SK Telecom launches one of Korea's largest GPU clusters

SK Telecom has launched a sovereign GPU-as-a-service (GPUaaS) cluster named Haein, using over 1,000 Nvidia B200 GPUs.

It’s one of the largest in South Korea and will be a core part of the company’s Gasan AI data center infrastructure.

The cluster was selected by the Ministry of Science and ICT for the Proprietary AI Foundation Model Project to develop Korea’s own AI foundation models.

SK Telecom collaborated with Penguin Solutions for integrated AIDC solutions and Supermicro for server procurement. It uses its proprietary Petasus AI Cloud and AI Cloud Manager platform to provide flexible GPU allocation and service management.

This move follows the opening of Lambda’s Korea region in December 2024 and supports SK Telecom’s broader AI infrastructure expansion.

🔗 Source: SK Telecom

🧠 Food for thought

1️⃣ South Korea’s AI sovereignty drive creates massive infrastructure opportunity

SK Telecom’s GPUaaS launch directly benefits from South Korea’s significant government commitment to AI independence.

The company was selected as one of five elite teams for the “Independent AI Foundation Model” project, which provides 200 billion won in government support specifically for developing homegrown AI capabilities1.

This timing aligns with South Korea’s enactment of comprehensive AI legislation through its Framework Act in January 2025, signaling a serious national commitment to AI leadership2.

The government’s “Digital New Deal” initiative is driving South Korea’s IT services market from $28.3 billion in 2023 toward a projected $61.8 billion by 20303.

SK Telecom’s consortium brings substantial research credibility to this national effort, having published over 800 research papers and holding 736 patents4.

2️⃣ Strategic infrastructure positioning in rapidly expanding market

By securing a role in the national AI program while launching commercial GPUaaS, SK Telecom is positioning itself as a key player in South Korea’s AI infrastructure.

The competitive landscape shows significant pricing pressure, with providers like Dataoorts charging $2.28 per hour for NVIDIA H100 GPUs and $1.62 for A100 models5.