At the keynote speech of 'e4ds Tech Day 2026', Dong-Joo Shin, CEO of Mobilint, stated that AI computation is shifting from GPU-centric to AI SoC and NPU-centric approaches. With the proliferation of generative AI, the proportion of inference is surging rapidly, and he projected that by 2027 it will account for approximately 67% of all AI computation. Additionally, NPUs are optimized for low-precision computation such as INT8 and FP8, providing higher efficiency (TOPS/W) compared to GPUs. He forecasted that future AI functionality will evolve into the form of AI SoCs (APs) that integrate CPUs, GPUs, NPUs, and ISPs, requiring more than 7 billion transistors. Furthermore, he emphasized that AI chips and memory will co-evolve on an HBM basis, and the establishment of a K-AP ecosystem leveraging Korea's memory strengths is crucial.