🇰🇷South Korea·AI News·13 Nov 2026·via 서울경제

    DeepX to Invest $100 Million in Next-Generation AI Chip

    DeepX plans to invest over $100 million into developing its DX-M2 AI chip, designed to run large language models at about 5 watts on Samsung Electronics' 2-nanometer manufacturing process, Bloomberg reported. The ultra-low-power AI chip developer aims to secure its first silicon by March 2027 and begin mass production in 2028.

    DeepX's reported investment targets the development of its DX-M2 artificial intelligence chip, with a core technical goal: running large language models (LLMs) and vision language models (VLMs) at approximately 5 watts. This is to be achieved using Samsung Electronics' advanced 2-nanometer process. DeepX has reportedly handed the DX-M2 design to Samsung's foundry business, with initial silicon expected by March 2027, followed by evaluation samples, initial production volumes, and full-scale mass production in 2028.

    The focus on a 5-watt operating target for LLMs suggests an effort to enable advanced AI capabilities directly on devices. Bloomberg characterized this as a significant move by DeepX to process AI models on devices themselves, rather than relying on remote data centers. This on-device approach could reduce energy consumption, operational costs, and the time involved in transmitting data to and from cloud services. Target applications for the DX-M2 include robots, industrial equipment, and personal computers with AI features.

    While the reported power efficiency is a key aspect, the specific size or complexity of the large language models the DX-M2 can support at approximately 5 watts is not detailed. Separately, DeepX also announced a collaboration with Intel Korea and South Korean server maker KTNF. This partnership will focus on on-premise inference servers, with DeepX contributing neural processing units and model optimization, Intel Korea providing CPU and extended processing unit platforms, and KTNF considering server design and integration.

    Share this article

    Go deeper
    Original reporting by 서울경제We don't republish, read the full story →

    Related reading

    6 stories