POSCO DX has partnered with AI semiconductor startup Mobilint to transition its AI infrastructure for industrial sites. The plan is to move away from a structure centered on foreign GPUs, apply domestic NPUs to manufacturing sites, and verify whether AI inference can be processed directly within factory automation systems. The industry is paying close attention to this initiative, as it aims to secure real-time processing and security while lowering costs and power consumption.

POSCO DX signed a business agreement with Mobilint on April 2 at its Pangyo headquarters to implement NPU-based industrial AI. Prior to this, in February, it invested 3 billion won in Mobilint through a corporate venture capital fund established jointly with POSCO Technology Investment. This agreement is interpreted as a follow-up step to fully connect technology verification and commercialization following the investment.

The two companies plan to combine POSCO DX's industrial automation platform with Mobilint's NPU technology to jointly develop AI hardware systems, conduct field trials, verify technology, and identify new business opportunities. Manufacturing, robotics, logistics, and safety are being considered as priority application sectors. In particular, the focus was placed on accelerating on-site response speed through an edge AI architecture that runs AI near facilities without going through data centers.

In industrial settings, AI operational methods that reduce reliance on central servers are emerging as an alternative due to control environments requiring immediacy and data security issues. NPUs, semiconductors specialized for AI inference, are being considered as a substitute for GPUs in terms of power efficiency and operating costs. POSCO DX plans to expand their application to steel and secondary battery material production sites, as well as the logistics sector.

Mobilint is also accelerating its business expansion focused on industrial sites after recently completing a large-scale investment round. This collaboration is expected to serve as a test case to gauge whether domestically produced AI semiconductors can move beyond the R&D stage and integrate into actual manufacturing operational systems.