Qualcomm Dragonwing’s New Edge AI Hardware, Compared

Qualcomm Dragonwing is appearing in a broader range of industrial hardware in 2026. Arduino has introduced a developer-friendly robotics board, Advantech has announced a family for multi-camera machine vision, and AAEON has added a compact SMARC module for embedded systems. Their specifications differ sharply, but all three emphasize local AI processing, connected sensors and deployment outside the data center.

At a glance

• Arduino VENTUNO Q: A dual-processor single-board computer for edge AI, robotics and actuation, with up to 40 dense TOPS, 16GB LPDDR5 and 64GB storage. A $299 preorder price is listed in the supplied product reporting, but the cited coverage says the price was not formally confirmed. 4147
• Advantech AOM-6741 and IQ-9075 family: Industrial SMARC, robotic-controller and fanless-system platforms offering up to 100 dense TOPS or 200 sparse TOPS and support for up to 16 concurrent cameras. Developer samples were announced, but public list pricing was not provided. 913
• AAEON uCOM-Q6490: An 82 × 50 mm SMARC 2.1 module based on Dragonwing QCS6490, with up to 12 dense TOPS, dual Gigabit Ethernet and applications spanning robotics, healthcare equipment, automation and digital signage. No public price was identified. 51720

Arduino VENTUNO Q: AI inference with a real-time control layer

Arduino’s VENTUNO Q is the most accessible of the three designs. It combines Qualcomm’s Dragonwing IQ-8275 application processor with a dedicated STM32H5 microcontroller in what Arduino calls a dual-brain architecture. The Qualcomm processor handles CPU, GPU and NPU workloads, while the STM32H5 is intended for low-latency actuation and motor control. The two processors communicate through an RPC bridge. 3941

The IQ-8275 configuration is rated for up to 40 dense TOPS of NPU performance. Arduino and Qualcomm position the board for computer vision, traditional AI, generative AI, local language and multimodal workloads. It includes 16GB of RAM and 64GB of expandable storage. 3841

Reported hardware features include:

• Qualcomm Dragonwing IQ-8275 octa-core application processor
• STM32H5 Cortex-M33 microcontroller for real-time I/O and actuation
• 16GB LPDDR5 memory
• 64GB eMMC storage
• M.2 NVMe expansion
• 2.5GbE, Wi-Fi 6 and Bluetooth 5.3
• HDMI and USB connectivity
• Linux support on the application processor and Zephyr support for the microcontroller

Arduino says the board will be sold through its store and official resellers. Retailer information cited in the supplied material pointed to Q2 2026 availability. 3739 The reported preorder figure is $299, while available third-party coverage describes a price below $300 and notes that Arduino had not confirmed it. 47

The important design choice is not just the TOPS rating. By assigning deterministic control tasks to a separate microcontroller, the VENTUNO Q is designed to avoid making motor timing and physical I/O depend entirely on the higher-level Linux application processor.

Advantech: IQ-9075 platforms for industrial vision

Advantech’s announcement is a broader platform strategy rather than a single board. The lineup includes the AOM-6741 SMARC module, ASR-A503 and AFE-A503 robotic controllers, and the AIR-055 edge AI system. The products target industrial vision, robotics, automated inspection, automation and intelligent surveillance. 915

The AOM-6741 is a full-size SMARC 2.2 module using Qualcomm’s Dragonwing IQ-9075 octa-core processor. It is specified for up to 100 TOPS of dense INT8 AI performance, or 200 TOPS for sparse workloads. The module can be configured with up to 36GB LPDDR5 memory and 128GB UFS storage. 4

Its connectivity and expansion options are aimed at camera-heavy systems, including:

• Four four-lane MIPI-CSI interfaces
• PCIe 3.0 x4 and x2 interfaces
• Two USB 3.2 Gen 2 ports, plus USB 2.0
• Dual 2.5GbE
• Multiple DisplayPort interfaces through the module connector
• Support for up to 16 concurrent cameras
• Ubuntu 24.04 and Advantech Robotic Suite support

The IQ-9075 platform combines general-purpose CPU processing with AI acceleration, image signal processing and video processing. That combination is suited to workloads such as object detection, tracking and optical character recognition, where camera ingest and inference must operate together. 91113

Advantech said samples of the AOM-6741, robotic controllers and AIR-055 were available. The supplied sources do not establish public list prices for any of them, so buyers should expect channel or quotation-based purchasing rather than a fixed retail price. 169

AAEON uCOM-Q6490: compact compute for embedded products

AAEON’s uCOM-Q6490 takes a lower-power, smaller-module approach. Announced on August 24, 2026, it is described as AAEON’s first Qualcomm-based SMARC module. The module uses Qualcomm’s Dragonwing QCS6490 octa-core SoC and follows the SMARC 2.1 form factor at 82 × 50 mm. Its intended applications include robotics, healthcare equipment and imaging, industrial automation and digital signage. 2021

Qualcomm’s QCS6490 product brief specifies an eight-core CPU running at up to 2.7GHz, together with GPU, DSP and an AI engine capable of up to 12 dense TOPS. 17

The uCOM-Q6490’s published feature set includes:

• Up to 16GB LPDDR5 memory in product listings
• Up to 64GB UFS storage
• Dual Gigabit Ethernet
• PCIe 3.0
• DisplayPort 1.4 output supporting 2K at 60Hz
• USB 3.1 and USB 2.0 interfaces
• UART, SPI, I2C, I2S and GPIO connectivity
• Optional extended-temperature operation from −40°C to +85°C

There is an important specification discrepancy. AAEON and product-listing coverage describe configurations of up to 16GB LPDDR5, while an available AAEON datasheet lists up to 8GB. Integrators should verify the memory option and current revision directly with AAEON before designing around the larger figure. 52126

No public module price was identified in the supplied material. As with many embedded modules, availability appears to be handled through AAEON or distribution channels rather than a standard consumer checkout listing.

What the three launches reveal about Qualcomm’s edge strategy

The products create a clear performance and integration ladder:

1. AAEON’s QCS6490 module targets compact embedded equipment at up to 12 dense TOPS.
2. Arduino’s VENTUNO Q brings up to 40 dense TOPS to a developer-oriented board while adding a separate real-time control processor.
3. Advantech’s IQ-9075 platforms scale to 100 dense or 200 sparse TOPS and up to 16 simultaneous cameras for industrial vision systems.

The common thread is heterogeneous edge computing. These products combine AI acceleration with camera or display interfaces, video and image processing, Ethernet, storage and real-time control. That makes them more directly useful for machines that must interpret local sensor data and respond quickly than a generic compute module would be.

The launches do not by themselves prove that Qualcomm has displaced established industrial platforms, and TOPS figures are not interchangeable across architectures or workloads. They do show Qualcomm working through multiple ecosystem layers: Arduino for developer adoption, Advantech for industrial system integration and AAEON for embedded product design. That is a meaningful expansion of Dragonwing’s role into long-lived, on-device industrial AI deployments.