Qualcomm unveils Dragonfly data center chip roadmap optimized for agentic AI workloads with high-bandwidth memory integration.
Qualcomm Technologies has unveiled a comprehensive data center roadmap designed for the agentic AI era, announcing a portfolio that spans data-center-class CPUs, AI inference accelerators, high-performance connectivity, and custom silicon. The company's new platform strategy prioritizes performance per watt and token throughput while reducing total cost of ownership.
The Dragonfly C1000 CPU serves as the cornerstone of this strategy—a purpose-built data center processor designed for agentic, general-purpose, and AI head node workloads. Built around custom-designed Qualcomm Oryon cores optimized for frequencies above 5 GHz, the C1000 features a chiplet architecture with more than 250 cores and is designed to deliver more than twice the performance per watt of existing competitive server CPU offerings, according to Qualcomm's estimates. Commercial availability is targeted for 2028.
Complementing the CPU is Qualcomm High Bandwidth Compute, a near-memory computing architecture that bonds compute with accelerated memory bandwidth through 3D-stacked silicon. This technology directly addresses AI's data movement bottleneck. With HBC Gen 1, the Qualcomm AI250 is designed to enable 133 TB/s per card—an 18-fold increase in effective memory bandwidth compared to the AI200 with LPDDR5X. Commercial sampling of HBC Gen 1 is expected in mid-2027. The third-generation Dragonfly AI300 builds on this foundation with HBC Gen 2, designed to deliver a 54-fold bandwidth increase over the AI200. Supporting both air and direct-liquid cooling, the AI300 targets high-throughput, low-latency inference for large language models, large multimodal models, and agentic AI workloads, with commercial sampling expected in 2028.
The roadmap expands to include custom silicon for next-generation AI and cloud infrastructure, developed through end-to-end co-design across silicon, systems, and software. Qualcomm's connectivity portfolio spans die-to-die, copper, optical, and campus-reach interconnects, supporting high-bandwidth 800G and 1.6T connectivity across optical, active optical cable, and active electrical cable applications, with campus-reach deployments extending to 20 kilometers.
Qualcomm has secured a multi-year, multi-generation agreement with Meta to power its next-generation server fleet with the Dragonfly C1000. More than 35 global technology and AI ecosystem leaders are supporting the company's data center vision and commercial solutions. As Qualcomm President and CEO Cristiano Amon noted, agentic AI is driving increased demand for data center inference capacity, playing directly to Qualcomm's strengths in high-performance, low-power computing. Tony Pialis, EVP and GM of Data Center at Qualcomm Technologies, emphasized the importance of orchestrating multiple compute types across distributed infrastructure and addressing key bottlenecks in memory bandwidth and power consumption through the unified, rack-scale Dragonfly platform.