Posco DX develops unstructured data analysis platform powered by domestic AI chips

Posco DX has developed an unstructured data analysis platform built on domestic AI semiconductors, and plans to use it to advance AI localization across industrial sites.

The company said Tuesday it developed the platform using a neural processing unit, or NPU — a domestic AI semiconductor designed specifically for AI computations such as deep learning and machine learning. Unlike a graphics processing unit, or GPU, an NPU is optimized for AI inference, reducing infrastructure costs and power consumption.

Posco DX also highlighted the NPU's suitability for direct installation in on-site equipment control systems. Because the chip does not require a connection to a remote AI data center or server, it enables edge AI — allowing real-time inference at manufacturing sites without transmitting data externally, a key advantage where data security is paramount.

The company said it has incorporated these technical strengths into its Vision AI platform. The newly developed platform manages diverse video data collected from industrial sites as training data within a single environment. It also provides standardized functions needed to deliver vision AI services, including AI model management, performance metrics and deployment history.

Posco DX said it expects the platform to shorten execution time for new projects and expand its services across multiple projects and industrial sites.

Posco DX said the platform's most distinctive feature is its hybrid architecture, which uses both GPUs and NPUs simultaneously. GPUs handle AI model training and development, while domestic NPUs manage real-time computation and decision-making on the factory floor. The company also built in an environment that allows NPU testing during the research and verification stage, letting engineers assess NPU-based inference performance and effectiveness before deploying the technology at actual industrial sites.

Posco DX has been pursuing AI localization across various fields through partnerships with domestic NPU developers including Deepx and Mobilint. Key applications include fire monitoring at industrial sites, worker safety monitoring and cargo load verification in logistics environments. The company said those efforts have confirmed that infrastructure construction and operating costs can be cut by about 50 percent compared with GPU-based systems delivering equivalent inference performance, while power consumption falls by roughly 90 percent.

Posco DX plans to migrate sites already running vision AI technology to its domestic NPU-based platform and expand the scope of deployment.

"The NPU-based vision AI platform will serve as a foundation for implementing AI optimized for industrial sites more efficiently," a Posco DX official said. "We will contribute to the systematic use of domestic NPUs and the enhancement of on-site inference capabilities, while working to invigorate the AI semiconductor ecosystem."