Raspberry Pi launched a $130 AI HAT+ 2 featuring a Hailo 10H chip and 8GB of dedicated RAM. This hardware aims to offload local LLM processing from the main CPU, boasting 40 TOPS of inference performance. It promises a compact, power-efficient alternative to bulky eGPUs for edge AI tasks, though the RAM upgrade is limited to the add-on board itself.

Despite the dedicated hardware, real-world benchmarks show the Pi's own CPU often outperforms the Hailo 10H chip. The Hailo's strict 3W power limit pales in comparison to the Pi SoC's 10W allowance, making the onboard processor faster for many inference tasks. The 8GB RAM ceiling also restricts running larger, more capable models, forcing users to compress files significantly.

Where the HAT shines is vision processing, running 10x faster than the CPU for object detection, though the previous AI HAT handles this just as well. The headline 'mixed mode'—running vision and LLMs simultaneously—failed in testing due to software errors. Ultimately, the board feels like a development kit for niche industrial hardware rather than a must-have upgrade for hobbyists.