South Korea's 614-Petaflop Supercomputer Launches This December With Shared Access Model

South Korea is preparing to activate a supercomputer with a theoretical capacity of 614 quadrillion floating-point operations per second, drawing interest from researchers, academic institutions, and businesses seeking access to its resources. Hangang, designated as the nation's sixth national supercomputer, will begin official operations in December and incorporate 8,496 GPUs, 9,936 CPUs, 205 petabytes of storage capacity, and network connectivity exceeding 400 gigabits per second.

The government intends to distribute computing capacity among national initiatives, academic researchers, universities, and commercial enterprises, creating an alternative for Korean organizations to gain access to advanced computing infrastructure without purchasing their own GPU hardware or depending on commercial cloud platforms. The bulk of Hangang's resources have already been reserved for research activities and government-backed projects.

Hangang packs 8,496 GPUs and 205 PB of storage

According to Seoul Economic Daily, Hangang employs Nvidia GH200 accelerators, which integrate Grace CPUs with H100 GPUs to enhance data-processing capabilities. The system's 614-petaflop specification translates to a maximum of 614 quadrillion floating-point operations per second.

In comparison to South Korea's previous fifth national supercomputer, Hangang delivers 24 times greater theoretical computing capacity, 10 times more storage resources, and four times faster network throughput. The Ministry of Science and ICT projects the system will achieve a position within the world's top 10 supercomputers upon becoming operational.

What 614 petaflops can actually be used for

Hangang targets large-scale computational science, data simulation, and artificial intelligence-driven scientific research. The government has designated 55% of the system's capacity for research and development innovation, with an additional 30% reserved for national strategic initiatives, including projects centered on critical technologies.

Per ChosunBiz reporting, extensive research endeavors demanding massive simulations or highly distributed computing can request resources totaling 50 nodes. Standard computational science and AI initiatives can obtain between two and 49 nodes, while a new exploratory category offers single-node access following a streamlined evaluation process. Each node houses four GH200 accelerators.

Emerging projects may begin with a single four-GH200 node configuration, whereas major research initiatives can secure allocations reaching 50 nodes.

What Hangang changes for Korea and APAC

Hangang aims to broaden access to advanced computing capabilities throughout Korea, and its collaborative-use framework may prove applicable to other Asia-Pacific nations.

Korean researchers and companies get another route to high-end compute

Hangang provides Korean researchers and enterprises with a pathway to high-performance computing without requiring them to develop their own equivalent infrastructure. ChosunBiz noted that organizations have previously been forced to procure premium GPUs or turn to commercial cloud solutions, introducing both financial burden and data-protection complications.

The 15% resource allocation designated for industrial participants targets small-scale and startup enterprises that may lack the financial capacity to establish independent high-performance computing environments. KISTI will furnish advisory services and engineering assistance alongside the computational resources themselves.

Pilot-phase applications are being accepted from September 23 through October 8, with chosen participants anticipated to receive access between late October and mid-November. Standard-service applications span September 28 to October 16, with resource assignments commencing in December.

Korea's approach offers an APAC case study

South Korea is consolidating costly GPU infrastructure within national systems and distributing access across public agencies, educational institutions, and private enterprises. This strategy could serve as a reference point for additional Asia-Pacific economies seeking to expand availability of artificial intelligence and scientific computing capabilities without mandating individual organizational hardware procurement.

The infrastructure investment exceeded initial projections significantly. Interesting Engineering reported that Hangang's final expenditure reached 448.3 billion won, equivalent to approximately $334.9 million, representing a 53% overrun from the original budget estimate due to elevated GPU pricing driven by market demand.

The system architecture depends on Nvidia accelerators and infrastructure constructed by Hewlett Packard Enterprise. Although South Korea is expanding its domestic computational resources, the nation remains subject to equipment expenses and availability constraints affecting other regional purchasers.

Pilot operations commence in late October, with comprehensive service deployment scheduled for December. The extent to which researchers and smaller enterprises utilize these resources may demonstrate whether Korea's distributed-computing framework represents a viable approach for other Asia-Pacific regions.