Qualcomm Dragonfly: New Data Center AI Chips Revealed

Qualcomm Technologies has unveiled a new data center roadmap and product portfolio for the agentic AI era. This includes new data center solutions like the Qualcomm Dragonfly C1000 CPU and the Qualcomm Dragonfly AI300 inference accelerator. The company states these new platforms are engineered to maximize performance per watt and token throughput, while also reducing total cost of ownership. The Dragonfly C1000 CPU, expected in 2028, is designed to deliver more than two times better performance per watt compared to existing server CPU offerings. What's interesting is the Qualcomm High Bandwidth Compute technology. This is a near-memory computing architecture designed to address AI’s data movement bottleneck. For instance, the AI250 with HBC Gen 1 is designed to enable a significant increase in effective memory bandwidth. The Dragonfly AI300 is a third-generation rack-level AI inference platform, also expected to sample commercially in 2028. It's designed for high-throughput, low-latency performance in large language and agentic AI inference workloads. The bottom line is Qualcomm is expanding its role in data center infrastructure, aiming to optimize it for the growing demands of AI.