What is KT NPU LLM Station and what does it include?

KT NPU LLM Station is an integrated AI appliance that combines Rebellions’ ATOM-MAX neural processing unit, KT’s Mi:dm K 2.5 Pro large language model and an application programming interface platform. KT said the package lets enterprises run hardware, model and related services together.

Why is KT positioning this AI server for security-sensitive companies?

KT said the appliance is installed inside a customer’s facilities, so data and AI computations are processed internally. It said this setup is aimed at organizations with strict security requirements and network-separation rules, including public agencies, defense companies and financial institutions.

What future AI features does KT plan for the NPU LLM Station?

KT said it plans to add AI agents for meeting-minute generation, coding assistance and workflow automation, including a planned agent called K-Claw. It also said it wants to work with specialized agent developers for customized systems and eventually expand the technology into edge data centers for physical AI.

KT puts Korean AI chips, language model into one enterprise server

Listen

Summary

KT has packaged a Korean-made AI processor, language model and operating platform into one enterprise server in a launch announced Wednesday. The KT NPU LLM Station uses Rebellions’ ATOM-MAX neural processing unit and KT’s Mi:dm K 2.5 Pro large language model. KT said the system lets companies run generative AI inside their own networks and target security-sensitive industries. The company also plans AI agents, customized customer systems and eventual expansion into edge data centers.

Key Facts

KT said the system includes an application programming interface platform so companies can operate the hardware, model and related services as a single package.
The appliance is installed inside a customer’s facilities, with data and AI computations processed internally.
KT said the system could help public agencies, defense companies, pharmaceutical firms, manufacturers and financial institutions adopt AI without relying on overseas generative AI services.
KT said customers can use retrieval-augmented generation for internal documents after installation without building separate AI infrastructure.
KT plans to add AI agents for meeting-minute generation, coding assistance and workflow automation, including a planned agent called K-Claw.

By Lee Kyung-min

Published Aug 19, 2026 2:17 pm KST

An engineer demonstrates the KT NPU LLM Station during a presentation in Seoul. Courtesy of KT

KT has packaged a Korean-made artificial intelligence (AI) processor, language model and operating platform into a single enterprise server, offering companies a way to run generative AI entirely within their own networks.

KT said Wednesday it launched the KT NPU LLM Station, an integrated AI appliance built around Rebellions’ ATOM-MAX neural processing unit and KT’s proprietary Mi:dm K 2.5 Pro large language model.

The system also includes an application programming interface platform, allowing companies to operate the hardware, model and related services as a single package.

KT said ATOM-MAX is an inference-focused Korean NPU that has demonstrated higher processing performance and power efficiency than comparable graphics processing units in certified testing.

Mi:dm K 2.5 Pro is designed for complex reasoning and agent-based applications, according to KT.

The system is installed inside a customer’s facilities, with data and AI computations processed internally, a feature aimed at industries with strict security requirements and network-separation rules.

KT said the appliance could help public agencies, defense companies, pharmaceutical firms, manufacturers and financial institutions adopt AI without relying on overseas generative AI services.

The company said customers can begin using functions such as retrieval-augmented generation for internal documents after installation without building a separate AI infrastructure.

The use of an inference-focused NPU is also intended to reduce power consumption compared with comparable GPU servers, while standard APIs can allow existing AI services to connect without code changes.

KT plans to add AI agents for tasks including meeting-minute generation, coding assistance and workflow automation, including a planned agent called K-Claw.

The company said it also plans to work with specialized agent developers to build customized systems for customers and eventually expand the technology into edge data centers for physical AI.

It said the product combines the company’s AI capabilities with Korean semiconductor technology to make sovereign AI immediately usable in industry.

This article was published with the assistance of generative AI and edited by The Korea Times.

Explore More