The Foundation behind the Raspberry Pi has announced the launch of a new expansion board that enhances local AI development on the single-board computer.
Raspberry Pi dominate the SBC segment and are used in a multitude of projects, including software development. Here, artificial intelligence continues to gain ground and is one of the sections most pursued by programmers. And the launch of the Raspberry Pi AI HAT+ 2 expansion board is aimed at them.
Single board computers can greatly benefit from a neural processing unit (NPU), as these allow data to be analyzed through neural networks. Possible use cases include pattern recognition in data or even images. The new expansion card aims to increase the AI computing performance of the Raspberry Pi 5, the latest version of these SBCs, up to 40 TOPS (INT4).
The expansion board uses a Known NPU, the Hailo-10Hwhich has its own built-in memory with a capacity of 8 Gbytes. As a result, running AI models ideally has little to no impact on the performance of the SBC itself, which as we know includes limited amounts of RAM despite the release of the new version with 16GB.
Installation of the expansion board is simple and follows the well-known HAT concept. The board mounts to the Raspberry Pi 5 with screws and then connects via the GPIO connector and a PCIe connection. Older Raspberry Pi models do not have a PCIe interface, which means that compatibility is limited to the mentioned version, the series 5. The Raspberry Pi 5 expansion board should be easy to use, something to be expected from an official add-on that is natively recognized by the Raspberry Pi operating system and can be used to run AI applications.
Local AI development through SBC
According to Raspberry Pi, the DeepSeek-R10-Distill, Llama3.2, Qwen2.5-Coder, Qwen2.5-Instruct, and Qwen2 models are available at launch. All (except Llama3.2) are 1.5 billion parameter models, and the foundation claims that there will be support for larger models in future updates.
Its capacity is incomparable with what the AI giants manage, the cloud-based LLMs of OpenAI, Meta or Anthropic and also what other machines can offer such as NVIDIA’s personal AI supercomputers that promise to handle “trillions” of parameters. However, the aforementioned models work well within the limitations of Raspberry hardware and offer another fundamental advantage: They are extremely economical and allow anyone to get started in local AI development.
The Raspberry Pi AI HAT+ 2 expansion board is priced at $130 and is now available at regular SBC distributors.
