Blog

DEEP RESEARCH · SK TELECOM AI INFRASTRUCTURE

[SK Telecom] Korea’s Leading Data-Center and AI-Related Company

A comparison of Korean big-tech and telecom companies across H200, GH200, H100 adoption and AI data-center strategy

Date: 2025-02-21 · AI data-center competitiveness comparison · Naver Blog

Investment decisions are your own responsibility. This material is research and is not a buy or sell recommendation.

0. Bottom line first

The more I look, the more SKT seems like the Korean company most serious about AI. Rebellions, early H200 adoption, and the Lambda investment all suggest it is allocating operating cash most aggressively into AI.

Chip

Rebellions investment

This is SK Telecom's AI semiconductor investment axis, and the source links a separate analysis post.

GPU

Lambda partnership

The investment and partnership with a U.S. GPU cloud operator is presented as part of the background for securing H200 supply.

AIDC

Gasan AI data center

SKT starts with H100-based GPUaaS, then targets H200 and later Blackwell-family products.

1. SKT AI investment links grouped in the source

The post links three internal notes for understanding SKT's AI strategy.

SK Telecom Rebellions investment linked post thumbnail SK Telecom Lambda partnership linked post thumbnail Lambda company analysis linked post thumbnail

2. Korean H200 and AI data-center comparison

CompanyGPU adoptionData center and scaleInvestment view
NaverNo official H200 adoption announcement in the source. It uses existing-generation GPUs, including 2,240 A100 Tensor Core GPUs in Sejong.Gak Sejong and Gak Chuncheon support HyperCLOVA X and Naver Cloud AI services.Naver is pursuing alternatives with Samsung Electronics, including an AI accelerator shown in 2023 with more than 8x power efficiency versus Nvidia GPUs.
KakaoUses GH200 rather than pure H200. GH200 is described as having 1.7x the memory capacity and 1.5x the bandwidth of H100.Ansan data center completed at end-2023 and started in Q1 2024. It has 4,000 racks and capacity for up to 120,000 servers.GPU equipment investment was about KRW 55 billion in 2023, with total GPU-related investment planned at KRW 100 billion through 2024. 2025 GPU investment is expected around the KRW 50 billion range.
SK TelecomPresented as the fastest Korean company to bring in H200, planned for Q1 2025 around March.Opened a dedicated AI data center in Gasan, Seoul, in January 2024. Rack power density is 44kW, about 9x the general IDC average of 4.8kW.Secured H200 through Lambda partnership. Invested USD 20 million in Lambda in 2023. Plans KRW 3.4 trillion in AIDC investment through 2029.
KTPlans H200 supply through Microsoft cooperation. Exact timing and quantity are not disclosed.Provides A100 and H100-based AI training infrastructure through Daedeok AIDC and AI Train.Plans about KRW 2.4 trillion in AI investment with Microsoft through 2029. Nationwide IDC and telecom edge infrastructure are its differentiators.

3. Why I rate SKT higher

Official fact: The source says SKT plans Korea's first H200 introduction in Q1 2025, secured supply through Lambda, started Gasan AIDC with H100-based GPUaaS, and plans to expand GPUs to the thousands within three years.

Interpretation: I rate SKT highly because it is allocating telecom cash flow into AI infrastructure, AI chips, and GPU cloud partnerships at the same time. It is not only announcing AI services; it is connecting GPU supply, data centers, and a cloud product.

SKT AI infrastructure stackInvestment and partnership turn into a data-center product
Cash flowTelecom core
PartnersLambda, Rebellions
InfrastructureGasan AIDC, H100/H200
ServiceGPUaaS, AI cloud
SKT's difference is building AI as infrastructure, not just as an app story

4. Risks and checkpoints

  • Whether the H200 introduction schedule proceeds as planned.
  • How quickly Gasan AIDC GPUaaS customer interest converts into revenue.
  • How the KRW 3.4 trillion AIDC investment plan through 2029 affects telecom cash flow and dividends.
  • Whether SKT can actually secure Blackwell-based B100 and later GPUs early.