DeepX Connects Millions of Developers in an 'Open Ecosystem'... Aiming for Standardization of Physical AI Execution

Partnerships with YOLO, PaddlePaddle, and Raspberry Pi
Open Execution Environment from Development to Industrial-Scale Production

DeepX, a company specializing in ultra-low power on-device artificial intelligence (AI) semiconductors, is accelerating the creation of a developer-oriented 'Open Physical AI Ecosystem' by increasing collaborations with global AI model and developer platforms.

DeepX plans to integrate worldwide software and hardware ecosystems such as Ultralytics YOLO, PaddlePaddle, and Raspberry Pi with its neural processing unit (NPU). Through this, the company aims to provide a unified execution environment that covers everything from idea development and proof of concept (PoC), to mass production for industrial use.

On July 14, Nohkwon Kim, CEO of DeepX, stated, "In the era of Physical AI, having just a good chip is no longer enough. There needs to be a full ecosystem in which developers can create models, validate them on hardware, and deploy them to real industrial sites." He continued, "Beyond simply supplying chips, DeepX will accelerate the construction of an Open Physical AI Ecosystem that connects developers, AI models, and hardware, aiming to build an open execution platform that enables global developers and companies to realize Physical AI more easily and more quickly."

DeepX Open Physical AI Ecosystem Conceptual Diagram. DeepX

Physical AI refers to technologies that enable offline devices—such as robots, intelligent cameras, and smart factory equipment—to directly perceive, decide, and control data in the real world. By linking general-purpose AI model ecosystems with its ultra-low power NPU, DeepX helps developers immediately deploy AI algorithms created in traditional research environments to actual industrial devices.

The company has also unveiled concrete plans to connect its NPU with the world's 'three major AI developer ecosystems.' One notable initiative is the collaboration with Ultralytics YOLO, which is widely used in the real-time vision AI field. 'YOLOv8,' relevant for smart cameras, robotics, and logistics, attracted 5 million users within a year of launch. In May, DeepX formed a strategic partnership with Ultralytics to enable developers to implement low-power, high-performance AI inference on platforms such as Raspberry Pi and robots using the familiar YOLO model. Unlike previous approaches, models can now be directly extended to edge systems after training, without the need for time-consuming optimization steps.

In addition, the connection with the open-source deep learning framework 'PaddlePaddle,' which has over 4 million developers, has been strengthened. Since signing an agreement in August last year, DeepX has been jointly developing products with the platform. As a result, it has equipped its M.2 form factor AI accelerator 'DX-M1' with PaddlePaddle's lightweight AI model 'PP-OCR 5th generation,' ensuring smooth operation on DeepX NPUs. This solution can be deployed immediately in industries requiring real-time inference, such as drones, smart cities, and robotics.

From a hardware perspective, DeepX is leveraging the Raspberry Pi ecosystem—a small-sized computer with cumulative sales exceeding 73 million units. In June, the company released an AI acceleration processor module for the fifth generation of Raspberry Pi and began supplying it to the global market. Developers can test functions on Raspberry Pi-based YOLO or PaddlePaddle models, then scale them up as final products for industrial equipment or smart factory systems. In this structure, Raspberry Pi serves as an experimentation environment while DeepX NPU acts as the ultra-low power engine for field devices.

Through this strategy, DeepX plans to establish a virtuous cycle in which developers' initial ideas evolve through proof of concept into final mass-production products. To this end, it will organically combine AI models, hardware, NPUs, and software development kits (SDKs), while continuously expanding enterprise-level field validation programs and hands-on examples to lower market entry barriers.

This content was produced with the assistance of AI translation services.