AI Efficiency Race Opens New On-Device and NPU Opportunities for Korean Firms

At Korea Investment Week (KIW) 2026, held on May 15 at The Shilla Seoul in Jangchung-dong, Korean AI companies preparing for IPOs said the center of AI competition is shifting from computing scale to efficiency and from data centers to everyday devices. Point2 Technology said interconnect bottlenecks linking GPUs can reduce computing performance by about 30%, and presented e-Tube, which offers up to 10 times the transmission distance of copper wires and uses one-third the power of optical communications. DeepX highlighted ultra-low-power chips for on-device AI, while Mobilint emphasized NPUs that run high-performance inference at around 20-25W. Elice Group proposed modular data centers that can be built within four months. Source: cited from hankyung.com.