Samsung SDS has introduced the latest GPUs for national AI research, enabling domestic researchers to utilize a world-class computing environment.
On August 11, Samsung SDS announced that it has been selected as the GPU resource supplier for the '2026 AI Research Computing Support Project,' supported by the Ministry of Science and ICT and the Institute for Information and Communications Technology Planning and Evaluation (IITP). The service commenced on July 20.
This initiative aims to provide AI research GPU resources to researchers in academia and research institutions in South Korea.
The project focuses on supplying computing resources to research entities that struggle to secure large-scale computing infrastructure independently, allowing them to conduct research on ultra-large AI models and large language models (LLMs).
The project has a budget of approximately 15.3 billion won and will provide services until March 2027. The evaluation criteria for selecting the supplier included GPU supply plans, resource scale, research environment quality, operational support systems, and security capabilities. According to the company, multiple suppliers were selected, with Samsung SDS ranked first.
Samsung SDS will provide not only NVIDIA H100 (Hopper) GPUs but also the latest B300 (Blackwell Ultra) based GPU cluster. The B300 exceeds the specifications required in the request for proposal (RFP) for Hopper-level GPUs. It offers significantly improved performance and large memory capacity for training and inference of large parameter models compared to previous generations.
The domestic research community has raised concerns about the urgent need to bridge the gap between available resources and global infrastructure levels due to competition for the latest GPUs. Samsung SDS's direct investment in the national AI research GPU computing project is seen as a foundation for domestic researchers to conduct their work in a top-tier computing environment.
Samsung SDS also provides a large-scale clustering configuration that pools multiple GPUs into a single resource pool, along with a high-speed interconnect network.
This setup allows researchers to flexibly respond to both long-term large projects and short-term experimental tasks. Beyond simply supplying GPUs, the company pre-configures and provides CPU, memory, storage, network, and development environments. This support alleviates the burden of infrastructure setup, allowing researchers to focus on their work.
Additionally, Samsung SDS launched NVIDIA's B300 GPU-based GPUaaS (GPU as a Service) through its Samsung Cloud Platform (SCP) in March, becoming the first in South Korea to do so. In July, it integrated the second-generation NPU, Renegade (RNGD), based on the domestically produced neural processing unit (NPU) called FuriosaAI, into SCP, launching NPUaaS (NPU as a Service).
Lee Ho-jun, Vice President and Head of Cloud Services at Samsung SDS, stated, "Through this project, we will provide stable services so that domestic researchers can achieve innovative results based on world-class AI infrastructure. We will continue to contribute to expanding the national AI research infrastructure and enhancing South Korea's AI competitiveness."
Samsung SDS has been building its capabilities in operating high-performance AI infrastructure, having introduced NVIDIA A100 and H100-based GPUaaS in 2021 and 2023, respectively. The company has experience supplying GPU resources to over 60 AI SMEs, startups, research institutions, and universities as a cloud service provider in the high-performance computing support project led by the Ministry of Science and ICT.
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.