SK Telecom (CEO Yoo Young-sang) is opening the Gasan AI Data Center based on the NVIDIA GPU H100 and is entering the GPUaaS business in earnest, and is making a full-fledged move into the 'AI infrastructure' market, a key growth engine for AI.

SKT announced on the 13th that it opened the Gasan AI Data Center (hereinafter referred to as AIDC) on December 30th and launched the AI cloud service, 'SKT GPUaaS (GPU-as-a-Service),' after completing a trial operation.

Since announcing its strategy to build an 'AI Infrastructure Super Highway' at the 'SK AI Summit 2024', SKT has been accelerating the full-scale launch of its 'AI Infrastructure' business through a series of actions, such as establishing the 'AI DC Business Division' as a separate organization in the year-end reorganization and launching a new GPUaaS.

The newly opened Gasan AIDC provides a data colocation environment optimized for high-density GPU server operation, with a power density per rack of 44kW, the highest in Korea, which is approximately nine times higher than the average power density per rack of domestic data centers, which is 4.8kW.

'SKT GPUaaS' is a subscription-based AI cloud that has been prepared in collaboration with Lambda, a US company, for the past year. To this end, Lambda's Korean region was attracted to Gasan AIDC, which is Lambda's first region in the Asia Pacific region.

In February, SKT invested in Lambda, a global GPU cloud company, as its first global step toward full-scale promotion of its AI DC business.

Lambda, founded in 2012 by AI engineers, is a GPUaaS company that provides AI cloud services to global big tech companies. SKT plans to fully advance into the AI cloud market based on stable acquisition of graphics processing units (GPUs) through cooperation with Lambda.

The GPUaaS being launched this time is based on the NVIDIA GPU H100. The H100 is the most powerful GPU currently introduced in Korea.

Corporate customers who wish to subscribe to 'SKT GPUaaS' can select the number of GPUs and period according to the scale or purpose of the AI service and configure a customized package including a dedicated server, firewall, and dedicated line.

Customers can use their own GPUs to perform stable AI learning and inference, and security can also be greatly enhanced by utilizing firewalls and dedicated lines.

Additionally, even if the customer's AI workload increases rapidly, GPUs can be easily added to a standalone server, which significantly reduces the burden of additional facility investment.

Additionally, SKT GPUaaS can be interoperable with public (shared) clouds. This allows corporate customers to utilize SKT’s GPU computing power while continuing to use the cloud they are already using.

By utilizing GPUaaS and SKT's 'AI Cloud Manager' together, GPU resources can be managed more efficiently.

'AI Cloud Manager' is a solution that manages numerous GPU resources as if they were a single computer to maximize GPU performance and shorten the learning time required for AI development.

SKT GPUaaS pricing has been flexibly set based on the contract period, number of GPUs, and prepaid billing type, greatly enhancing customer convenience.

For example, if a customer wants 32 GPUs for 24 months, the price will be set accordingly. Short-term services such as 1 month or 2 months are also available.

SKT is also running a special promotion to commemorate the launch of GPUaaS. We are offering a 20% discount on subscription prices for approximately two months until the end of February.

We also plan to implement events such as GPU replacement compensation programs and cloud cost optimization consulting.

For more information, please visit the SKT Enterprise website (https://www.sktenterprise.com).

SKT said that there was a lot of industry interest in the service, with over 100 companies inquiring about specific details even before the official launch of GPUaaS.

In particular, not only large corporations and small and medium-sized enterprises, but also various universities and research institutes researching AI are showing great interest, confirming the market's high expectations for the 'SKT GPUaaS' service.

Companies cited the following as reasons for wanting to use GPUaaS: △Shortening the AI model training period by utilizing high-performance GPUs, △Ability to use GPUs at a reasonable price for a short period of time, and △Use of GPU resources immediately without worrying about purchasing servers and operating data centers.

In addition, SKT plans to introduce the latest GPU, H200, during the first quarter. By introducing H200 for the first time in Korea, we plan to create an environment where domestic companies can develop AI technology and services more quickly, and also accelerate the expansion of GPUaaS customers.

Kim Myeong-guk, head of SKT’s GPUaaS business division, said, “This launch of GPUaaS is the first case of the AI data center business reaching out to customers, and is significant in establishing SK Telecom as a key AI infrastructure operator.” He added, “We will build it as a GPU farm that will enhance the country’s AI competitiveness.”