SK Telecom CEO Ryu Young-sang (center) inspects the preparation status of GPUaaS at SK Broadband's Gasan IDC. (SK Telecom)
SK Telecom CEO Ryu Young-sang (center) inspects the preparation status of GPUaaS at SK Broadband's Gasan IDC. (SK Telecom)

SK Telecom said Monday that it has launched an artificial intelligence cloud service, SKT GPUaaS, following pilot operations of its new AI data center in Seoul on Dec. 30.

Since unveiling its AI infrastructure superhighway strategy at the SK AI Summit 2024 in November, SK Telecom has accelerated efforts to establish its AI infrastructure business.

This includes forming a dedicated AIDC business unit during a year-end organizational restructuring and the launch of graphics processing units-as-a-service, solidifying its commitment to this new growth area.

The AI infrastructure superhighway strategy focuses on three pillars: AI data center GPUaaS, edge AI and nationwide AI infrastructure development.

The Gasan AI data center offers a colocation environment optimized for high-density GPU server operations, boasting a power density of 44 kilowatts per rack -- nine times higher than the national average of 4.8 kW per rack.

SKT GPUaaS is a subscription-based AI cloud service co-developed with US-based GPUaaS provider Lambda over the past year. As part of this collaboration, Lambda’s first Asia-Pacific data center region was established within the Gasan AIDC.

SK Telecom’s partnership with Lambda enables stable GPU resource acquisition, a critical factor for competing in the AI cloud market.

The newly launched GPUaaS is powered by US tech giant Nvidia’s H100 GPUs, the highest-performing GPUs currently available in Korea.

Business subscribers can customize their package by selecting GPU quantities and subscription periods based on their AI service needs. They can also opt for dedicated servers, firewalls and private lines for enhanced security.

Subscribers gain access to exclusive GPUs for stable AI training and inference. The service’s flexibility allows businesses to scale GPU resources as needed, reducing the financial burden of additional infrastructure investments.

Additionally, SKT GPUaaS is compatible with public cloud platforms, enabling companies to integrate their existing cloud environments with the telecom carrier’s GPU computing power.

By leveraging AI cloud manager, an advanced solution for managing multiple GPUs as a unified resource, businesses can optimize GPU performance and reduce AI training time.

The pricing for GPUaaS is highly adaptable, depending on factors such as subscription duration and GPU quantity.

For example, a subscriber requiring 32 GPUs for 24 months can negotiate a tailored pricing plan. Short-term subscriptions, including one- or two-month options, are also available.

To celebrate the launch, SK Telecom is offering a two-month promotional discount of 20 percent until the end of next month. Other incentives include a GPU trade-in program and consulting services for cloud cost optimization.

According to SK Telecom, the service has received inquiries from over 100 companies, including large corporations, universities and research institutions, even before its official launch.

SK Telecom plans to introduce Nvidia’s latest H200 GPUs in the first quarter, making it the first company in Korea to do so. The move is aimed at enhancing the environment for rapid AI technology and service development, while also expanding the GPUaaS subscriber base.

“The launch of GPUaaS marks a significant milestone as the first direct customer-facing AI data center service,” said Kim Myoung-gook, head of SK Telecom's GPUaaS business. “Our goal is to establish a GPU farm that enhances national AI competitiveness while positioning SK Telecom as a leader in AI infrastructure.”