Geniatech is launching pre-sales and planning the first shipments for the AIM-M2 M.2 AI inference module, powered by Kinara's Ara-2 NPU.

The AIM-M2 M.2 module targets edge AI by delivering local inference without cloud dependence, and is capable of running large models on-device.

Supporting materials include a product page, shop link, and a video for additional context.

On-device support enables running models like Stable Diffusion and LLaMA-2, including large models such as LLMs and diffusion models.

Geniatech positions itself as a long-standing embedded and IoT hardware provider with a global footprint, focusing on industrial applications.

Founded in 1997, Geniatech specializes in embedded hardware, AI edge devices, and IoT solutions, with HQ in Shenzhen and global partners.

The AIM-M2 delivers about 40 TOPS INT8 compute while maintaining low power, targeting edge AI for embedded and industrial devices.

In addition to 40 TOPS INT8 performance, the module supports edge AI use cases across embedded systems, industrial devices, and AIoT.

Pricing is set at $168, aiming to democratize on-device generative AI and complex inference for developers, startups, and enterprises.

Target customers include developers, startups, and enterprises seeking accessible, rapidly deployable edge AI acceleration.

The module uses the M.2 2280 form factor (M-Key) for easy plug-and-play integration into host systems.

Designed for seamless integration into embedded platforms, mini-PCs, industrial gateways, and AIoT devices using the M.2 2280 (M-Key) standard.

First shipments are scheduled to begin on September 25, 2025, with pricing and benchmarking details available on Geniatech’s official site.

Additional purchasing information is accessible via Geniatech’s shop portal and official site.

The module supports multi-model concurrent workloads (e.g., object detection, semantic segmentation, NLP) and ships with drivers and an SDK to run models such as Stable Diffusion and LLaMA-2.

With drivers and an SDK, developers can benchmark and rapidly deploy multi-model workloads on-device.

Additional resources include a product page, shop page, and a related video; Emily from Geniatech is a contact for details.