Search
IDCNOVA

Samsung SDS launches NPU-as-a-Service with FuriosaAI RNGD chips in South Korea

By: IDCNOVARegion: East Asia
Samsung SDS, the IT and cloud services unit of Samsung, has introduced an NPU-as-a-Service (NPUaaS) offering in South Korea, powered by FuriosaAI's RNGD AI inference accelerators. The service is now available through the Samsung Cloud Platform, marking what the companies describe as the first NPUaaS offering in the country.

The launch reflects a broader industry shift toward flexible, on-demand AI inference infrastructure. As enterprises deploy increasingly complex agentic AI workloads, demand is growing for compute resources that combine high performance with greater efficiency and lower total cost of ownership. By offering access to FuriosaAI's RNGD chips without requiring customers to manage dedicated hardware, Samsung SDS aims to lower the barrier to entry for organizations seeking scalable AI inference capabilities.

FuriosaAI's RNGD accelerators are built on a 5nm Tensor Contraction Processor (TCP) architecture and deliver 512 teraflops of FP8 compute per chip, with a thermal design profile of 180W. Each server can house up to eight RNGD accelerators, resulting in a 3kW thermal design per server. According to FuriosaAI, this configuration enables high inference capacity per rack, making it well-suited for demanding agentic AI workloads.

Samsung SDS has already deployed the RNGD accelerators at its data center in Dongtan, located in Gyeonggi Province. The company has virtualized FuriosaAI's NXT RNGD server and is offering instances in on-demand configurations of 1, 2, 4, or 8 cards. The Dongtan data center, which opened in March 2023, spans six floors with four server rooms and achieves a power usage effectiveness (PUE) of 1.12 through natural air cooling. By relying on outside wind and the region's colder climate instead of traditional air-conditioning, the facility is expected to save 21,443 tons of greenhouse gas emissions by 2035.

June Paik, CEO of FuriosaAI, said the partnership gives enterprises a new way to deploy AI inference infrastructure without the burden of managing dedicated hardware. "As organizations deploy increasingly complex AI workloads, they need inference infrastructure that delivers exceptional performance while improving efficiency and lowering the total cost of ownership," Paik said. "Making RNGD available through Samsung SDS gives enterprises a new way to deploy AI inference infrastructure without managing dedicated hardware. We're excited to support Samsung SDS as it expands access to efficient AI inference infrastructure."

Founded in 2017 and headquartered in Seoul, FuriosaAI launched its NXT RNGD server in September 2025. The collaboration with Samsung SDS is expected to strengthen FuriosaAI's position in the AI inference market while enabling Samsung SDS to expand its cloud portfolio with high-performance, energy-efficient compute options. The move also signals growing momentum in South Korea's AI infrastructure sector, as domestic providers seek to offer alternatives to traditional GPU-based cloud services.