Third-generation rack-level AI inference platform · integrated Qualcomm High Bandwidth Compute (HBC) Gen 2 · air and direct-liquid cooled · 4–8× better performance per watt versus GPU-based architectures
Third-generation, air- and direct-liquid-cooled rack-level AI inference platform, following AI200 and AI250 (October 2025).
Accelerator card form factor for scale-out inference deployments inside standard racks.
Rack-scale inference system integrating accelerator, memory and networking into one platform.
Qualcomm's annual-cadence inference accelerator line, now in its third generation.
Qualcomm Technologies, Inc.
Qualcomm Dragonfly AI300
Rack-scale AI inference accelerator
Third generation (AI200 → AI250 → AI300)
Qualcomm High Bandwidth Compute (HBC) Gen 2
4–8× better performance per watt vs GPU-based architectures
LLM, LMM, agentic AI, long-context inference
Disaggregated inference
UALink (Ultra Accelerator Link), ESUN (Ethernet for Scale-Up Networking)
Copper and optical networking
Air and direct-liquid cooling
Card and rack
Annual accelerator roadmap
Commercial sampling expected 2028
Available
Quote
The Qualcomm Dragonfly AI300 is the third generation of Qualcomm's data center inference accelerator line, announced on June 24, 2026 at the company's Investor Day alongside the Dragonfly C1000 CPU and Qualcomm High Bandwidth Compute (HBC). It succeeds the AI200 and AI250 launched in October 2025 on an annual cadence.
The AI300 integrates HBC Gen 2 — Qualcomm's 3D-stacked near-memory computing architecture — to pair compute with dramatically increased effective memory bandwidth. Qualcomm positions it for disaggregated inference deployments, targeting 4–8× better performance per watt than existing GPU-based architectures on a memory-bandwidth-per-watt-per-card basis, with industry-leading memory capacity for LLM, LMM and agentic AI serving.
Connectivity is built for modern AI factories: UALink and ESUN for scale-up, plus copper and optical scale-out. The AI300 ships in both card and rack form factors with air and direct-liquid cooling. Commercial sampling is expected in 2028.
Need Qualcomm Dragonfly AI300?
Contact QS Compute for availability, configuration, and volume pricing.
Request Quote