Qualcomm's proposed solution to catch up in AI infra: Bury the compute under the DRAM

Qualcomm is developing a near-memory compute architecture called High-Bandwidth Compute (HBC), stacking DRAM on XPUs to improve AI infrastructure performance. The technology will debut in 2025 as part of the AI250-series Dragonfly systems, claiming 768 GB memory and 133 TB/s effective bandwidth, though skepticism surrounds its 'effective' bandwidth claims compared to competitors like Nvidia.