Near-memory computing architecture bonding compute with accelerated memory bandwidth in a 3D-stacked silicon solution · 133 TB/s per card (HBC Gen 1) · 6× bandwidth per watt versus HBM
First-generation HBC bonded to the Qualcomm AI250 inference accelerator.
Second-generation HBC integrated into the Dragonfly AI300 accelerator platform.
HBC is engineered to beat both HBM and on-die SRAM on the metrics that govern inference economics.
Qualcomm's multi-generation roadmap against HBM, leveraging 3D integration leadership and LPDDR expertise.
Qualcomm Technologies, Inc.
Qualcomm High Bandwidth Compute (HBC)
Near-memory computing / 3D-stacked memory architecture
Compute bonded with accelerated memory bandwidth
133 TB/s per card (designed)
18× vs AI200 with LPDDR5X
54× vs AI200 (designed)
6× vs HBM (card level)
200× vs SRAM (rack level)
Multi-generation roadmap vs HBM
Mid-2027 (with AI250)
Commercial sampling expected 2028 (AI300)
Available
Quote
Qualcomm High Bandwidth Compute (HBC) was announced on June 24, 2026 as the memory pillar of the Qualcomm Dragonfly data center portfolio. HBC is a near-memory computing architecture that bonds compute directly with accelerated memory bandwidth in a 3D-stacked silicon solution, attacking the data-movement bottleneck that dominates modern AI inference cost.
Qualcomm frames HBC as a multi-generation roadmap against HBM. In its first generation, attached to the AI250 inference accelerator, HBC is designed for 133 TB/s per card — an 18× increase in effective memory bandwidth versus the AI200 with LPDDR5X. HBC Gen 2, integrated into the Dragonfly AI300, targets a further stepwise improvement of 54× over AI200.
On efficiency, Qualcomm claims a 6× increase in bandwidth per watt versus HBM (card-level, against published competitor specifications) and a 200× increase in capacity per watt versus SRAM (rack-level). The architecture targets continuous-reasoning agents that need sustained memory bandwidth and real-time responsiveness. HBC Gen 1 sampling with the AI250 is expected mid-2027, with Gen 2 following in the AI300 generation.
Need Qualcomm High Bandwidth Compute?
Contact QS Compute for availability, configuration, and volume pricing.
Request Quote