NHN launched a 7,656-GPU cluster in Seoul, South Korea, according to Data Center Dynamics. The cluster is housed in a Yangpyeong-dong data center and targets large-scale AI workloads.
Key facts
- 7,656 GPUs in a single cluster
- Located in Yangpyeong-dong, Seoul
- No GPU model or investment disclosed
- Targets domestic enterprise AI workloads
- NHN competes with Naver and Kakao in cloud
NHN has deployed a 7,656-GPU cluster in Seoul, South Korea, according to Data Center Dynamics. The cluster is located in a data center in the Yangpyeong-dong district, a known tech hub in the capital. The company did not disclose the GPU model, total investment, or power capacity for the facility.
This deployment adds to South Korea's growing AI infrastructure race. Naver, Kakao, and major telcos have all announced similar clusters in the past year as domestic demand for LLM training and inference scales up. NHN, primarily known for its cloud services and webtoon platform, is positioning itself to capture enterprise AI workloads from financial services, gaming, and e-commerce clients.
Unique take: NHN’s cluster signals a shift from hyperscaler dependency
Unlike most Asian AI clusters that are built by AWS, Google Cloud, or Azure, NHN’s deployment is fully owned and operated by a domestic company. This reflects a broader trend of regional cloud providers building their own GPU fleets to avoid hyperscaler lock-in and data sovereignty concerns. South Korea’s strict data localization laws make this particularly relevant: financial and healthcare customers increasingly require on-premises or domestic cloud inference.
The cluster's scale — 7,656 GPUs — is modest compared to the 100,000-GPU superclusters from Meta or Tesla, but it is significant for a regional player. For comparison, Naver’s hyperscale AI cluster in Chuncheon reportedly houses 20,000+ GPUs. NHN’s cluster likely targets inference workloads rather than frontier model training, given the smaller scale and lack of announced training partnerships.
Competitive landscape
NHN competes directly with Naver Cloud and Kakao i Cloud in the domestic market. Naver has invested heavily in its own LLM, HyperCLOVA X, and offers GPU-as-a-service for startups. Kakao has partnered with local chip designers like Rebellions for inference acceleration. NHN’s cluster gives it a differentiated offering for customers who want dedicated GPU capacity without going to a hyperscaler.
The company has not disclosed a timeline for when the cluster will be fully operational or whether it will be used for internal AI products (e.g., NHN’s own AI assistant or content generation tools) or resold as cloud compute.
What to watch
Watch for NHN to announce GPU model details and customer commitments. If the cluster uses Nvidia H100 or B200 GPUs, it signals a long-term commitment to Nvidia's roadmap. Also track whether NHN launches an LLM-as-a-service product tied to this cluster.
