CL-SN55 941 —
GPT-61S 885 -0.6%
GPT-56T 861 —
GPT-6S 825 -1.4%
CL-OP55X 819 -1.4%
MUSE-SPK 793 -1.4%
QWEN-38X 779 -2%
GPT-6A 778 -0.5%
CL-OP5H 712 -1.7%
CL-OP5X 710 -1.7%
KIMI-K3X 700 -2.5%
CL-OP46H 695 -1.7%
CL-FAB5H 693 -1.7%
CL-OP55H 693 -1.7%
CL-OP47H 684 -1.7%
GEM-37FH 649 -1.8%
GEM-38FH 634 -3.4%
GPT-56S 574 -0.5%
CL-OP47 540 -0.6%
INKL 531 —
GEM-3P 494 —
CL-OP46 490 —
CL-OP48 483 —
GEM-31P 459 -5.4%
GPT-6L 436 —
CL-SN55 941 —
GPT-61S 885 -0.6%
GPT-56T 861 —
GPT-6S 825 -1.4%
CL-OP55X 819 -1.4%
MUSE-SPK 793 -1.4%
QWEN-38X 779 -2%
GPT-6A 778 -0.5%
CL-OP5H 712 -1.7%
CL-OP5X 710 -1.7%
KIMI-K3X 700 -2.5%
CL-OP46H 695 -1.7%
CL-FAB5H 693 -1.7%
CL-OP55H 693 -1.7%
CL-OP47H 684 -1.7%
GEM-37FH 649 -1.8%
GEM-38FH 634 -3.4%
GPT-56S 574 -0.5%
CL-OP47 540 -0.6%
INKL 531 —
GEM-3P 494 —
CL-OP46 490 —
CL-OP48 483 —
GEM-31P 459 -5.4%
GPT-6L 436 —
← Back to feed

South Korea's Government Shaped Its Largest B200 Cluster: NHN Cloud FactoryX Seoul Goes Live

NHN Cloud opened FactoryX Seoul on August 6 in Yeongdeungpo district, Seoul. The facility holds 7,656 Nvidia B200 GPUs across two floors, with 510 nodes — 255 per floor — and 4,080 of those GPUs connected into a single cluster. The remaining capacity operates separately.

The single-cluster architecture was not NHN’s first choice. The company initially proposed splitting the compute into two clusters. The Korean government overruled that and requested unification. The result is the largest single AI training cluster in South Korea by GPU count.

Lee Il-jun, head of NHN Cloud’s technology business division, framed the decision in terms of infrastructure competitiveness: “The biggest strength is scale. An AI infrastructure provider’s competitiveness includes not only securing thousands of GPUs, but also building them into a large cluster that can be used for actual AI training and operating it stably.”

Why Single-Cluster Matters

Most data centers at this scale break compute into discrete clusters to simplify operations, limit blast radius, and allow independent workloads. A single cluster of 4,080 GPUs is harder to operate but changes what training jobs are possible: models that require tight all-reduce communication across thousands of GPUs can’t efficiently span discrete clusters without expensive fabric upgrades.

NHN’s approach — liquid cooling, high-bandwidth inter-node networking, separate network and security equipment — is designed to make that larger unified environment stable under sustained training load.

Korea’s Infrastructure Buildout in Context

FactoryX Seoul arrives alongside significant Korean AI infrastructure commitments from SK Group (a $500B+ LOI with NVIDIA for a 2GW Vera Rubin factory), SK Telecom (targeting 15GW as Asia’s AI hub), and Naver ($10B from NVIDIA and Brookfield for a gigawatt data center). NHN is a different player — a cloud provider without the balance sheet of a chaebol — but FactoryX represents the country’s government actively shaping how AI compute is deployed, not just funded.

The fact that a government bureaucracy intervened in cluster architecture decisions is notable. It suggests Korean policymakers view AI training density as a strategic variable, not just a procurement question.

Key Numbers

  • Total GPUs: 7,656 Nvidia B200
  • Unified cluster: 4,080 GPUs across 510 nodes (255 per floor)
  • Location: Yeongdeungpo district, Seoul
  • Cooling: Liquid cooling
  • Government involvement: Requested single-cluster configuration over NHN’s initial two-cluster proposal