SK Telecom to build sovereign AI infrastructure

South Korean telecoms giant SK Telecom (SKT) is building a sovereign artificial intelligence (AI) infrastructure powered by Nvidia Blackwell servers from Supermicro and an AI operating system (OS) from Vast Data.

Dubbed Haein Cluster, the new infrastructure is core to SKT’s transformation from a traditional telco into an AI-first company. Designed to support the telco’s Petasus AI Cloud service, it will meet the needs of AI model development while ensuring all data and processing remains in South Korea. The infrastructure, already selected for the Ministry of Science and ICT’s AI compute rental programme, will be used to develop national AI foundation models.

Traditional AI infrastructure built on bare-metal environments often requires days or weeks to provision new workloads, a major bottleneck for developers and enterprises. SKT has slashed this time dramatically by deploying a virtualised AI infrastructure, which can spin up GPU (graphics processing unit) environments in as little as 10 minutes while maintaining performance levels comparable to bare-metal systems.

DK Lee, vice-president and head of the AI datacentre lab at SKT, said Vast Data has been instrumental in helping the telco move from legacy bare-metal deployments to a fully virtualised, production-grade AI cloud. “The Vast AI OS powers the performance, simplicity and flexibility needed to support the next generation of sovereign AI workloads and gives us the confidence to scale fast and securely,” he added.

Essentially a GPU-as-a-service, Petasus AI Cloud is built on a disaggregated, shared-everything (DASE) architecture from Vast, running on Supermicro’s Nvidia HGX servers. This creates a high-throughput and secure multi-tenant environment capable of supporting both large-scale AI training and real-time inference.

The service is designed for enterprise-grade uptime and resilience with minimal operational overhead, allowing for dynamic allocation of GPUs, storage, and networking resources for each tenant. AI application developers can also expect a unified pipeline for full lifecycle AI development and deployment. There’s also protocol flexibility without the need for client-side gateways, simplifying access for government, research and enterprise AI customers across South Korea.

“SK Telecom is defining the future of national-scale AI infrastructure, and Vast is proud to support their vision,” said Sunil Chavan, vice-president for Asia-Pacific at Vast Data. “By eliminating traditional bottlenecks around data movement, provisioning and security, Vast is enabling SKT to launch a sovereign and secure AI infrastructure that offers speed and flexibility at scale.”

SKT has been driving the use of AI across its business, from delivering AI-powered services to improving customer experience and expanding its robotics capabilities, through its AI to Everywhere initiative.

The initiative builds on a slew of AI capabilities that SKT has rolled out over the years, including an AI chatbot that uses OpenAI’s GPT models to handle customer requests. The telco has also developed its own Sapeon AI chips which are being used in its own neural processing unit farms, as well as by NHN Cloud, an OpenStack-based cloud service provider, and semiconductor firm SK Hynix’s smart factory.