Samsung SDS (018260) announced on the 23rd that it has launched Korea's first GPU-as-a-Service (GPUaaS) offering based on Nvidia's latest B300 graphics processing unit.
The B300 GPUaaS was introduced to meet surging demand for high-performance computing as companies move beyond artificial intelligence model development into the AI inference stage, where models are deployed in real-world services. The B300 GPU is equipped with 12-layer HBM3E high-bandwidth memory. It delivers 288 gigabytes of memory capacity per GPU and bandwidth of 8 terabytes per second. Compared with Nvidia's H100 model, memory performance has improved by 3.6 times in capacity and 2.4 times in bandwidth for AI inference tasks requiring complex computations. This dramatically reduces data bottlenecks, where slow memory data transfer speeds degrade overall GPU performance when running large language models.
The B300 GPUaaS is available through Samsung Cloud Platform (SCP), Samsung SDS's proprietary cloud service. Samsung SDS has been building a GPUaaS ecosystem since launching A100-based services in 2021 and H100-based services in 2023, enabling GPUs to serve as dedicated AI infrastructure.
Customers adopting the B300 GPUaaS can efficiently process large AI models using high-capacity memory. The service minimizes latency for performance-intensive AI applications such as AI agents and image, video, and code generation and analysis.
The subscription model, which charges only for actual GPU usage, helps companies optimize upfront investment costs. Even amid tight GPU supply conditions, businesses can immediately access Nvidia's latest architecture through SCP. Customers also benefit from processing sensitive corporate data within a secure cloud environment backed by Samsung SDS's security capabilities.
Samsung SDS plans to introduce new GPUaaS subscription models in the third quarter of this year. These include a "serverless inference service," which charges only for tokens consumed without separate infrastructure fees, and an "AI training service," which automatically runs distributed AI training when developers input code and data.
"We will actively support the industry's AI transformation with Korea's first B300 GPU service," said Lee Ho-jun, executive vice president of the cloud services business unit at Samsung SDS.