KT Cloud announced on the 24th that it is providing optimized high-performance AI infrastructure by applying the NVIDIA H200 to GPU as a Service (GPUaaS).
KT Cloud is currently operating GPUaaS, which allows usage-based GPU resource utilization, targeting public institutions and AI startups in South Korea. It has provided various NVIDIA-based services such as 'AI Train' for AI training that requires large-scale GPUs and 'AI SERV' for inference that requires continuous GPU supply. With the introduction of the H200, it plans to further enhance the performance of the existing AI Train service centered around the H100.
The NVIDIA H200 is a GPU based on the latest Hopper architecture, which has improved computational performance and power efficiency compared to the previous H100, optimizing the performance of AI training and inference infrastructure with expanded memory bandwidth.
Additionally, it is expanding its business in the Neural Processing Unit (NPU) area. It is currently operating the AI SERV NPU service in collaboration with Rebellion and is optimizing small language models (sLM) using 'Atom Max,' which is scheduled for release in May this year. Furthermore, it is also reviewing collaboration on AI semiconductor projects with American semiconductor design firm Ampere.
KT Cloud will proceed with the CSAP certification process for inference-only products targeting public institutions in the second half of the year and plans to strengthen its response to public AI demand.