HBM Shortage Makes Frontier AI a Luxury Good by 2030, RIKEN Study Finds - techtimes.com

A recent study by RIKEN highlights that the shortage of high-bandwidth memory (HBM) is reshaping the competitive landscape of the artificial intelligence industry, potentially making frontier AI a lux... Read original →

The HBM bottleneck is triggering a structural reshaping of AI infrastructure. Technologically, while 3nm nodes and EUV boost logic performance, yield constraints in silicon interposers and TSVs make HBM the critical bandwidth chokepoint, accelerating GPU architectures toward chiplets and near-memory computing. On compliance, U.S.-Japan-South Korea export controls on advanced packaging tools heighten supply chain fragility, exposing Taiwan, China and Hong Kong, China assembly hubs to geopolitical scrutiny and inflating inventory costs. Strategically, SK Hynix and Samsung lock in Meta and xAI via long-term HBM3E deals, while Micron pushes CoWoS alternatives; Chinese firms like Zhipu AI adopt Mixture-of-Experts to reduce memory dependency—a defensive downgrade. Within 18 months, AI will stratify: proprietary models retain premium pricing for agentic tasks, while open-weight models dominate low-end inference but hit performance ceilings. Memory bandwidth has become the hidden gatekeeper of Moore’s Law.