Qualcomm challenges Nvidia's AI grip with chip that ditches HBM - Nikkei Asia

Qualcomm is challenging NVIDIA's dominance in the AI chip market with a new data center chip that dispenses with traditional high-bandwidth memory (HBM) technology. This innovative design features low... Read original →

Qualcomm’s shift away from HBM toward stacked low-power DRAM fundamentally redefines the performance-power-cost trade-off in AI accelerators. This move directly pressures SK Hynix and Samsung’s HBM pricing power and accelerates TSMC’s development of hybrid packaging like CoWoS-L. From a supply chain resilience standpoint, reducing HBM reliance mitigates geopolitical exposure tied to Korea’s concentrated memory production. NVIDIA will likely counter with CUDA ecosystem lock-in, but if Dragonfly achieves superior inference efficiency, hyperscalers may diversify procurement. Over the next 12–24 months, a ‘post-HBM’ niche could emerge, steering sub-3nm designs toward heterogeneous integration rather than raw bandwidth—reshaping AI chip architecture norms.