KT on the 19th unveiled the KT NPU LLM Station, a corporate AI device that combines a domestic AI Semiconductor with its own large language model (LLM).
The new product is an all-in-one device that integrates AtomMax, a Neural Processing Unit (NPU) for inference from domestic fabless Rebellions, KT's language model Faith K 2.5 Pro, and an operations API platform into a single server. A key feature is that the core components—from the AI Semiconductor to the model and operating system—are built with domestic technology.
The device is installed inside the client company so data storage and AI computation are all handled in-house. Accordingly, it targets demand in the public, defense, finance, pharmaceutical, and manufacturing sectors, where using external Generative AI has been difficult due to network separation or security regulations.
With the server and software pre-integrated, users can immediately employ retrieval-augmented generation (RAG)-based Q&A using in-house documents after installation. By applying an NPU specialized for inference, it improves power efficiency over GPUs of the same class, and it supports standard APIs so existing AI services can be linked without major code changes, the company said.
KT plans to add meeting minutes drafting, coding assistance, and a work automation agent, K-Claw (tentative name), and will pursue customized deployments with specialized developers. In the long term, it plans to expand use to edge data centers for physical AI.
Executive Director Lee Jin-hyung, head of KT's AX Business Division, said, "It is a product that secures Data Sovereignty while enabling on-site AI transformation right away."