Overview
DDN is supporting SK Telecom (SKT), South Korea’s leading telecommunications company, as it reshapes the nation’s AI landscape with its pioneering GPU as a Service (GPUaaS) offering. At the heart of this commercial offering is DDN’s EXAScaler, which provides the scalable, high-performance storage backbone powering SKT’s Petasus AI Cloud. Together, DDN and SKT are enabling GPUaaS innovation —helping organizations scale, secure, and consume GPU-accelerated AI workloads with commercial SLAs, flexible deployment models, and operational simplicity.
GPUaaS: A Practical Market Need
There is growing global demand for flexible GPU consumption models that let organizations rapidly access GPU compute without long procurement cycles. In response, DDN’s EXAScaler was integrated with SKT’s Petasus AI Cloud as a commercial offering that pairs scalable, high-performance storage with GPU-accelerated compute. This integrated solution is intended to simplify deployment and operations, helping enterprises and service providers rapidly provision commercial-grade GPUaaS across on-prem, colocation, and hybrid environments to meet performance, cost, and operational requirements.
The Challenge: Scaling AI for Enterprises
With the rapid increase in model complexity, dataset size, and SLA requirements, enterprises face unprecedented challenges in scaling AI.
Jian Li, Principal Engineer at SKT, explains:
“Enterprises are grappling with exploding model sizes, tighter SLAs for inference, and larger dataset scales, all while requiring robust multi-tenancy security, efficient hardware utilization, and streamlined operations. The key drivers are model scale and I/O pressure, which demand low-latency, high-bandwidth storage to support large models and datasets — especially when you want cloud-like elasticity for GPU resources.”
For GPUaaS providers and customers, these challenges are compounded by the need to maximize GPU utilization, support mixed workloads (training, fine-tuning, inference), and provide predictable cost and performance.
Alex Shim, Country Manager for Korea, comments:
“Our collaboration with SK Telecom gives customers shorter training cycles, higher GPU utilization, and a clear path from on-prem to cloud—without re-architecting. It’s practical performance and a consumption model that accelerates AI value for GPUaaS.”
The Solution: Petasus AI Cloud Packaged as a GPUaaS Platform
SKT’s Petasus AI Cloud evolved from the company’s telco cloud experience. Lessons from building robust, automated cloud operations—such as orchestration, lifecycle automation, and operational rigor – directly informed the design of GPUaaS.
Jian explains:
“As a telco, SKT developed strong operational practices around automation and scale. We applied the same principles to our GPU cloud, automating GPU VM provisioning, workload scheduling, and tenant isolation so developers and data teams can instantly access AI compute resources with cloud-like agility.”
In this commercial GPUaaS package, DDN’s EXAScaler serves as a Lustre-based parallel file system purpose-built for AI and HPC workloads. The integration enables SKT to deliver high-performance, low-latency data access alongside cloud-like agility – allowing enterprises to consume GPUs and data infrastructure with the ease of standard cloud services while maintaining fine-grained operational control.
“We applied telco-grade operational experience to our GPU cloud, particularly in addressing networking, scheduling, and storage challenges. This allowed us to achieve consistent, production-grade performance and user experiences in GPUaaS use cases,” Jian notes.
Market Expansion: Targeting Japan with Technical PoCs
Leveraging this completed solution, DDN and SKT have identified demand among customers in Japan for cloud-like, enterprise-grade GPUaaS. To validate operational fit and accelerate commercial adoption, DDN and SKT are conducting technical proof-of-concept (PoC) projects with prospective Japanese customers and partners. These PoCs aim to demonstrate that the integrated Petasus AI Cloud + DDN EXAScaler solution can be rapidly provisioned and operated at production-grade levels to meet performance, operational, and compliance expectations in enterprise environments.
Why DDN Stands Out
DDN’s EXAScaler was selected for this commercial GPUaaS package integration, not only for performance but also for its open-source leadership.
“EXAScaler is built on the open-source Lustre project, which is rare among storage vendors who typically don’t open-source their core features. DDN leads Lustre’s development, ensuring new features originate from them, which drives significant customer traction. The open-source model also allows global contributions, reducing development cycles compared to proprietary vendor solutions,” Jian emphasizes.
This approach ensures rapid innovation, adaptability, and long-term sustainability—critical for GPUaaS infrastructure that must evolve alongside fast-moving AI workloads and operational requirements.
Results and Benefits
The collaboration between DDN and SKT has produced tangible solutions for enterprises and service providers who want to adopt GPUaaS with strong performance, operational control, and flexible deployment:
- On-Demand GPU Access – Developers and data teams can provision GPU resources instantly, democratizing access to AI compute.
- High Performance at Scale – Low-latency, high-bandwidth storage supports massive AI models and datasets.
- Flexible Deployment Models – Deploy on-premises, in customer datacenters, or via colocation; hybrid and cloud-bursting options allow customers to align deployment with cost, latency, and policy needs.
- Operational Efficiency & Observability – Integrated orchestration, multi-tenant isolation, and monitoring streamline operations and shorten time-to-insight.
- Cost Predictability & GPU Utilization – Better scheduling, pooling, and utilization reduce idle GPU time and optimize TCO for AI workloads.
- Telco-Grade Reliability – SKT delivers enterprise-class availability and resiliency, built on telco-grade operational experience.
- Low latency – 200x faster than the competition and close to bare metal performance.
Conclusion
DDN’s commercial collaboration with SK Telecom demonstrates how AI infrastructure innovation can support organizational objectives for performance, cost efficiency, and operational simplicity. By combining HPC-grade performance with cloud agility, the GPUaaS platform— packaged on Petasus AI Cloud and supported by DDN EXAScaler—offers a practical path for organizations to deploy GPUaaS infrastructure with enterprise-grade performance and operational simplicity. This collaboration represents a commercial technology solution intended to help customers address the needs of modern GPUaaS deployments. It is designed to provide secure, high-performance infrastructure for enterprises and research institutions and to support those looking to adopt rapid, production-grade GPUaaS. The ongoing PoCs in Japan illustrate the solution’s international applicability and the joint commitment of DDN and SKT to support customers seeking fast GPUaaS adoption in markets that value performant, locally managed deployments.
