"In the AI industry, orchestration is drawing more attention than faster HBM or larger GPUs" - 매일경제

The semiconductor industry's focus in AI is shifting from faster HBM and larger GPUs to system orchestration and overall efficiency optimization. In early AI stages, training was the priority, making ... Read original →

The AI hardware race is pivoting from peak compute to system-level orchestration—a paradigm shift driven by the exhaustion of Moore’s Law. Technically, CXL and optical interconnects will increasingly supplant HBM bandwidth demands, while VRM/IVR integration pushes power delivery closer to die, elevating silicon capacitors and passive components. Geopolitically, tighter U.S.-EU export controls on advanced packaging and optical engines may force Intel and AMD to deepen localized co-development with suppliers in Taiwan, China and South Korea. NVIDIA’s GPU dominance now faces systemic threats: if Grace and Orin fail to tightly unify CPU, memory, and interconnect, AMD’s MI300X and Intel’s Gaudi3 could capture inference workloads through superior data-flow control. Over the next 12–24 months, Agentic AI will prioritize scheduling efficiency over raw FLOPS, rewarding firms with heterogeneous integration capabilities and early-customer co-design pipelines.