SK Hynix looks beyond HBM as inference workloads diversify

SK Hynix's strategic pivot signals a shift in AI inference workloads from GPU-centric HBM reliance to diversified, latency-sensitive and capacity-intensive applications. This evolution will reshape the DRAM and HBM supply chain, compelling upstream foundries to accelerate advanced process development and downstream server vendors to adapt to new architectures. Geopolitical tensions, particularly U.S. export controls on China, are heightening supply chain risks, pushing companies toward regionalization. Competitors like Micron and Samsung are accelerating inference-specific memory products to capture emerging market share. Over the next 12–24 months, memory bandwidth and energy efficiency will define competitive advantage, with traditional HBM demand declining while high-capacity memory for long-context models emerges as a growth driver.