We explore the latest trends in edge AI development boards, from microcontroller (MCU) devices aimed at makers and enterprises to single-board computers (SBC) designed for home users and professionals.
As September arrives, what better way to immerse ourselves in the summer of edge AI than by reviewing the best edge AI hardware and devices of the season?
Due to the wide variety of hardware, we decided to categorize several different types of devices into similar groups. In this comparison, we will adopt the following four categories of hardware:
-
Microcontrollers (MCU): with built-in NPU hardware acceleration
-
Application Processors (CPU): with built-in higher-performance NPU or TPU hardware acceleration
-
Software Acceleration: does not include dedicated AI hardware accelerators (such as NPU or TPU), and its AI capabilities mainly rely on general processors through software solutions.
-
Others: does not include traditional standalone NPU (Neural Processing Unit) or GPU (Graphics Processing Unit), but achieves AI computing capabilities through dedicated architectures.
Now that we have defined the categories, let’s dive into the best edge AI development boards for the summer of 2025! We will start with the most beginner-friendly and popular category, which is microcontroller (MCU) level devices suitable for hobbyists and makers. Then we will move on to enterprise-grade microcontrollers, followed by single-board computers for beginners and home users, and finally, we will introduce professional-grade Linux-based single-board computers (SBC).
Microcontroller Units (MCU):
These development boards are known for being beginner-friendly and cost-effective, and due to their microcontroller-based design, they offer excellent battery life for edge AI scenarios where continuous power supply is not feasible.
This type of board is mainly based on STM32 and ESP S3.
Seeed Studio XIAO ESP32 – S3 Sense
For beginners looking to learn computer vision, the Seeed Studio XIAO ESP32 – S3 Sense is an excellent choice. While it does not have as many sensors as the Arduino Nano 33 BLE Sense, it adds a camera (and microphone) and is powered by a more powerful Xtensa LX7 dual-core 32-bit ESP32 processor, with a maximum clock speed of 240MHz, along with 8MB PSRAM and 8MB flash memory. Best of all, it is priced at just $15! Seeed also provides a wealth of tutorials and example projects to help you get started quickly.
Starry Sky Board K10 AI Programming Development Board
The K10 AI Programming Development Board, developed by the DFRobot team, may be the easiest board for beginners to get started with. The Yushu K10 comes pre-installed with various AI examples that can be run directly, including face detection, cat and dog detection, action classification, and more. This device features a 2.8-inch LCD display for visual interactivity, guiding users as they learn about edge AI. The development board also supports a visual “block” builder and MicroPython, making it easy for novice developers to program the K10. Given its educational attributes, DFRobot also provides a series of excellent learning tutorials that users can follow to start building their own edge AI applications.
OpenMV AE3 and N6 Cameras
These two devices bring dedicated AI acceleration capabilities to the OpenMV ecosystem for the first time and will be compatible with the powerful OpenMV IDE. The AE3 version is powered by the Alif Ensemble E3 microcontroller on a stamp-sized development board, while the N6 version will adopt the classic OpenMV camera size and shape, powered by the STM32N6 microcontroller. Once these two development boards are fully launched, be sure to keep an eye on them, as we expect to see many outstanding projects built with them!
STMicroelectronics STM32N6570 Development Kit
The STM32N6 exploration kit was launched earlier this year and quickly sold out due to high demand and popularity! It is still somewhat difficult to purchase, but the supply chain is catching up, and it is now available again. This exploration kit is feature-rich, with an excellent out-of-the-box experience, featuring a demo application that starts up immediately, showcasing high-speed video inference and smooth video playback on its integrated 5-inch touchscreen, thanks to its integrated Neural-ART AI accelerator. The kit also includes a camera, onboard USB ports, Ethernet interface, a micro SD card slot, and a microphone for easy building, expansion, and integration as needed.
Special Feature: Tria RaSynBoard
Occupying a special position in this category is a truly unique development board that stands out from others on the market. The Tria RaSynBoard is a very small system-on-module and carrier board, featuring a Renesas RA6 microcontroller and a Syntiant NDP120 neural decision processor. It is entirely focused on audio applications, such as wake word detection, keyword recognition, audio classification, or similar use cases, and its compact size allows for easy integration into product designs, such as smart remotes, appliances, and home automation hubs, as well as low-power always-on environmental monitoring solutions.
Application Processors (CPU):
Transitioning from microcontrollers to the Linux world on microprocessors enables developers to build more powerful applications while running multiple services and leveraging stronger processing power and larger memory space. These development boards particularly emphasize ease of use and are an excellent starting point for expanding edge AI capabilities compared to microcontroller options. 
Raspberry Pi 5 with AI HAT+
For developers with little or no experience running AI on single-board computers, the Raspberry Pi 5 combined with the AI HAT+ expansion board is the easiest way to get started! Like all Raspberry Pi products, its focus on documentation, ease of use, example projects, and user support creates a smooth development experience. The AI HAT+ comes in two sizes, one with a 13 TOPS option containing the Hailo-8L accelerator, and another with a 26 TOPS version that integrates the Hailo-8 accelerator. Installation is straightforward, and the AI HAT+ can be mounted on the Raspberry Pi’s GPIO pins like any other HAT, using just a PCIe flexible cable. The built-in camera application in the Raspberry Pi operating system can natively utilize the NPU for post-processing and includes examples for object detection, pose estimation, and image segmentation, with more examples available in the Hailo model library.
NVIDIA Jetson Orin Nano
Developers looking to learn and explore generative AI topics (such as large language models and visual language models) should opt for the NVIDIA Jetson Orin Nano. As a successor to the original Jetson Nano, it is indeed priced higher than its predecessor, but it offers significantly more power with a modern software stack. With the Jetson AI Lab project by Dustin Franklin, it can run large language models, visual language models, Whisper, LlamaSpeak, etc., through simple containerized installations and detailed documentation. The integrated Ampere architecture GPU also supports robotics and autonomous mobile projects, making it an excellent starting point for developers interested in exploring these topics, with CUDA and Deepstream compatibility. Before upgrading to NVIDIA’s large data center-level products, it is a great entry point for learning the basics of GPU programming and the software ecosystem.
BeagleY-AI
There is a passionate and supportive community around the BeagleBoard series, making BeagleY-AI an excellent platform for prototyping, teaching, learning, and exploring robotics and edge AI. This development board adopts the form factor of the Raspberry Pi but is based on Texas Instruments’ AM67A SoC and includes a built-in 4 TOPS AI accelerator. The development board is open-source, allowing its design to be iterated or modified to suit specific use cases.
AMD Kria KV260
This development board has been around for a while but has recently been updated. If you haven’t looked at it in a while, it’s worth a second glance. Some believe that field-programmable gate arrays (FPGAs) should be classified in Group D, alongside professional-grade and production-scale microprocessor development boards, but AMD Xilinx’s entry-level FPGA development board, which uses the Zynq UltraScale+ MPSoC, is relatively low-cost, capable of running Ubuntu on its Arm Cortex core, and is designed to help developers get started with FPGAs, unlike those expensive professional-grade FPGA products. It includes example applications, has excellent documentation, and is a great practical way to learn FPGA programming.
Particle Tachyon
In our professional engineer category, we first introduce a turnkey solution for edge AI projects that require off-grid connectivity—the Particle Tachyon. It is user-friendly and offers an excellent out-of-the-box experience, which means it could have been grouped with consumer-grade development boards, but it truly stands out due to Particle’s seamless cellular network and fleet management capabilities. Tachyon is powered by the Qualcomm Snapdragon 6490 system-on-chip (SoC), an octa-core CPU paired with a Hexagon NPU, enabling 12 TOPS of accelerated AI inference. After a successful Kickstarter campaign, Tachyon devices have just begun shipping, with the team maintaining good communication throughout the process.
NXP i.MX 8M Plus Evaluation Kit (EVK)
The i.MX8MP is used in many professional development boards across the ecosystem, covering almost every imaginable shape and size. However, the best way to familiarize yourself with this SoC and NXP’s software stack before entering production and large-scale deployment is to use the official NXP i.MX 8M Plus evaluation kit. In addition to four Arm Cortex-A cores, it also features a 2.3 TOPS AI accelerator, allowing your edge AI applications to benefit from accelerated inference running directly on the device. Of course, NXP also provides excellent documentation, support, and long-term availability options, enabling developers to confidently bring their products to market.
Jiu Ding X8390
Based on the MediaTek MT8390 (Genio 700) chip.
• Key Advantages: Since the MT8390 and MT8370 (Genio 510) use a PIN to PIN package, this means that the X8390 core board designed based on the MT8390 can be directly compatible with the MT8370 chip. This greatly enhances hardware design flexibility and platform scalability.
Renesas RZ/V2H Evaluation Kit
The Renesas RZ/V2H is one of the most powerful edge AI SoCs on the market (as of July 2025), providing complete 100 TOPS inference capability in a small form factor suitable for edge deployment. To simplify application and product development, Renesas has also just announced that they are collaborating with Canonical to bring Ubuntu support to this development board—allowing for faster prototyping, and if the product development cycle later requires the security and reliability of Yocto, adjustments can be made accordingly. In addition to the official Renesas RZ/V2H evaluation kit, there are various other boards based on the same SoC from different vendors if different form factors or price points are needed.
Software Acceleration:
Arduino Nano 33 BLE Sense Rev2
The Arduino Nano 33 BLE Sense Rev2 may be the gold standard for micro machine learning and opening the door to the world of edge AI. It is affordable, easy to use, and readily available almost anywhere in the world. It comes with a range of onboard sensors, including an inertial measurement unit (IMU) for detecting motion and orientation, gesture sensors, light sensors, proximity sensors, color sensors, temperature sensors, pressure sensors, humidity sensors, and a microphone. These built-in sensors, combined with rich expansion capabilities for other sensors, make it an excellent starting point for edge AI projects. Keep in mind that its limited processing power and memory will exclude most computer vision use cases, but sensor and small audio applications run well on this development board.
Due to the lack of dedicated AI acceleration hardware, the Nano 33 BLE Sense relies on TensorFlow Lite for Microcontrollers (TFLite Micro) and other TinyML tools (such as Edge Impulse, Qeexo AutoML) to run machine learning models. These frameworks are optimized for microcontrollers and support 8-bit integer quantization (int8), significantly reducing computational and storage requirements.
Infineon CY8CKIT – 062S2 – AI
For enterprise microcontroller projects that do not focus on computer vision, the Infineon PSOC 6 AI evaluation kit is hard to beat. It is compact and includes various onboard sensors, but unlike the development boards used by hobbyists, it also includes USB expansion, battery interfaces, and Qwiic expansion. It is also supported by Infineon’s machine learning platform DEEPCRAFT™ Studio (formerly Imagimob Studio) for building custom AI models. This feature alone can significantly accelerate product development, as it has a web-based native platform for model development and can be quickly and easily deployed to the development board.
Nordic Semiconductor Thingy:91x
The latest product in Nordic Semiconductor’s “Thingy” series, Thingy 91:X is an upgrade to the previous generation Thingy:91, and is unique in our microcontroller category as it is the only development board that includes cellular network connectivity. For those edge AI projects that will operate in environments without Wi-Fi or Ethernet, the addition of cellular network connectivity can be a significant factor. The Thingy 91:X integrates an accelerometer and gyroscope for motion sensing, as well as temperature, humidity, air quality, and pressure sensors for climate monitoring. This development board communicates with Nordic Semiconductor’s nRF Cloud service, supporting firmware updates, location services, etc., essentially enabling your IoT products to have AI capabilities.
Others:
Synopsys Astra Machina SL1680 Development Kit
As a new member of the edge AI ecosystem, the Synopsys Astra Machina base series looks like an excellent edge AI solution, with a form factor similar to the NVIDIA Jetson Orin Nano. The SL1680 SoC features a quad-core Cortex-A73 design, integrating an 8 TOPS NPU for AI acceleration. Browsing Synopsys’s product website, you can find high-quality documentation, a comprehensive developer zone with resources, examples, and model libraries, as well as various GitHub repositories to help developers quickly get started with using the device for development. We look forward to trying one of these out in the near future!
This concludes our list of top edge AI products, covering development boards suitable for various price ranges and use cases. But remember, while this list is a starting point, the ultimate “best” edge AI development board is the one that best meets your project’s exact needs… so if you are using other products, that’s great too! In fact, let us know about your experiences, and who knows, it might just appear on our list in the future!
Wishing everyone a pleasant autumn of edge AI development boards!