Ha Jung-woo, Vice Chairman of the National Artificial Intelligence (AI) Strategy Committee, emphasized the need for South Korea to advance as a 'smart export nation' by connecting semiconductors, AI models, and manufacturing capabilities. He proposed supplying high-value services through 'on-premises AIDC,' which encompasses an AI ecosystem that includes power, data centers, manufacturing, and robotics to the global market.
During a discussion on industrial and social policies for sustainable AI transformation held at the National Assembly in Yeouido, Seoul, on October 26, Ha stated, "We must enhance the competitiveness of the entire industry by connecting power, semiconductors, data centers, models, manufacturing sites, and robotics, and move towards becoming a smart export nation that supplies technology and services to the global market."
He suggested that an open full-stack AI strategy is essential for this goal. Ha noted, "Our AI semiconductors should be compatible with various models, allowing companies to choose model infrastructure based on cost, performance, and security requirements. This openness and collaboration should lead to the growth of more domestic companies."
Ha also highlighted the role of on-premises AIDC within companies, as it provides a foundation for safely utilizing internal data while implementing AI tailored to specific operational needs.
He argued that to spread AI across industries, the capabilities of companies and small to medium-sized enterprises must be supported. He particularly stressed the importance of regional AI transformation, stating, "The success of regional AI transformation will determine the success or failure of industrial AI transformation (AX). There is a gap in delivering many strategies planned at the central level to the industrial field." He added, "The strategy committee will focus on creating results that the public and businesses can feel in this regard."
On the same day, Park Ji-yong, an industrial AX PD at the Korea Institute for Advancement of Technology (KAIT), proposed the installation of on-premises AIDC within manufacturing sites as a strategy to grow the domestic AI ecosystem. This plan involves setting up rack-based infrastructure that performs AI learning and inference without external network connections to prevent process data leaks and utilize AI suited for the operational environment.
To achieve this, he outlined a strategy to develop three standard models that combine domestic neural processing units (NPUs) with semiconductors, software, power, and cooling equipment. The small model will range from 10 to 25 kilowatts (kW) for vision inspection and equipment monitoring in small businesses, the medium model will range from 25 to 50 kW for predictive maintenance and process optimization in mid-sized companies, and the large model will range from 50 to 100 kW to support advanced visual inspection, robot control, and production planning optimization in large enterprises.
Park stated, "We will conduct domestic demonstrations starting in 2028 and aim for widespread adoption by 2030. Models that show results in manufacturing sites could also be packaged for export."
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.