Qualcomm unveils 'High Bandwidth Compute' as HBM alternative, enters AI chip race - 디지털투데이

Qualcomm has unveiled a new memory architecture called 'High Bandwidth Compute' (HBC), positioning itself as an alternative to high-bandwidth memory (HBM) in the AI chip market. The technology uses LP... Read original →

Qualcomm’s HBC architecture isn’t just an HBM alternative—it’s a strategic pivot to redefine AI chip efficiency by stacking LPDDR directly atop compute dies. This move pressures Samsung and SK Hynix to reallocate LPDDR5X/6 capacity and may disrupt Micron’s HBM3E roadmap. If adopted at scale by Meta or Microsoft, it will strain EDA and advanced packaging ecosystems, demanding alternatives to TSMC’s CoWoS. Geopolitically, reliance on foundries in Taiwan, China heightens supply chain vulnerability amid tightening U.S. export controls. Facing NVIDIA’s Grace Hopper dominance, Qualcomm must leverage power-efficient inference for cloud-edge hybrid workloads. Within 18 months, verified bandwidth-per-watt superiority over HBM could catalyze an industry-wide shift from peak FLOPS to sustainable throughput as the core AI chip metric—rewriting memory-compute co-design principles.