Qualcomm announces 'Qualcomm Dragonfly' products for AI data centers, including the 'C1000' CPU for agent AI and the 'AI300' AI rack.
On June 25, 2026, Qualcomm announced a range of products for data centers, which are described as enabling agent-based AI to run with lower power consumption and lower costs than ever before.
Qualcomm Unveils Comprehensive Data Center Roadmap for the Agentic AI Era with New Qualcomm Dragonfly Portfolio | Qualcomm
The following products and services were announced by Qualcomm:
・Qualcomm Dragonfly C1000 CPU
Qualcomm Dragonfly AI300 Inference Accelerator
・Qualcomm High Bandwidth Compute(HBC)
Improved connectivity, new custom silicon solutions, etc.
・Qualcomm Dragonfly C1000 CPU
The Qualcomm Dragonfly C1000 CPU is a data center-specific CPU designed for agent-based, general-purpose, and AI head node workloads. It features custom-designed Qualcomm Oryon CPU cores optimized for operating frequencies above 5GHz, and its chiplet design with over 250 cores delivers high throughput and high scalability while maintaining exceptional performance per core.
Designed specifically to meet the memory error tolerance and on-site replacement needs of data centers, this low-power memory subsystem utilizes low-power memory technology and is built to deliver superior bandwidth, capacity, latency, and power efficiency, projected to more than double the performance per watt compared to existing server CPU competitor benchmarks. In addition to state-of-the-art PCIe Gen 7 connectivity exceeding 2TB/s, it also supports
Three products will be offered: an agent-based CPU for high-throughput agent control and low-latency conversational AI applications; a general-purpose CPU optimized for performance per total cost of ownership for first-party workloads and performance per vCPU for third-party applications; and an AI head node CPU that maximizes XPU utilization in generative AI computations through low-overhead host processing with a high-speed CPU.
C1000 Custom CPU Power Efficient Server | Dragonfly
https://www.qualcomm.com/data-center/products/qualcomm-dragonfly-c1000
Qualcomm Dragonfly AI300
The AI300 is the third-generation AI inference platform following the AI200 and AI250 announced in 2025. The AI300 integrates Qualcomm HBC Gen 2 technology, which features unified memory for computational acceleration and enhanced effective memory bandwidth, achieving industry-leading memory capacity and effective bandwidth, and providing high-throughput, low-latency performance for large-scale language model and multimodal model inference and agent-based AI workloads.
In terms of memory bandwidth per watt per card, it is expected to achieve 4 to 8 times better power efficiency compared to existing GPU-based architectures. Commercial samples are expected to be available around 2028.
AI300 Rack‑Level Inference Platform for Agentic AI | Dragonfly
https://www.qualcomm.com/data-center/products/qualcomm-dragonfly-ai300
・Qualcomm High Bandwidth Compute(HBC)
Qualcomm High Bandwidth Compute (HBC) is a near-memory computing architecture that delivers faster, more efficient, and scalable processing with lower total cost of ownership and higher energy efficiency compared to high-bandwidth memory (HBM). It is a product that is said to be able to address the fundamental data movement bottleneck in AI.
The first generation of HBC (Gen 1) was designed to deliver industry-leading memory bandwidth of 133 TB/s per card for the AI250, an 18x improvement in effective memory bandwidth compared to the AI200 with LPDDR5X. The AI300, with HBC Gen 2, is designed to deliver an even greater leap forward, a 54x improvement over the AI200.
In addition, it boasts a 6x improvement in bandwidth per watt compared to HBM and a 200-fold improvement in capacity per watt compared to SRAM, enabling efficient scaling of AI agents.
Data Center AI Inference Accelerators | Dragonfly
https://www.qualcomm.com/data-center/expertise/ai-accelerators
To provide these product lines, Qualcomm has partnered with several companies.
Meta, which has signed a multi-year, multi-generational contract with Qualcomm, plans to use the Qualcomm Dragonfly C1000 as the power source for its next-generation server fleet. Production of the product is scheduled to begin in the second half of 2028 to accommodate future data center capacity expansions.
Qualcomm and Meta Announce Strategic Multi-Generation Agreement on Data Center CPUs | Qualcomm
https://www.qualcomm.com/news/releases/2026/06/qualcomm-and-meta-announce-strategic-multi-generation-agreement-
Hugging Face is expanding its existing partnership with Qualcomm to seamlessly connect Qualcomm's data center infrastructure with Hugging Face's AI storage infrastructure, aiming to deliver a more immersive AI experience. This integration will allow models deployed on Hugging Face to run through Qualcomm's products.
Qualcomm and Hugging Face Expand Relationship to Advance Open, Developer-Driven AI from Device to Cloud | Qualcomm
https://www.qualcomm.com/news/releases/2026/06/qualcomm-and-hugging-face-expand-relationship-to-advance-open--d
Qualcomm CEO Cristiano Amon said, 'Agent-based AI is significantly increasing the demand for AI inference in data centers. As these become the main workloads, infrastructure needs to deliver much higher performance with lower power and cost. With Qualcomm Dragonfly, we are bringing high-performance, low-power computing to data centers and securing multi-year, multi-generational contracts with key customers.'
Related Posts:
