Samsung SDS has launched an infrastructure service featuring domestically developed artificial intelligence (AI) semiconductors, aiming to expand its proprietary AI technology ecosystem.
On Monday, Samsung SDS unveiled its NPU as a Service (NPUaaS), a cloud service based on Furiosa AI’s second-generation neural processing unit (NPU), now integrated into the Samsung Cloud Platform (SCP).
Furiosa AI’s second-generation NPU, dubbed Renegade (RGND), is a homegrown NPU tailored for AI inference. It outperforms GPUs in energy efficiency and cost-effectiveness when applying pre-trained AI models to real-world applications such as response generation, document analysis, and image recognition.
While NPUs, hailed as next-generation AI semiconductors, are less versatile than graphics processing units (GPUs) capable of both training and inference, they are seen as a promising infrastructure solution. Their ability to deliver high inference performance with low power consumption could overcome the limitations of costly GPU-centric AI architectures.
This new service allows corporate clients to access NPU servers through SCP on a subscription basis, eliminating the need for direct purchases or in-house data center installations. Clients can also scale NPU resources to match their specific demands and data volumes.
By expanding its offerings to include NPU-based computing resources alongside existing GPU-based cloud services, Samsung SDS aims to broaden AI infrastructure options. This initiative is expected to drive the adoption of domestically developed NPU-based AI infrastructure and contribute to the establishment of sovereign AI capabilities.
Lee Ho-jun, Vice President and Head of the Cloud Services Division at Samsung SDS, emphasized that the introduction of NPUaaS goes beyond launching a new cloud product. It represents a significant step in enabling the customers to leverage high-performance AI technology with greater flexibility and cost-efficiency.