Samsung SDS Bets Big on Domestic Silicon for AI Inference
Samsung SDS has made a pointed bet on Korean silicon, fitting into a larger global pattern of seeking alternatives to traditional hardware dominance. The company launched a new cloud service running on chips from domestic startup FuriosaAI, pushing aggressively into the market for cheaper AI infrastructure. This marks Korea’s first deep learning service powered entirely by a homegrown chip.
Introducing NPUaaS and the RNGD Chip
The new service introduces Neural Processing Unit as a Service, or NPUaaS, which is built on FuriosaAI’s second-generation RNGD (Renegade) chip.
While traditional GPUs are typically used to train massive AI models, NPUs specialize in AI inference.
Inference is the phase where an already-trained model is actually put to work to generate text, analyze data, or identify images.
Samsung SDS is not competing on raw training power; instead, it is targeting the cheaper, higher-volume inference workloads.
The pitch to enterprise businesses is focused heavily on cost efficiency. The RNGD chip promises drastically lower power consumption compared to traditional GPUs. Under the subscription model, clients lease computing power through the cloud rather than purchasing and maintaining their own physical servers. Customers have the flexibility to scale their infrastructure across one, two, four, or eight chips depending on their specific needs, making it ideal for companies wary of huge upfront hardware spending.
The Drive for Sovereign AI and Data Security
Beyond cost savings, the deeper motive behind this launch is technological sovereignty.
The NPUaaS runs inside a sovereign cloud environment on the Samsung Cloud Platform (SCP). This is aimed squarely at public-sector and defense-adjacent clients who demand strict data security and compliance with rigorous regulations. This strategy not only provides cheaper compute but also ensures that sensitive Korean workloads remain entirely off foreign hardware.
This move mirrors other global efforts to build domestic semiconductor capabilities and reduce reliance on foreign technology. For example, China is building chips like the DF1000 to escape US controls, the UAE recently won license-free NVIDIA access through diplomacy, and now Korea backs a domestic challenger for inference.
Samsung SDS Executive Vice President Lee Ho-jun noted that this launch allows customers to deploy high-performance AI technology more flexibly and cost-efficiently. As the market evolves, inference—not just training—may be where affordable, localized AI infrastructure realistically begins.