DEEPX, a South Korea-based AI semiconductor firm specializing in low-power on-device artificial intelligence, has entered a strategic partnership with Chinese technology giant Baidu to advance global industrial AI adoption in sectors including drones, robotics, and optical character recognition (OCR).
The companies signed a “PaddlePaddle Technology Ecosystem Partnership” agreement in Shenzhen, China, on August 18. Under the deal, DEEPX becomes an official ecosystem partner of PaddlePaddle, Baidu’s open-source deep learning framework, and will collaborate on industrial AI projects, co-develop products, and engage in global market outreach.
PaddlePaddle—short for “PArallel Distributed Deep LEarning”—is China’s first open-source deep learning framework and is widely used across industries such as smart cities, autonomous driving, image recognition, and speech processing. Positioned as a full-stack AI platform, it offers pre-trained models, model compression tools, and visualization capabilities, and is adopted by thousands of companies and institutions in China.
Prior to the agreement, DEEPX demonstrated its DX-M1 chip running Baidu’s fifth-generation PP-OCR and vision-language models (VLM) in real time, achieving high frame rates and strong energy efficiency on edge devices with limited battery resources. Baidu said it plans to expand collaboration beyond OCR into drone-based AI models and VLM-driven obstacle avoidance for robotics, with deployment targeted for real-world commercial projects.
The partnership will also see DEEPX compile and integrate 10 widely used OpenVINO-based AI models for compatibility with the DX-M1, making them available to Baidu’s PaddlePaddle ecosystem partners worldwide. The aim is to ease adoption of DEEPX’s neural processing unit (NPU) technology for industrial customers.
DEEPX is concurrently developing the V-NPU, a vision AI-dedicated processor card designed to serve as a GPU alternative, with mass production expected in September. Discussions are underway for joint work on DEEPX’s upcoming DX-M2 chip, built on Samsung Foundry’s 2nm process, to run Baidu’s large-scale model ERNIE-4.5-VL-28B-A3B.
The collaboration extends to joint marketing and public demonstrations. Baidu and DEEPX plan to co-exhibit at the 2025 Shenzhen International Artificial General Intelligence Conference in August, showcasing live demonstrations of AI applications to global customers.
Baidu’s PaddlePaddle and ERNIE teams said in a joint statement that DEEPX’s low-power AI chips will add “powerful edge computing capabilities” to the ecosystem, enhancing the real-world applicability of PaddlePaddle-based AI in OCR, robotics, and drones. The companies expect the partnership to accelerate the integration of AI into diverse industrial environments.
“This partnership with Baidu marks an important turning point in the rapid spread of DEEPX’s AI semiconductor technology to global industrial sites,” said DEEPX CEO Lokwon Kim. “We will strengthen our competitiveness in the global AI market through practical productization based on the PaddlePaddle and ERNIE ecosystem.”
Founded with the vision of making AI as pervasive as electricity and Wi-Fi, DEEPX holds more than 350 pending patents in the U.S., China, and Korea, making it one of the largest on-device AI chip patent portfolios globally. The company works with over 120 organizations worldwide in fields including smart cameras, surveillance, smart factories, robotic platforms, and AI servers, with ongoing collaborations with Hyundai Kia Motors Robotics Lab, POSCO DX, LGU+, and Inventec.
