On the 19th, KT announced the launch of 'KT NPU LLM Station,' an enterprise-grade AI appliance that combines domestic AI chips and a domestic large language model (LLM) into one server. This all-in-one product features the AtomMax NPU from domestic AI chipmaker Revelion, KT's independently developed LLM 'Mitum K 2.5 Pro,' and an operational API (application programming interface) platform.
The product is notable for being built entirely with domestic technology—from AI chips to models and operating platforms—enabling immediate implementation of the government's sovereign AI policy direction at the enterprise level.
AtomMax (ATOM-MAX) is a domestic NPU specialized for inference, which has been verified through official testing to deliver higher processing performance and power efficiency compared to equivalent GPUs (graphics processing units). 'Mitum K 2.5 Pro' is KT's proprietary model known for its strengths in complex reasoning and agent utilization.
A key feature of the KT NPU LLM Station is its robust security. The device is installed within client organizations, ensuring that all data and AI computations are processed exclusively internally. It is expected to serve as a catalyst for AX transformation in regulated industries such as public sector, defense, pharmaceuticals, manufacturing, and finance, which have been unable to adopt foreign AI due to network segregation regulations and stringent security requirements.
The product also lowers adoption barriers. As a fully integrated device combining servers, AI models, and platforms, it enables immediate use of AI functions such as internal document-based question-and-answer (RAG) starting from the day of installation, without requiring separate setup. By adopting an inference-specialized NPU, it achieves higher power efficiency compared to equivalent GPU servers. Additionally, by providing industry-standard APIs, companies can integrate the system simply by updating access information for their existing AI services—without any code modifications.
KT plans to sequentially equip the platform with its own developed AI agents, including meeting minute creation, coding support, and an automation agent tentatively named 'K-Claw.' The company also intends to expand into customized implementation projects through collaboration with specialized agent development firms.
Furthermore, KT aims to expand the application scope of 'KT NPU LLM Station' to edge data centers (Edge DC) for physical AI, paving the way for the realization of low-power, low-latency AI infrastructure.
KT AX Business Division Head Lee Jin-hyeong (SVP) stated, "The KT NPU LLM Station combines KT's full-stack AX capabilities with domestic semiconductors to enable immediate use of sovereign AI in real industrial settings." He added, "It will offer the most practical option for companies and institutional investors seeking to begin their AI transformation while safeguarding data sovereignty."
Meanwhile, KT plans to contribute to the realization of sovereign AI and strengthen the competitiveness of Korea's AI ecosystem by utilizing domestic NPU in the Ministry of Science and ICT's 'AI for All' project.
in TECH
