A South Korean startup is about to shake up cloud AI infrastructure by offering a specialized chip that processes artificial intelligence workloads far more efficiently than the graphics processors that currently dominate the market. Starting in July, Samsung SDS will offer Puriosa AI's Renegade neural processing unit (NPU), a specialized chip designed specifically for running AI models in data centers. This marks the first time a domestic Korean cloud company has commercialized a native NPU-based service, potentially opening a new front in the competition against Nvidia's entrenched GPU ecosystem .

What Makes Renegade Different From Traditional AI Chips?

The Renegade is a second-generation NPU built from the ground up to handle AI inference, the process of running trained AI models to generate predictions or responses. Unlike general-purpose graphics processors that Nvidia designed for gaming and later adapted for AI, the Renegade was engineered specifically for this task. The efficiency gains are striking: according to Puriosa AI, the Renegade can handle up to 7.4 times more user requests while consuming the same amount of power as Nvidia's RTX Pro 6000, a high-end GPU commonly used in professional AI applications .

To demonstrate this capability, Puriosa AI's leadership showed a live demonstration where the Renegade processed 512 simultaneous requests using Exemite 4.032B, a large language model (LLM), which is an AI system trained on vast amounts of text to understand and generate human language, developed by LG AI Research Institute. The chip achieved performance nearly equivalent to Nvidia's RTX Pro 6000 while showing superior power efficiency .

How Will Samsung's Cloud Service Make NPUs Accessible to Customers?

• Flexible Scaling: Samsung SDS is designing a service model called NPUaaS (NPU as a Service) that lets customers purchase computing capacity in modular units of 1, 2, 4, or 8 NPU chips, paying only for what they need rather than buying entire systems.
• Cloud and On-Premises Options: The Renegade can be deployed either in Samsung's cloud infrastructure or in customers' own data centers, giving organizations flexibility in how they want to run their AI workloads.
• Expanded Market Access: By integrating the Renegade into Samsung's cloud platform, external customers gain a new way to access specialized AI hardware without building their own infrastructure.

"We are preparing a service-type NPU that allows customers to use NPU as much as they need in units of 1, 2, 4, and 8," explained Lee Ju-pyeong, managing director of Samsung SDS. "Samsung's external customers will be able to increase their contact points by accessing NPU through Samsung Cloud."

Lee Ju-pyeong, Managing Director at Samsung SDS

Why Is This Launch Significant for the AI Infrastructure Market?

The cloud AI market has been almost entirely dominated by Nvidia, which supplies the GPUs that power most large language models and AI services. Nvidia's dominance gives the company enormous pricing power and makes it difficult for cloud providers to differentiate their offerings. By introducing a purpose-built alternative, Samsung and Puriosa AI are attempting to break that monopoly and offer customers a more cost-effective path to AI inference at scale .

Puriosa AI began mass production of the Renegade in January and plans to manufacture approximately 20,000 units throughout 2026, with these chips flowing into Samsung's cloud service and potentially other partners . This production volume suggests the company is serious about competing for real market share rather than simply testing the waters.

"It is a product that can implement AI agent systems most efficiently in the cloud or on-premises environment," said Baek Joon-ho, CEO of Puriosa AI.

Baek Joon-ho, CEO at Puriosa AI

The timing matters too. As enterprises increasingly deploy AI agents, autonomous systems that can perform tasks with minimal human intervention, the demand for efficient inference hardware is growing rapidly. A chip that can handle more requests per watt of power consumed directly translates to lower operational costs for cloud providers and their customers, making it an attractive alternative to Nvidia's offerings .

For Samsung, this partnership represents a strategic move to strengthen its cloud business and create a competitive moat against rivals like AWS and Google Cloud. By offering exclusive access to specialized hardware, Samsung can attract customers who are price-sensitive or concerned about vendor lock-in with Nvidia. The success of this initiative could reshape how cloud providers compete on AI infrastructure, shifting the focus from raw performance to efficiency and cost per inference.