Raspberry Pi launches AI HAT+ 2 to bring generative AI to the edge
The new add-on board pairs a dedicated neural network accelerator with on-board memory to let developers run generative AI models locally on Raspberry Pi 5, reducing reliance on the cloud.
Raspberry Pi has launched the Raspberry Pi AI HAT+ 2, a hardware add-on for Raspberry Pi 5 designed to run advanced artificial intelligence workloads, including generative AI, directly on the device.
The micro-computer maker said the AI HAT+ 2 carries the Hailo-10H neural network accelerator alongside 8GB of dedicated on-board dynamic random-access memory (DRAM). Together, these components provide the compute and memory bandwidth required to run large language models, vision-language models and other generative applications locally, rather than sending data to remote cloud servers.
In practical terms, the neural network accelerator is a specialised processor optimised for AI inference, the process of running trained models to generate outputs such as text, images or classifications. Unlike general-purpose central processing units, accelerators such as the Hailo-10H are designed to execute neural network operations efficiently and at low power, which is critical for edge devices. The inclusion of dedicated DRAM means AI models and intermediate data can be stored and accessed at high speed without competing with the Raspberry Pi’s main system memory, improving performance and predictability.
Building on the original AI HAT+ launched in October 2024, Raspberry Pi said the new board expands the range of workloads that can be run on-device. By keeping AI processing local, developers can reduce latency, improve data privacy and lower operating costs by avoiding ongoing cloud subscriptions.
Related reading
- IBM unveils Sovereign Core to help organisations run AI under local control
- Deepgram outlines blueprint for sub-500ms real-time sentiment analysis on live audio
- Microsoft agrees long-term purchase of soil carbon credits from Indigo Carbon
Eben Upton, chief executive officer of Raspberry Pi, said the product reflects growing demand for edge-based AI. “The Raspberry Pi AI HAT+ 2 extends the capabilities we introduced with the AI HAT+ and enables developers to run a wider range of AI workloads, including generative models, directly on Raspberry Pi 5. It reflects our continued focus on providing accessible, efficient on-device compute for both vision-based and emerging AI applications,” he said. Upton added that strong demand for the first-generation AI HAT+ underpinned the decision to invest in a higher-performance successor, as customers look to build “true edge AI capabilities” with greater privacy and security.
The company said the AI HAT+ 2 retains high performance for computer vision tasks and integrates with Raspberry Pi’s existing camera hardware and software stack. This compatibility is intended to make it easier for existing users to upgrade while also opening opportunities in areas such as security systems, premises management and industrial process control, where local, real-time AI inference is increasingly required.
The Recap
- Raspberry Pi launches AI HAT+ 2 add-on for Raspberry Pi 5.
- Board includes Hailo-10H accelerator and 8GB dedicated DRAM.
- Company provides contact details for further information and enquiries.
