NHN Cloud topped Korea's TOP500 supercomputer list using its FactoryX GPU clusters. The achievement marks the first time a domestic cloud provider leads the national HPC ranking.
Key facts
- NHN Cloud ranks first in Korean TOP500.
- FactoryX clusters deliver 1.2 exaflops peak.
- System uses NVIDIA H100 GPUs with InfiniBand.
- First Korean cloud provider to top domestic list.
- Previous leader KISTI Nurion achieved 25.7 petaflops.
NHN Cloud has secured the number one position on Korea's TOP500 supercomputer list, powered by its FactoryX GPU clusters. According to ChosunBiz, the clusters leverage NVIDIA H100 GPUs interconnected via InfiniBand, achieving 1.2 exaflops of peak performance. This surpasses previous leaders like the KISTI Nurion system, which topped out at 25.7 petaflops.
The FactoryX Architecture
FactoryX is NHN Cloud's proprietary GPU cluster design optimized for large-scale AI training workloads. The system uses a three-tier network topology: leaf switches connect GPU nodes within racks, spine switches aggregate across racks, and a super-spine layer handles inter-cluster traffic. This design minimizes latency for distributed training across thousands of GPUs. NHN Cloud did not disclose the exact node count or total GPU deployment but confirmed the cluster is fully operational for customer workloads.
Implications for Korea's AI Ecosystem
Korea has been aggressively investing in AI infrastructure, with government initiatives like the "AI Gwangju" project and private sector builds from Naver and Kakao. NHN Cloud's TOP500 leadership signals that domestic cloud providers can compete with global hyperscalers like Google Cloud and AWS for HPC workloads. The ranking also puts pressure on Korean semiconductor firms like Samsung to develop competitive HBM memory solutions for GPU clusters, as NVIDIA H100s currently dominate the market.
Comparison to Global Standards
While 1.2 exaflops places NHN Cloud at the top of Korea's list, it remains modest compared to global leaders. The current world number one, Frontier at Oak Ridge National Laboratory, delivers 1.2 exaflops on the HPL benchmark—roughly equivalent to NHN's peak. However, NHN's system is likely optimized for AI training (mixed precision) rather than traditional HPL, meaning real-world AI performance could be higher. The company did not release HPL scores, making direct comparison difficult.
What to watch
NHN Cloud's next TOP500 submission will reveal whether FactoryX scales beyond 1.2 exaflops. Watch for customer adoption metrics from Korean AI labs and potential expansion into Japan or Southeast Asia markets. The Q3 2026 TOP500 list will show if NHN retains the lead against Naver Cloud's emerging clusters.
Source: news.google.com
