Targeting memory bandwidth
Qualcomm sees memory bandwidth as a central constraint in AI inference: every generated token requires model data to be fed to compute units. The company is developing a memory design called High Bandwidth Compute (HBC) to overcome this bottleneck.
First silicon validated
The company announced that its first-generation HBC silicon has been validated in its San Diego lab. The chip ran the complete memory read-write path and near-memory computations at production-representative dimensions. Tests used industry-standard automated test equipment with randomized weights and activations.
Data center product in the works
Qualcomm's first data center product, Dragonfly AI250, is said to have completed tape-out. The product is expected to sample in the first half of 2027, marking a significant step in Qualcomm's entry into the data center AI hardware market.
