A single USB can now add AI inference functions to existing devices. Mobilint said on Aug. 7 it has launched the MLD-R1 as a commercial product. The MLD-R1 was first unveiled in reference form at COMPUTEX 2026 in June and has now been completed as a commercial version that customers can deploy in the field.

The MLD-R1 is the first commercial AI accelerator in which Mobilint implemented REGULUS, a standalone AI computing chip it developed this year, in a USB form factor. By simply connecting it to laptops, industrial PCs and embedded systems that support USB 3.1 Gen1 Type-C, users can add AI inference functions without installing an expansion card or changing the system. It supports Windows and Linux operating systems, and both x86- and ARM-based hosts.

The product delivers AI computing performance of up to about 10 TOPS (INT8) at about 3 watts of low power, and includes 4 GB or 8 GB of LPDDR4X memory. It can handle computer vision and video analysis as well as generative AI models, including small language models with 1 billion to 3 billion parameters. The company said support for major AI development frameworks such as PyTorch, ONNX, TensorFlow and Hugging Face lowers barriers for developers.

Mobilint sees the MLD-R1 as a platform that will reduce the cost of adopting on-device AI and shorten development timelines in areas including smart factories, machine vision, industrial automation, robotics, security, medical devices and digital signage.

Kim Seong-mo (김성모), head of the business development division, said, "MLD-R1 is a product that confirmed high market expectations for commercialization, as pre-orders and inquiries about adoption continued not only from domestic AI national projects but also from customers at home and abroad immediately after its unveiling at COMPUTEX 2026." He added, "We will continue to expand our product portfolio from AI semiconductors to USB AI accelerators, AI cards and AI Boxes to build an ecosystem where anyone can use on-device AI easily and efficiently."