[Weekpick Cloud] Government-Supplied GPUs Running at Less Than 50% Capacity... AIDC Could Generate Up to KRW 3.5 Trillion in Output Effect
IT DAILY ·
✦ AI Summary
Of the 620 GPUs allocated under the government's high-performance computing support program, 443 were running at less than 50% capacity in 2025, and 350 of 478 were below 50% in 2026.
The Ministry of Science and ICT said the low utilization was temporary and due to the nature of the research projects, but on-site observers pointed to a structural shortage of operating personnel for distributed learning design and job scheduling.
The Ministry of the Interior and Safety inspected the Daejeon headquarters of the National Information Resources Service, checking battery removal measures and recovery progress, and said it plans to complete disaster recovery (DR) systems for 97 major systems by 2028.
[Weekpick Cloud] summed up major cloud news from the past week. More than 7 out of 10 rented GPUs allocated under the government's high-performance computing support program were found to be running at less than 50% capacity. This raised the need to review the resource allocation system, as well as the operational capabilities of academia, industry, and research institutions.
Separately, the Ministry of the Interior and Safety visited the Daejeon headquarters of the National Information Resources Service. At the site, the ministry confirmed battery removal measures inside the computer room.
The ministry announced plans to complete disaster recovery (DR) systems for 97 major systems by 2028. The announcement was summarized as laying out the direction for expanding disaster recovery systems, along with the confirmation of safety measures in the computer room.
According to an analysis by the National Assembly Research Service, the production-inducing effect across industry of a 100-megawatt (MW) AI data center was estimated at up to about KRW 3.5 trillion. The report presented figures for the broader industrial ripple effects of a 100-megawatt (MW) AI data center and highlighted the economic significance of AIDC.
A large share of the GPUs allocated under the government's high-performance computing support program were found to be underutilized. According to data analyzed by Rep. Kim Woo-young from the Ministry of Science and ICT and the National IT Industry Promotion Agency (NIPA), 443 of the 620 GPUs allocated in 2025 had utilization rates below 50%, and the share below 50% in 2025 was 71.5%. In 2026, 350 of 478 allocated GPUs were also below 50% utilization, and the share below 50% that year was 73.2%.
The ministry said the GPUs were temporarily idle due to the nature of the research projects and were not dormant resources. However, at the site, the lack of specialized operating personnel for tasks such as distributed learning design and job scheduling was cited as a structural cause.
It was also found that the allocated domestic neural processing units (NPU) had seen little practical use in commercial inference services. Accordingly, calls grew for assigning domestically produced NPUs to suitable areas such as inference, and for a flexible resource reallocation management system tailored to performance and cost efficiency. The issue is expected to come up at the National Assembly audit.
The Ministry of the Interior and Safety conducted a comprehensive inspection of the recovery situation and digital safety management system at the Daejeon headquarters of the National Information Resources Service. Minister of the Interior and Safety Yun Ho-jung personally took part in the inspection by visiting the site.
Minister Yun checked whether the fire-damaged computer room had been restored in phases and confirmed that the computer room was operating normally. He also reviewed the 24-hour monitoring system for immediate response in case of an emergency.
Given the planned closure of the Daejeon center in 2030, the ministry focused on progress in building a real-time data replication system between the Daejeon center and the Gongju backup center. The government this year finalized a plan to build disaster recovery systems for 97 major administrative information systems at the Daejeon center.
The government plans to gradually complete the DR systems by 2028. The goal is to prevent interruptions in public administrative services even in disaster situations. Meanwhile, a 100MW AI data center was presented as having up to KRW 3.5 trillion in production effects.
A written response submitted by the National Assembly Research Service to Rep. Hwang Jeong-a's office presented the economic ripple effects of one 100MW AI data center in terms of production, added value, employment, and tax revenue. According to the response, the estimated industrial production effect of one 100MW AI data center ranged from about KRW 500 billion to up to KRW 3.5 trillion. It also highlighted effects on employment and tax revenue.
The response cited the case of a 105MW center construction project in Busan and explained that the production-inducing effect from the build investment was large. The Busan case involved an investment project worth KRW 1.8 trillion, with an expected production-inducing effect of KRW 3.492 trillion and an expected added value of KRW 1.332 trillion. It also referred to an analysis based on the Bank of Korea's input-output table.
The response also estimated that if the supply of 100MW-class infrastructure is delayed, production in related industries would fall by up to KRW 3.06 trillion. It also said that hiring one person for permanent operations at a data center would create an additional 2.5 to 4.5 jobs in other industries. Tax revenue related to the U.S. data center industry reached USD 204.4 billion in 2024, up 24% from the previous year.
Source: IT DAILY · Kwon Young-seok
Original: https://www.itdaily.kr/news/articleView.html?idxno=242025
References
This article was produced with the help of an automated content generation algorithm.
Source: IT DAILY
View originalThis article was summarized and organized by BizCrush based on the original article from IT DAILY. For exact quotations and full details, please refer to the original article.