Qualcomm has officially outlined its long-term data center strategy with the new Dragonfly platform family, targeting the growing demand for AI inference and agentic AI workloads.
The company introduced several new products, including the Dragonfly C1000 server CPU, High Bandwidth Compute (HBC) memory technology, the Dragonfly AI300 inference accelerator, and new connectivity solutions aimed at hyperscale AI infrastructure, which joins the previously announced AI200 and AI250 accelerators in Qualcomm’s roadmap. Qualcomm says the entire platform is built around performance-per-watt efficiency and lower operating costs, something becoming increasingly important as AI inference workloads continue to grow.
“Agentic AI is driving a significant increase in demand for AI inference in the data center. As these become the dominant workloads, infrastructure has to deliver much higher performance at lower power and cost,” said Cristiano Amon, President and CEO, Qualcomm Incorporated. “That plays directly to Qualcomm’s strengths, and we’re well positioned for this shift. With Qualcomm Dragonfly, we’re bringing our high-performance, low-power computing into the data center, with multi-year, multi-generation agreements with leading customers.”
The Dragonfly C1000 is Qualcomm’s first major server CPU push in years. It uses custom Oryon CPU cores running at over 5 GHz and features a large chiplet-based design with more than 250 CPU cores. Qualcomm claims the chip is optimized for agentic AI workloads, AI orchestration, and cloud-scale deployments while offering more than twice the performance-per-watt of current competing server CPUs based on internal estimates.
The processor will also support PCIe Gen 7, CXL connectivity, and advanced low-power memory technologies. Qualcomm says the C1000 can be deployed in both air-cooled and liquid-cooled servers.
One of the more interesting announcements is Qualcomm’s new High Bandwidth Compute technology, or HBC. Instead of relying entirely on traditional HBM memory, Qualcomm is using a near-memory compute design with stacked silicon and LPDDR-based memory systems.
According to Qualcomm, HBC Gen 1 used in the upcoming AI250 accelerator can deliver up to 133 TB/s of effective memory bandwidth per card, while HBC Gen 2 planned for AI300 should significantly increase bandwidth again. The company claims much better bandwidth-per-watt efficiency compared to current GPU platforms.
Qualcomm also confirmed its next-generation Dragonfly AI300 inference accelerator, which follows the previously announced AI200 and AI250 products. The AI300 platform is designed for large language models, multimodal AI workloads, and agentic AI systems, with both air-cooled and liquid-cooled rack-scale configurations planned. The company says its AI roadmap will be updated on an annual basis, which is a great way to attract customers and investors.
Qualcomm additionally revealed a multi-year agreement with Meta, which plans to use the upcoming Dragonfly C1000 CPU in future server deployments. More than 35 ecosystem partners, including Lenovo, Supermicro, Samsung SDS, Quanta, Foxconn, and Gigabyte, are also backing Qualcomm’s expanding AI data center plans.
Commercial availability for the Dragonfly C1000 and AI300 platforms is expected around 2028, while HBC Gen 1 sampling with AI250 is planned for mid-2027.
