Qualcomm Launches Dragonfly Data Center Line With C1000 CPU and AI300 Accelerator
Qualcomm is moving beyond mobile and automotive into the data center AI market with a dedicated CPU and inference accelerator designed for agentic AI workloads, aiming to compete on power efficiency rather than raw peak performance.
Reporting from 1 sources: GIGAZINE.
Qualcomm announced the Dragonfly portfolio for data centers on June 25, 2026, including the C1000 CPU with over 250 custom Oryon cores and the third-generation AI300 inference accelerator. The company claims the C1000 offers more than double the performance per watt of competing server CPUs and the AI300 achieves 4 to 8 times better memory bandwidth per watt per card than existing GPU architectures.
The C1000 CPU uses a chiplet design with over 250 custom Oryon cores optimized for frequencies above 5GHz. Qualcomm says it delivers more than double the predicted performance per watt of existing competing server CPU benchmarks. The CPU supports PCIe Gen 7 connectivity exceeding 2TB/s and CXL, and comes in three variants: an agent-type CPU for high-throughput agent control, a general-purpose CPU, and an AI head node CPU for maximizing XPU utilization.
The AI300 is the third-generation inference platform after the AI200 and AI250 announced in 2025. It integrates Qualcomm's second-generation High Bandwidth Compute technology with integrated memory, and the company expects 4 to 8 times better memory bandwidth per watt per card compared to GPU-based architectures. Commercial samples are planned for around 2028.
Qualcomm also introduced the High Bandwidth Compute architecture, a near-memory computing design that aims to address the data movement bottleneck of AI with lower total cost of ownership and higher energy efficiency than high-bandwidth memory.
- Qualcomm Dragonfly C1000 CPU: Data center CPU with over 250 custom Oryon cores, over 5GHz frequencies, PCIe Gen 7, three variants
- Qualcomm Dragonfly AI300 Inference Accelerator: Third-gen inference platform with HBC Gen 2, 4-8x better memory bandwidth per watt per card vs GPUs, samples around 2028
- Qualcomm High Bandwidth Compute (HBC): Near-memory computing architecture for lower TCO and higher energy efficiency than HBM
Synthesized by Yomimono from the 1 cited source below, including Japanese-language reporting where cited, then editorially reviewed before publishing.