SK Telecom CEO Yoo Ryu Young-sang, front, visits SK Telecom's new data center in western Seoul, Dec. 23. Courtesy of SK Telecom
By Nam Hyun-woo
SK Telecom will launch a cloud-based subscription service for artificial intelligence (AI) processing to capitalize on the rapidly growing demand for computing power required for AI services, the telecom company said Monday.
SK Telecom announced the official launch of SKT GPU-as-a-Service (GPUaaS) following the completion of pilot operations at its new AI data center in western Seoul, which opened on Dec. 30.
GPUaaS is a service that allows users to rent powerful Graphics Processing Units (GPUs) from data centers via the cloud, instead of purchasing expensive processors. Through the service, subscribers can scale their computing power for various high-performance tasks such as AI, machine learning and big data analysis services, paying only for the resources they actually use.
SK Telecom launched the service in collaboration with Lambda, a U.S.-based company. As part of the service, SK Telecom is hosting Lambda's regional base at its new data center, the company’s first in the Asia-Pacific region.
Founded in 2012, Lambda is a GPUaaS provider that offers AI cloud services to major tech companies. As part of its strategy to become an AI infrastructure provider, SK Telecom formed a strategic investment with Lambda in February last year, securing a reliable supply of GPUs for its cloud services.
Servers at SK Telecom's new data center in Western Seoul / Courtesy of SK Telecom
SK Telecom’s GPUaaS will be based on Nvidia’s H100 GPU, which is the most advanced data center GPU currently available in Korea.
Companies subscribing to SK Telecom’s GPUaaS can customize their subscription package by selecting the number of GPUs, subscription duration and additional options such as dedicated servers, firewalls and private lines based on the scale and purpose of their AI services.
Through the service, clients can perform AI training or inference reliably and enhance service security with firewalls and private lines. Even if their AI workloads increase rapidly, clients can easily add more GPUs, reducing the need for additional infrastructure investment, SK Telecom said.
The company did not disclose specific price ranges but stated that subscription fees would be determined based on the number of GPUs and the subscription duration.
SK Telecom reported that over 100 companies have already expressed interest in GPUaaS, inquiring about the service's details. In addition to corporate clients, the company said universities and research institutions are also showing strong interest.
Citing client research, SK Telecom said clients expressed expectations of accessing high-performance GPUs for faster AI model training, renting GPUs for shorter periods of time and saving costs associated with purchasing servers or establishing data centers.
SK Telecom said it plans to introduce its latest H200 GPUs within the first quarter of this year, which will also be the first among Korean data centers.
“The launch of GPUaaS marks the first step in bringing our AI data center business closer to customers, and it holds significant meaning in establishing SK Telecom as a key player in AI infrastructure,” an SK Telecom official said.
“Building upon this service, SK Telecom will establish a GPU farm that enhances national AI competitiveness.”