Milestone project validates viability of virtual machines for high-performance AI workloads — enhancing speed, efficiency, and scalability.
Seoul, South Korea – June 25, 2026 – NetApp® (NASDAQ: NTAP), the Intelligent Data Infrastructure company, and SK Telecom today announced the completion of a joint proof of concept (PoC) demonstrating a solution that enables enterprises to scale AI infrastructure in virtualized environments while maintaining high performance. Using NetApp AFX systems with SK Telecom’s Petasus AI Cloud, the companies achieved a performance breakthrough in this project, running high-performance AI workloads on virtual machines (VMs) with NVIDIA GPUDirect Storage (GDS) at 99+ percent of the performance of bare metal workloads.
This breakthrough demonstrates that organizations can reliably deploy large-scale AI workloads in virtualized environments, unlocking the operational benefits of virtualization while maintaining high levels of performance. Traditionally, running AI workloads on VMs incurs a performance trade off. While VMs deliver flexibility and operational efficiency by enabling more effective use of server resources, they can introduce additional processing overhead as multiple workloads share and manage those resources, which may reduce performance for compute-intensive AI workloads. This adversely impacts industries that require ultra-low latency like electronic design automation, financial services, manufacturing, and telecommunications - key growth engines in many markets.
The PoC conducted by NetApp and SK Telecom directly addressed this limitation. By optimizing the software stack and infrastructure design, NetApp and SK Telecom significantly narrowed the performance gap between virtualized and bare metal AI environments, substantially reducing the performance penalty typically associated with virtualization in AI infrastructure.
“The results of this PoC with SK Telecom underscore our unswerving mission to deliver data infrastructure that is optimized, secured and ready for enterprise AI. By virtually eliminating the performance gap between bare metal and virtual machines, we’re opening the door for businesses to unlock faster, more efficient AI training and inference in the cloud, operate more efficiently, and push the envelope in innovation,” said PD Prasad, Vice President and General Manager, Enterprise Storage at NetApp.
“Through this collaboration with NetApp, we have significantly reduced one of the key performance challenges in AI cloud infrastructure—performance degradation in virtualized environments—and at the same time validated Petasus AI Cloud as a production-ready platform for next-generation AI workloads. The Petasus AI Cloud validated through this PoC is expected to deliver real value to organizations seeking to support high-performance AI training and inference while also benefiting from the operational and cost efficiencies of virtualization,” said Minyoung Jeong, Head of AI DC Solution at SK Telecom.
Background and detailed results of the PoC
The PoC was designed to verify interoperability between SK Telecom’s Petasus AI Cloud, a virtualized cloud infrastructure that is catered to build and operate AI data centers, and NetApp AFX, disaggregated storage purpose-built for AI workloads. Additionally, the project aimed to evaluate the feasibility of VM-based AI clusters using NVIDIA GDS, as well as identify and eliminate virtualization overheads in GPU-to-storage direct I/O paths.
NetApp and SK Telecom confirmed interoperability between NetApp AFX and the Petasus AI Cloud, commanding a seamless operation across various hardware and system configurations, and the complete NVIDIA GDS software stack. The project realized industry performance for NVIDIA GDS on virtual machines, reaching 32.7 GB/s in both bare metal and VM environments. CPU usage was reduced by 40% to 50%, freeing system resources. In addition, unnecessary processing overhead was significantly reduced, enabling GPUs to focus more of their cycles on AI training and inference and improving overall compute efficiency.
Powered by ONTAP, NetApp’s foundational operating system
At the core of this achievement is NetApp AFX, a storage architecture designed to meet the extreme performance demands of AI while preserving enterprise-grade data management and operational consistency. Disaggregated by design, NetApp AFX allows independent scaling of performance and capacity, giving cloud service providers like SK Telecom the flexibility to right-size AI infrastructure as workloads evolve. Built on NetApp ONTAP®, AFX delivers the same data management, automation, security, governance and resilience features already deployed in environments worldwide.
Improving AI inference performance and memory utilization efficiency
Petasus AI Cloud is rapidly evolving beyond AI training infrastructure into an inference-focused platform designed to reduce the cost and processing time required for AI systems to generate responses.
This project demonstrated that NVIDIA GDS with virtualization can improve not only AI training performance, but also AI inference efficiency. In particular, by storing and using KV cache data on NetApp AFX, the PoC confirmed that AFX can serve as a high-performance expansion layer that complements GPU memory.
NetApp AFX's high-bandwidth, ultra-low-latency disaggregated storage architecture played a pivotal role in processing the large-scale data I/O demands generated during AI inference, helping reduce data access time and improve inference throughput. This project confirms that the combination of NVIDIA GDS with virtualization and NetApp storage not only supports high-performance training but also drives essential performance and efficiency optimizations for AI inference.
NetApp and SK Telecom’s collaboration
NetApp has been a storage partner to SK Telecom for more than a decade. Building on the success of this PoC, NetApp and SK Telecom plan to continue collaborating on AI data center solution integration, technical validation, and joint customer engagement opportunities for enterprise AI workloads. The companies will also explore how the validated architecture can support GPU-intensive cloud service use cases as customer requirements and deployment models continue to evolve.
Moving forward, NetApp and SK Telecom plan to further advance the AFX-based storage expansion technology validated through this PoC into next-generation memory technologies, enabling temporary data required for AI response generation to be stored and accessed more quickly and efficiently.
Building on the technical insights validated through this PoC, SK Telecom plans to continue enhancing Petasus AI Cloud’s platform capabilities to support high-performance and cost-efficient AI training and inference workloads.
Statements by NetApp about unreleased offerings and future plans are for informational purposes only, are subject to change without notice, and should not be relied upon for purchasing or other decisions. Such statements do not constitute a commitment, obligation, guarantee, or warranty of any kind by NetApp, including about availability, functionality, pricing, or timing.
About NetApp
For more than three decades, NetApp has helped the world’s leading organizations navigate change – from the rise of enterprise storage to the intelligent era defined by data and AI. Today, NetApp is the Intelligent Data Infrastructure company, helping customers turn data into a catalyst for innovation, resilience, and growth.
At the heart of that infrastructure is the NetApp data platform – the unified, enterprise-grade, intelligent foundation that connects, protects, and activates data across every cloud, workload, and environment. Built on the proven power of NetApp ONTAP, our leading data management software and OS, and enhanced by automation through the AI Data Engine and AFX, it delivers observability, resilience, and intelligence at scale.
Disaggregated by design, the NetApp data platform separates storage, services, and control so enterprises can modernize faster, scale efficiently, and innovate without lock-in. As the only enterprise storage platform natively embedded in the world’s largest clouds, it gives organizations the freedom to run any workload anywhere with consistent performance, governance, and protection.
With NetApp, data is always ready – ready to defend against threats, ready to power AI, and ready to drive the next breakthrough. That’s why the world’s most forward-thinking enterprises trust NetApp to turn intelligence into advantage.
Learn more at www.netapp.com or follow us on X, LinkedIn, Facebook, and Instagram.
NETAPP, the NETAPP logo, and the marks listed at www.netapp.com/TM are trademarks of NetApp, Inc. Other company and product names may be trademarks of their respective owners.
.png?width=5760&format=avif&disable=upscale)