• Major Asian digital service provider (DSP) is positioning itself in South Korea’s burgeoning enterprise cloud market
• Its ambitious recipe – to manage enterprise cloud applications for customers at scale – is going to require a huge dollop of AI
• Can SK Telecom conquer the complexity challenge?

SK Telecom (SKT) is pushing forward with an AI-enabled strategy it hopes will position it as a key player in Korea’s enterprise cloud computing market. So what’s the overall plan?

SKT is building its cloud capability from the ground up with a national arrangement of what it claims will be ‘stable’ AI datacentres. Using that infrastructure foundation, it’s constructing large-scale graphics processing unit (GPU) clusters, which it intends to scale up as corporate demand for cloud applications builds which, at this point, looks like a safe bet given that generative AI (GenAI) enthusiasm shows few signs of a “bubble-like” implosion.

That means that whatever happens with Korean corporate preferences for cloud delivery and the business models with which they would prefer to engage, SKT is off to a good start having installed a robust underlying cloud platform.

The decision then became one that faces all telcos/DSPs with cloud service ambitions: Where do you position yourself in the cloud stack and what cloud application delivery model do you select?

SKT has calculated that its best approach, given the way its target market appears to be headed, is to provide cloud capacity for corporates to run their own or software partners’ applications and it would, therefore, engage with partners of its own to provide or sell applications on to corporate users. To do that it would need a way of onboarding and charging enterprise customers for platform resources and for managing all the issues that go along with that. This is the big challenge and doing it at scale (multiple enterprises with multiple applications) is where it expects its AI investment is going to prove its worth. 

Enter SKT’s Enterprise AI Cloud Manager 

The really important element in SKT’s cloud strategy is the recent launch of the SKT Enterprise AI Cloud Manager (announced here in Korean).

According to SK Telecom CEO Ryu Young-sang the AI Cloud Manager is designed to commercialise the management and operational know-how of SKT’s large-scale GPU cluster resources. It works with SKT’s AI Job Scheduler software, organising the GPU resources contracted by its enterprise customers and managing them as if they were onboarded “to a single computer”.

A workable analogy is to see the AI Cloud Manager as the equivalent of a sophisticated computer operating system, able to multi-task and dynamically and equitably share out the underlying memory and other resources to the many cloud applications each customer enterprise will (hopefully) eventually be operating across the thousands of GPUs.

Once the system is up and efficiently allocating GPU time and other resources for each customer’s growing and constantly changing collection of cloud applications, it can use ‘spare’ cycles to optimise the applications, minimise electricity consumption and shorten the learning time required for AI development, according to the company.

In addition the AI Cloud Manager will systematically manage the entire process of AI development, claims SKT, including: Data preprocessing and storage/management; model development/learning; model distribution; and model inference.

It will also provide an ‘MLOps’ (machine learning operations) environment so that all the processes may be organically linked.

SKT has high expectations for its approach. According to Kim Myeong-guk, head of SKT Cloud CO, with AI Cloud Manager, SKT “will lead the AI ​​datacentre solution business sector [in South Korea] by stably supplying AI datacentres and GPU servers and also providing essential solutions to manage them.”

 – Ian Scales, Contributing Editor, TelecomTV