Software

[WeekPick Cloud] Data Center

IT DAILY ·

[Photo: Pixabay]

✦ AI Summary

There are growing concerns that infrastructure investment could be disrupted as it takes as long as more than 2 years for KEPCO's advance review notice on power supply during domestic data center construction.

A review of 154 KEPCO data cases by Rep. Eo Gi-gu found that 57 cases took more than 6 months to receive a reply, 13 took more than 1 year, and 4 took more than 2 years.

The share of applications in the Seoul metropolitan area last year was 80.4%, and all 3 cases of power supply being unavailable were in Gyeonggi Province.

There are growing concerns that infrastructure investment could be disrupted as it takes as long as more than 2 years for Korea Electric Power Corporation's advance review notice on power supply during domestic data center construction. In the process of building data centers, which are regarded as a key element of AI infrastructure, cases in which it took more than 6 months to confirm whether Korea Electric Power Corporation could supply power accounted for 4 out of 10 sites, while concentration in the Seoul metropolitan area is also deepening.

A review of KEPCO data by Rep. Eo Gi-gu showed that the analysis covered 154 responses received between 2021 and April 2026. Of those, 57 cases took more than 6 months to receive a reply, putting the share that took at least 6 months at 37.0%. There were 13 cases that took more than 1 year and 4 cases that took more than 2 years. There were also 7 cases still under review, and all 7 were found to have been waiting for more than 6 months. Rep. Eo's office said the analysis shows that advance review for data center power supply is taking a long time.

Overseas, Google is pushing ahead with launching its in-house AI chip, TPU, into orbit and has also begun a demonstration of a next-generation space data center. In contrast to the situation in Korea, where it takes a long time to confirm whether power can be supplied for new data centers, overseas there is also a growing trend of technological demonstrations around data centers and AI chips.

In this regard, the Export-Import Bank of Korea forecast that the AI chip market for data centers will grow to KRW 1,200 trillion by 2030. The bank also suggested that government support is needed for the early commercialization and market establishment of domestic NPU companies.

If review of power supply plans is delayed, operators cannot finalize data center designs, cannot secure project financing (PF) loans from the financial sector, and cannot finalize orders for key equipment such as graphics processing units (GPU). As a result, delays in reviewing power supply plans disrupt the overall business process.

Along with this situation, the share of applications in the Seoul metropolitan area last year was 80.4%. This was a sharp increase from 32.0% the previous year, indicating that concentration in the Seoul metropolitan area has intensified.

Among the 154 responses, there were 3 cases of 'power supply unavailable' due to insufficient supply capacity at nearby substations. This accounted for 1.9% of the total, and all 3 cases of 'power supply unavailable' were in Gyeonggi Province.

Meanwhile, Google is set to launch its Tensor Processing Unit (TPU) into space on October 1. The purpose of the launch is research on next-generation space data centers, and it will use the Google-Planet jointly developed experimental satellite 'MVP.' The 'MVP' will carry 4 sixth-generation AI chips, 'Trillium' TPUs, and the launch vehicle will be a SpaceX Falcon 9 rocket, with the mission named Transporter-18 and the launch orbit set in low Earth orbit. This launch is being pursued as the first orbital test of 'Project Suncatcher,' unveiled last year.

The space data center is being pursued as a way to overcome the limits of ground-based data center power supply and demand. Power will be secured through solar generation in orbit, with output up to 8 times that of on Earth. The test satellite will generate about 1 kW of power and operate the TPU on a 15-minute cycle. The satellite is in the stage of verifying Google AI model 'Gemini' computation, radiation durability, and heat dissipation performance in a vacuum for about 1 year.

Google expects that if launch costs fall in the mid-2030s, operating costs for space data centers and ground-based data centers will become comparable. It also plans to launch 2 additional satellites equipped with laser ultrahigh-speed optical communications next year.

The AI chip market for data centers is projected to grow from USD 124 billion (about KRW 170 trillion) in 2024 to USD 860 billion (about KRW 1,200 trillion) in 2030. The projected average annual growth rate is 38%. These forecasts are included in a report by the Overseas Economic Research Institute of the Export-Import Bank of Korea.

NVIDIA currently holds a 78.2% market share, making the market structure one of NVIDIA dominance. However, as the center of gravity in the generative AI market shifts from training to inference, entry opportunities are expanding for latecomers in neural processing units (NPU) that lower prices and power consumption.

Companies targeting the domestic inference market include FuriosaAI and Rebellions. While the technological capabilities of FuriosaAI, Rebellions and others have been proven, domestic companies are assessed to lag global leading companies by more than 3 years. This assessment of the gap was presented based on financial strength, the software ecosystem, and actual data center commercialization references.

The report projected the next 3 to 5 years as a turning point when the global supply chain structure will harden. Accordingly, it called for tax benefits and subsidy support for companies adopting domestic AI chips. It also called for stronger intergovernmental cooperation to advance into overseas data center supply chains, including in the Middle East.

The related event is hosted by the Korea Artificial Intelligence Cloud Industry Association (KACI). The event, a specialized technical conference titled 'The 3rd AI-Cloud Big Tech 2026,' will be held on October 1 at the Grand Hall of the L Tower in Yangjae, Seoul.

The theme of this year's event is 'AI-Cloud Opened by Advance, Connect, Transform.' The event will take a look at the AI infrastructure~platform/service technology ecosystem and cover the latest changes in industrial settings.

The event is designed to present in one place the full AI technology cycle spanning AI infrastructure, including AI data centers (AIDC) and GPUs, AI-PaaS, AIOps, SaaS, work environments, and security. With about 300 attendees, including government and related agency officials, local governments and public institution officials, and corporate representatives, technical lectures as well as exhibition and promotional booths by participating companies will be set up, through which practical introduction plans and cooperation measures are expected to be discussed.

Source: IT DAILY · Kwon Young-seok
Original: https://www.itdaily.kr/news/articleView.html?idxno=241861

References

This article was produced with the help of an automated content generation algorithm.


Source: IT DAILY

View original

This 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.