This page was machine-translated and may differ from the original. View original

KT Launches Domestic NPU and LLM Integrated Sovereign AI Appliance

Google 우선 소스Published2026.08.19 09:27



Equipped with Rebellio ATOM-MAX and Midum K 2.5 Pro, Immediately Usable from Day of Installation


 A fully-integrated enterprise AI appliance combining domestic AI semiconductors and a proprietary language model in a single server has emerged. With a closed-structure architecture where data does not leave the premises, it targets AI Transformation (AX) adoption demand in regulated industries such as public, defense, and finance sectors where adopting overseas generative AI has been difficult due to security regulations.


On the 19th, KT announced the launch of 'KT NPU LLM Station,' an enterprise sovereign AI appliance integrating Rebellio's domestic AI semiconductor company NPU 'ATOM-MAX,' its proprietary LLM (Large Language Model) 'Midum K 2.5 Pro,' and an operations API platform.


KT explained that the structure aligns with the government's sovereign AI policy direction by filling all core components from AI semiconductors to models and operations platforms with domestic technology.

 
ATOM-MAX is an inference-specialized domestic NPU that has verified superior processing performance and power efficiency compared to equivalent-class GPUs in certified tests.


'Midum K 2.5 Pro' is a proprietary KT model with strengths in high-difficulty inference and AI agent utilization.


The product is installed within the customer's premises, and all data and AI operations are processed only in-house.


KT anticipates that adoption pathways in public, defense, pharmaceutical, manufacturing, and financial sectors, where using overseas generative AI has been difficult due to security regulations such as network segregation, will open up.


As a fully-integrated product combining server, AI model, and platform, KT states that AI features such as retrieval-augmented generation (RAG) based on in-house documents can be utilized from the day of installation without separate construction.


Supporting industry-standard APIs enables compatible operation without code modifications by simply changing existing AI service connection information.


KT plans to sequentially integrate meeting minutes generation, coding support, and work automation agent 'K-Claw (tentative name),' and expand utilization areas to include edge data centers (Edge DC) for physical AI.


KT plans to participate in the Ministry of Science and ICT's 'AI for All' project using domestic NPUs.


KT AX Division Head Lee Jin-hyung, Senior Executive, stated, "It will become the most practical choice for companies and organizations seeking to begin AI transformation while protecting data sovereignty."

본 기사에 대한 정정·반론·추후보도 청구는 보도 청구 안내를, 그간 게재된 보도문은 정정·반론보도 모아보기를 참고해 주세요.
명세환 기자
명세환 기자