The Korea Electronics Technology Institute (KETI) is a leading Korean research institute driving digital transformation and AI transformation (AX) across the electronics and manufacturing industries. KETI bridges research and industry by developing commercially viable technologies that accelerate innovation.

KETI’s research spans key manufacturing sectors, including semiconductors, displays, automotive and electric vehicle components, electronic materials, and energy. In recent years, accelerating the adoption of AI across the manufacturing lifecycle—from product design to shop-floor operations—has become one of its strategic priorities.

To support this vision, KETI plans to establish a dedicated Manufacturing AI Data Center (AIDC) in 2026. The platform will serve as the foundation for developing, training, validating, and deploying AI models tailored for manufacturing applications.

Challenge: Transitioning from Standalone Infrastructure to a Shared AI Platform

To enable multiple researchers and developers to work on AI models simultaneously, KETI needed a high-performance GPU infrastructure with flexible resource sharing.

The platform also required integrated MLOps capabilities covering AI development, training, validation, deployment, and management, along with centralized management of GPUs, storage, and networking resources to improve efficiency.

However, building a scalable and shared AI infrastructure for long-term operation presented significant challenges. KETI identified three key challenges:

  • High Barriers to Building High-Performance AI Infrastructure:
  • High Barriers to Building High-Performance AI Infrastructure: KETI needed to establish a GPU AI training environment that could support multiple researchers simultaneously. Beyond computing performance, the infrastructure had to ensure stability, user isolation in a multi-tenant environment, and efficient allocation of limited GPU resources.
  • Inconsistent management across the AI development lifecycle: AI projects span far more than model development and training — they include validation, deployment, operations, and continuous updates. Without an integrated MLOps platform, tools and data at each stage had to be managed separately, adding operational complexity and slowing down both R&D and collaboration.
  • Fragmented heterogeneous resources, low utilization: When GPUs, storage, and networking are managed separately across disconnected systems, resource silos form. Some resources sit idle while others face contention, and as operations grow more complex, overall infrastructure utilization drops and management costs climb.

In short, KETI’s challenge wasn’t simply adding more hardware — it was building an intelligent AI infrastructure platform that could unify heterogeneous compute resources, simplify AI development workflows, and balance efficiency, manageability, stability, and scalability in a multi-tenant environment. This infrastructure needed to support not only the development of manufacturing AI models and applications, but also serve as the foundation for lab-stage validation, testbed verification, and real-world deployment on the manufacturing floor.

The Impact and Value of AI-Stack

With AI-Stack in place, KETI built a shared AI infrastructure environment for multiple researchers and projects, transforming resource management and operations across its manufacturing AIDC. Researchers can now provision GPU resources on demand, cutting wait times for model development and training. Platform administrators manage GPUs, storage, and core infrastructure through a single interface, reducing operational complexity and improving allocation efficiency. INFINITIX’s intelligent resource scheduling and sharing capabilities minimize GPU idle time and allocate resources dynamically based on project priority and workload characteristics — boosting GPU utilization while laying the groundwork for lower total cost of ownership (TCO).

AI-Stack also gives KETI a development infrastructure built to scale as GPU and compute needs grow. On this foundation, KETI can provide the computing environment required for manufacturing-specific foundation models (MFMs) and smart manufacturing research, systematically supporting the development, validation, and real-world testing of AI technologies.

Looking Ahead

AI-Stack will continue to expand KETI’s manufacturing AIDC infrastructure and platform capabilities. By offering a shared environment where more researchers and developers can build, train, test, and validate AI models, it aims to lower the barrier to accessing high-performance AI infrastructure.

Through heterogeneous resource orchestration, sharing, and unified management, AI-Stack is designed to keep improving infrastructure efficiency over time — maintaining consistent operations even as compute scale and user numbers grow, and supporting a scalable, shareable manufacturing AI computing platform. KETI plans to use this infrastructure to help Korean manufacturers and research institutions advance their AI capabilities, accelerating the validation and real-world deployment of manufacturing-specific AI models and smart manufacturing applications.