Kim Nok-won, CEO of South Korean artificial intelligence (AI) semiconductor startup DeepX, presented a “blueprint” for expanding its presence in the global market with ultra-low-power on-device artificial intelligence (AI) chips.

In an interview with a Dong-A Ilbo reporting team on 8 January (local time) in Las Vegas, United States, where “CES 2026” is being held, Kim said, “Just three years ago, DeepX had around 30 global competitors,” adding, “That number dropped to around 10 last year, and this year only 2 or 3 companies are competing with DeepX.”

Kim described DeepX’s competitiveness using a “butter benchmark.” The company aims to realize generative AI operating inside devices through an ultra-low-power neural processing unit (NPU) that generates so little heat that even butter placed on it would not melt. DeepX has partnered with Hyundai Motor·Kia Robotics Lab to develop an AI chip based on DeepX’s first-generation chip “DX-M1.” The two companies plan to mass-produce delivery and service robots equipped with this chip and expand its use to other sectors in the future.

At CES 2026, DeepX also unveiled its development roadmap for its second-generation AI chip DX-M2 targeting “generative physical AI.” Utilizing Samsung Electronics’ 2-nanometer process, the DX-M2 is an AI chip that runs large language model (LLM) inference at ultra-low power consumption of under 5W (watts). Kim stated, “The design is almost complete, and samples are expected to be available by the end of this year,” adding, “The goal is to begin mass production in mid-2027.”

Popular News