Qualcomm Roadmap Puts HBC In The HBM Debate

StorageReview reported that Qualcomm used its 2026 Investor Day to outline a broader data center push, including the planned acquisition of Modular, the Dragonfly C1000 CPU, an expanded inference accelerator portfolio, and partnerships with Hugging Face and Meta. For RamTrend, the most relevant part is Qualcomm's High Bandwidth Compute architecture. The report describes HBC as a near-memory computing design positioned against HBM, with HBC Gen 1 in the Dragonfly AI250 and HBC Gen 2 in the AI300. The article also says the AI250 design reaches 133 TB/s of effective memory bandwidth per card, compared with the prior AI200 using LPDDR5X, while AI300 is aimed at higher memory bandwidth and capacity for LLM, multimodal, and agentic AI inference workloads. The timing keeps the pricing read-through limited. HBC Gen 1 sampling is expected in mid-2027, AI300 sampling is expected in 2028, and the C1000 CPU is also targeted for 2028. Qualcomm's C1000 platform also includes CXL support, which keeps memory disaggregation part of the roadmap. This is not a near-term DRAM or HBM price signal, but it adds another large silicon vendor to the group trying to reduce AI memory bottlenecks through architecture, packaging, and bandwidth-per-watt gains.