SKT Launches Subscription-Based AI Cloud Service Using NVIDIA GPU

Pilot Operation of Gasan AIDC Completed... Launch of GPUaaS
Deployment of NVIDIA H100, the Highest-Performing GPU in Korea
Over a Year of Preparation with Lambda... Pricing Tailored to Corporate Needs

SK Telecom has launched a subscription-based artificial intelligence (AI) cloud service based on NVIDIA's graphics processing units (GPUs) through over a year of collaboration with the US company Lambda. This is expected to provide a reasonable option for companies that want to utilize AI but have low GPU accessibility due to high costs and other factors.

On December 23 last year, SKT CEO Yoo Young-sang visited SK Broadband Gasan IDC to check the preparation status of the subscription-based GPU service (GPUaaS). Photo by SKT

On the 13th, SK Telecom announced that it officially launched the subscription-based AI cloud service (SKT GPUaaS) after completing the pilot operation of the Gasan Artificial Intelligence Data Center (AIDC) on December 30 last year. The service is currently provided based on NVIDIA's 'H100', the highest-performing GPU introduced in Korea.

At the 'SK AI Summit 2024' held in November last year, SKT announced the 'AI Infrastructure Super Highway' strategy to build AI infrastructure nationwide centered on three pillars: ▲AIDC ▲GPUaaS ▲Edge AI. Subsequently, through a year-end organizational restructuring, SKT established the 'AIDC Business Division' as a separate organization and officially began the AI infrastructure business by newly launching GPUaaS.

The Gasan AIDC offers a data colocation environment optimized for high-density GPU server operation, with a power density per rack of 44 kW, which is about nine times the domestic data center average power density of 4.8 kW per rack, the highest level in Korea. (Data colocation is a service where specialized companies lease space such as computer rooms and manage and operate customer equipment on their behalf.)

SKT has been preparing to launch 'SKT GPUaaS' for over a year in collaboration with the US GPUaaS company Lambda. The Gasan AIDC is Lambda's first region (a geographic location where a cloud provider operates a data center) in the Asia-Pacific region. As the first global move to actively promote the AIDC business, SKT invested $20 million in Lambda in February last year.

Artificial Intelligence Data Center (AIDC) built in SK Broadband's Gasan Internet Data Center (IDC) / Photo by SKT

Companies wishing to subscribe to 'SKT GPUaaS' can select the number of GPUs and the duration according to the scale and purpose of their AI service and configure customized packages such as dedicated servers, firewalls, and dedicated lines. Prices are determined according to the selected options, and short-term service use of 1 to 2 months is also possible.

Clients can perform AI training and inference based on dedicated GPUs, and security can be enhanced by utilizing firewalls and dedicated lines. Additionally, even if AI workloads increase sharply, GPUs can be added to dedicated servers, reducing the burden of additional facility investment.

It is also possible to interoperate with shared cloud services. This means that clients can use SKT's GPU computing power while continuing to use their existing cloud services.

When used together with SKT's AI Cloud Manager, GPU resources can be efficiently managed. This solution manages numerous GPU resources as if they were a single computer, maximizing performance and supporting the reduction of training time required for AI development.

According to SKT, more than 100 companies inquired about specific details even before the official launch of GPUaaS, and universities and research institutes conducting AI research have also shown interest. Companies cited reasons for wanting to use the service as ▲shortening AI model training time by utilizing high-performance GPUs ▲the ability to use GPUs for a short period at a reasonable price ▲immediate use of GPU resources without concerns about server purchase and data center operation.

SKT plans to accelerate the expansion of GPUaaS customers by introducing the latest GPU, H200, for the first time in Korea in the first quarter of this year.

Kim Myung-guk, head of SKT GPUaaS Business Division, said, "This GPUaaS launch is the first case of the AIDC business approaching customers," adding, "We will build a GPU farm that enhances the nation's AI competitiveness."