Samsung SDS has launched a subscription-based neural processing unit (NPU) service on its Samsung Cloud Platform, built on FuriosaAI's second-generation "Renegade" chip, allowing companies to access specialized AI hardware without purchasing expensive servers outright. The service marks a shift in how organizations can deploy artificial intelligence inference, the process of applying trained AI models to real-world tasks like document analysis, image recognition, and text generation.

What Is an NPU and Why Does It Matter?

A neural processing unit is a specialized chip designed specifically for running AI workloads, as opposed to general-purpose graphics processing units (GPUs) that handle many different computing tasks. FuriosaAI's Renegade NPU is a Korean-made processor engineered for AI inference, offering what the company describes as superior power efficiency and cost-effectiveness compared to GPUs when deploying trained models into production. Think of it this way: if GPUs are like Swiss Army knives that can do many things reasonably well, NPUs are precision tools built for one job, and they do it with less energy and lower costs.

The distinction matters because inference, the stage where AI models actually answer questions or process images for real users, consumes enormous amounts of computing power. Companies running large language models or image recognition systems at scale face steep electricity bills and infrastructure costs. Specialized chips like the Renegade promise to reduce both.

How Does Samsung's Subscription Model Work?

Rather than forcing customers to purchase and maintain their own NPU servers, Samsung SDS offers the service on a pay-as-you-go basis. Customers can select NPU capacity in modular units, choosing configurations of 1, 2, 4, or 8 cards depending on their workload demands and data scale. This flexibility appeals to companies with fluctuating AI needs; they can scale up during peak periods and scale down when demand drops, paying only for what they use.

The service integrates with other Samsung Cloud Platform components, including high-performance storage, compute resources, and high-speed networks, creating a complete environment for AI workloads. Importantly, Samsung SDS plans to offer this NPUaaS (NPU-as-a-Service) in a sovereign cloud environment, a setup designed for government and public-sector customers subject to strict data residency and security regulations.

What Are the Key Benefits for Organizations?

• Cost Reduction: Companies avoid the capital expense of purchasing NPU hardware and building dedicated data center infrastructure, shifting to operational expenses instead.
• Flexibility and Scalability: Organizations can adjust their NPU capacity in units of 1, 2, 4, or 8 cards, matching their actual inference workload without overprovisioning.
• Regulatory Compliance: The sovereign cloud option allows public-sector and regulated industries to run AI inference while maintaining data residency requirements and security standards.
• Power Efficiency: FuriosaAI's Renegade chip delivers lower power consumption than GPU-based inference, reducing operational costs and environmental impact.

What Do Industry Leaders Say About This Move?

"The launch of this NPUaaS is significant not simply as the introduction of a new cloud product, but in that customers can use high-performance AI technology more flexibly and cost-effectively," said Lee Ho-jun, executive vice president and head of the Cloud Services Division at Samsung SDS.

Lee Ho-jun, Executive Vice President and Head of Cloud Services Division at Samsung SDS

Lee added that Samsung SDS plans to continue expanding its cloud offerings to meet evolving customer needs and strengthen its position in the cloud AI ecosystem. This statement reflects a broader industry trend: as AI inference becomes a critical business function, companies are seeking alternatives to expensive GPU-based cloud services and looking for ways to run models closer to their data.

Why Is This Timing Significant?

The launch arrives as organizations worldwide grapple with the rising costs of AI deployment. Large language models and computer vision systems require constant inference capacity, and GPU cloud services from major providers have become expensive as demand surges. By offering NPU-based inference as a subscription service, Samsung SDS positions itself as a cost-conscious alternative, particularly for companies processing high volumes of AI tasks that don't require the flexibility of general-purpose GPUs.

The sovereign cloud component is equally important. Many governments and regulated industries cannot store sensitive data on public cloud infrastructure, creating a market gap that Samsung SDS is now addressing. This opens the service to sectors like healthcare, finance, and defense, where data sovereignty is non-negotiable.

Samsung SDS's move reflects a maturing AI infrastructure market where specialized hardware, flexible pricing models, and regulatory compliance are becoming table stakes. As more organizations move beyond AI experimentation into production deployment, the demand for efficient, scalable, and compliant inference infrastructure will only grow.