Qualcomm targets Nvidia, AMD, Huawei with Dragonfly AI accelerator rack loaded with 43TB of LPDDR5x, future generations set to smash 7PB/s bandwidth (www.techradar.com)

Qualcomm has announced the Dragonfly AI200 AI accelerator rack, marking its entry into the data center market that is currently dominated by Nvidia and AMD. This accelerator, packed with 43TB of LPDDR5X memory, offers a memory bandwidth of 414TB/s per rack, positioning Qualcomm to compete more effectively in the lucrative AI space. The company plans further releases, including the Dragonfly AI250, which will employ its proprietary High Bandwidth Compute (HBC) technology to potentially deliver an astonishing theoretical bandwidth of up to 7.4PB/s. This development is significant as it underscores Qualcomm's strategic pivot towards data centers, aiming to differentiate itself by focusing on low-power memory solutions in contrast to competitors relying on high-bandwidth memory (HBM). While Qualcomm's architectures target inference workloads—ideal for deploying complex AI models—the potential to cater to hyperscalers like Microsoft and Meta indicates a promising trajectory for the company. As data centers increasingly emphasize inference capabilities, Qualcomm's unique approach could carve out a vital niche, challenging existing norms in the AI hardware landscape.