KT launches private AI server built on domestic chip technology

KT has launched the KT NPU LLM Station, a corporate AI server designed to keep data management and AI processing inside customer facilities. The appliance combines KT’s Mi K 2.5 Pro LLM, Rebellions’ ATOM-MAX neural processing unit and an operational API platform in a single system.

The product is aimed at organizations that face limits on using foreign cloud-based generative AI because of network segregation, data protection or sovereignty requirements. Target sectors include financial services, manufacturing, pharmaceuticals, the public sector and defense.

KT says certified testing of the ATOM-MAX NPU showed gains in processing speed and power efficiency compared with similar GPUs. The system also includes RAG capabilities, allowing companies to connect internal documents to the AI system for company-specific question answering and information retrieval.

KT plans to expand the platform with AI agents for meeting-minute creation, coding support and business-process automation, including an agent tentatively called K-Claw. The company also intends to explore edge data center uses for physical AI, focusing on low-power and low-latency computing in industrial environments.