Best Dev Boards for Edge AI Projects

Raspberry Pi 5 and the new Nvidia Jetson Orin Nano Super top our picks for the best dev board for your edge AI projects. But they aren’t your only options. If you’re looking for something smaller, lower cost, something with vision built-in, or even a microcontroller, these are our favorites in 2026.

Raspberry Pi 5

Released in 2023, Raspberry Pi’s flagship single-board computer (SBC) is still capable of crunching the numbers required for many AI applications, including lightweight large language models (LLMs). Add a Raspberry Pi Camera Module or USB webcam for image classification or object detection. You also have the option of adding an official AI HAT+, which offers up to 13 (Hailo-8 variant) or 26 (Hailo-8L) tera operations per second (TOPS), allowing you to process larger machine learning (ML) models faster.

Nvidia Jetson Orin Nano Super

The Jetson Orin Nano Super is an update to the Jetson Orin Nano, and cheaper, too! Boasting an Nvidia Ampere GPU with 1024 CUDA cores and 32 Tensor cores, the Super variant can achieve up to 67 TOPS. Much like the Raspberry Pi, the Jetson Orin Nano Super is an SBC, but installing the required operating system and drivers can be a little trickier than the Pi experience. However, the Jetson Orin Nano Super’s MSRP is $249, around half the price of the original Jetson Orin Nano, making it a strong competitor in the edge AI SBC market.

If you already have an Nvidia Jetson Orin Nano 8GB model, you can simply apply a software patch to upgrade it to the Super version. See more information at Seeed Studio wiki.

OpenMV N6 and AE3

For small, low-power vision processing, OpenMV’s new MicroPython-based offerings are hard to beat.

The flagship N6 is based on the STM32N6 Arm Cortex-M55 microcontroller with the ST Neural-ART accelerator. It can achieve 0.6 TOPS at around 0.75W and is capable of performing full object detection (with the YOLOv8 model) at 30 frames per second (FPS) with a color image resolution of 256×256.

The AE3 is a small, production-ready module built around the Alif Ensemble E3 microcontroller, which is based on a dual-core Arm Cortex-M55 that includes a dual-core Ethos-U55 AI accelerator. Maximum processing power clocks in at around 0.2 TOPS at around 0.25W, and the board is capable of full object detection (YOLOv8) at 13 FPS with 256×256 color
image resolution.

Microcontroller options

If you are OK getting your hands dirty by laying out your own PCB (or don’t mind expensive vendor-built development boards) and navigating low-level vendor libraries, then the following two microcontrollers offer some promising edge AI features.

The powerful new Renesas RA8P1 features an ARM Cortex-M85 core with an Ethos-U55 accelerator, achieving up to around 0.26 TOPS. Renesas’ e² studio is similar to other vendor software experiences: an Eclipse-based development environment with hardware libraries provided through a hardware abstraction layer (HAL).

Alif offers a line of microcontrollers based around the Cortex-M55 core and Ethos-U55 and U85 accelerators. The Ensemble E3 (the same microcontroller used in the OpenMV AE3) offers a particularly enticing sweet spot, and it can reach up to 0.2 TOPS. Alif relies on a software development kit (SDK) that includes HAL libraries, which need to be manually included in your development environment. They also started adding support for Zephyr in 2024, which makes developing professional, cross-platform applications much easier.

This article appeared in Make: Vol. 95.