South Korean AI semiconductor developer DeepX has struck a deal to supply chips for Chinese tech giant Baidu, targeting applications in drones and robotics.

The partnership will integrate DeepX’s processors with Baidu’s PaddlePaddle open-source deep learning framework to co-develop AI solutions for global industrial markets.
According to Yonhap News Agency and TheElec, Baidu said the agreement will broaden edge AI applications from optical character recognition (OCR) in documents and images to autonomous aerial and robotic systems.

Performance rivals Nvidia, with lower power draw
DeepX’s DX-M1 AI chip is now fully compatible with the fifth-generation PP-OCR model on PaddlePaddle. Baidu said the jointly developed recognition system outperformed Nvidia’s GeForce RTX 2080 Ti in key benchmarks, delivering superior performance and energy efficiency.
The chip’s low-power, high-performance profile makes it well-suited for battery-constrained industrial AI devices such as inspection robots, equipment monitors, and automated document processors, enabling faster, more accurate, and longer-lasting operations.
Baidu plans to deliver the DX-M1 to around 20 customers, including drone manufacturers, with mass production to follow initial orders. The companies aim to extend their collaboration beyond OCR into a broader range of physical AI applications.

Next-gen chips for Baidu’s ERNIE 4.5
The two firms have also begun proof-of-concept testing of DeepX’s upcoming DX-M2 chip with Baidu’s large vision model (VLM) ERNIE 4.5. Early results show that ERNIE 4.5 maintains high accuracy while operating stably in ultra-low-power environments.
Because DeepX’s semiconductors are designed for edge computing rather than large-scale AI training servers, the products fall outside the scope of US export restrictions on China-bound AI chips.

News Source – https://www.digitimes.com/news/a20250812PD207/deepx-baidu-industrial-chips-ocr.html