electronics' 2-nanometer process. It is pursuing mass production in 2027, aiming to run generative AI, including large language models (LLMs), on approximately 5W of power.
Original Korean article: Yonhap News
English brief by Korea Wind Intelligence, koreawindintel.com
