Qualcomm announced on November 2 that its first-generation high-bandwidth computing (HBC) chip has completed hardware validation, successfully testing the full memory read-write path and near-memory computing capability, the company said. The technology places compute units close to memory to cut data movement, raising bandwidth and energy efficiency while lowering the power and cost of AI inference.
Qualcomm said a data-center product has been taped out, with initial tests underway on its logic base and stacked DRAM. The Dragonfly AI250 product built on HBC technology is expected to provide samples in the first half of 2027, targeting inference platforms where efficiency is paramount.
The validation marks progress in addressing the cost and energy bottlenecks of large-scale inference as agent applications spread, with the industry metric shifting from peak performance toward effective inference throughput per unit cost.
Source: Qualcomm
