South Korea’s semiconductor leaders say the country must abandon its traditional “fast-follower” approach to meet the surging demand for artificial intelligence. Speaking at the Herald Business Forum 2026 in Seoul, executives warned that relying on copying established technologies will no longer secure growth. Instead, they called for the development of new memory architectures and stronger computing and energy infrastructure.
Beyond the fast-follower strategy
Herald Business Forum attendees examined the technologies underpinning the industry’s next stage of growth. Jusung Engineering Chairman Hwang Chul-joo said the nation’s past success depended on quickly following knowledge and markets created by others, but he believes that strategy is insufficient now. He argued that companies must establish their own technological direction and define a clear philosophy rather than merely improving existing solutions.
Lim Eui-cheol, vice president of solution AT at SK hynix, noted that AI model computing requirements are increasing about 25 times every two years. In contrast, GPU performance has only improved about three times and memory capacity about 1.6 times during the same period. This disparity makes memory a significant bottleneck in AI systems, highlighting the importance of advanced solutions like high-bandwidth memory, or HBM.
Kim Kyung-ryun, vice president of DRAM product and technology at Samsung Electronics, added that larger AI models and rising user expectations are placing further pressure on memory technology. He stated that existing memory technologies cannot fully meet these demands, necessitating new approaches to memory architecture.
Infrastructure and collaboration
Beyond memory hardware, experts discussed how AI agents could improve productivity by easing coordination across corporate departments. LG AI Research Vice President Lee Hwa-young noted that when these agents enter organizations, they can reduce delays between departments and help teams achieve their goals faster. He also observed that natural-language interaction is changing how people use computers, allowing users to communicate with machines without relying on traditional programming methods.
Sungkyunkwan University Associate Professor Kwon Seok-joon argued that AI competitiveness ultimately rests on both computing resources and energy infrastructure. He stated that neither factor is sufficient on its own and that they must be understood and developed together to operate increasingly demanding AI systems at scale.
