Raspberry Pi AI HAT+ 2 — Hailo-10H 40 TOPS Generative AI Accelerator

Specifications

Overview

The Raspberry Pi AI HAT+ 2 is the second-generation AI accelerator HAT for Raspberry Pi 5, built around the Hailo-10H neural accelerator with 8GB of dedicated on-board LPDDR4X memory. Where the first-generation AI HAT+ (Hailo-8/8L) relied on the host Raspberry Pi 5's memory and was limited to computer vision, the AI HAT+ 2 carries its own RAM and can therefore run large language models and vision-language models of up to roughly 6 billion parameters entirely locally — leaving the Raspberry Pi 5 free for application and I/O duties. It conforms to the Raspberry Pi HAT+ specification and is supplied with a heatsink and stacking hardware so it can be fitted alongside the official Raspberry Pi 5 Active Cooler.

Key Benefits

Generative AI on a $200 board: 40 TOPS of INT4 inferencing and 8GB of dedicated memory bring local LLM, VLM, speech-to-text and translation workloads to Raspberry Pi 5 without any cloud subscription. Keeps the host free: because the accelerator owns its memory, model weights do not compete with Raspberry Pi 5 system RAM — useful for camera pipelines, Frigate-style security systems and robots that must both record and reason. Drop-in integration: HAT+ compliant, PCIe-attached, fully integrated into the Raspberry Pi camera software stack, with Hailo's model zoo and a GitHub repository of ready generative-AI models plus LoRA fine-tuning support. Fleet-scale edge deployment: identical software across many units makes it straightforward to roll out private, offline inference to distributed sites.

Applications

Offline LLM chat and document Q&A, vision-language scene understanding, security-camera analytics (Frigate and similar), robotics perception, smart retail and kiosk assistants, speech-to-text and on-device translation, computer-vision education and prototyping.

Related Products