As securing grid access emerges as the biggest challenge in building global artificial intelligence (AI) data center infrastructure, the refining and petrochemical industry, with its extensive unused land and on-site generation facilities, is emerging as a potential source of solutions.
According to industry sources on October 28, large AI data center (AIDC) projects in South Korea and overseas are facing development difficulties because of transmission grid connection delays and insufficient power supply capacity.
Power-intensive AI data centers increase interest in privately developed infrastructure
Intensifying global competition in AI models has driven a surge in demand for power-intensive inference computing infrastructure. Unlike conventional data centers, AIDCs consume far more than 100 kW per server rack, with power requirements rising by dozens of times, making them difficult to accommodate on existing transmission and distribution grids.
The key constraint on infrastructure expansion has shifted from securing GPUs to securing electricity. In the United States, data center buildings themselves can be completed in 18 to 36 months, but connecting generation facilities to the grid and securing transmission infrastructure can take 4 to 5 years or longer in systems such as the Texas grid operated by ERCOT. These grid delays are holding up development.
As national grid expansion encounters physical constraints, a private AI utility business model in which private capital directly builds generation and transmission infrastructure has emerged as an alternative. In the United States, the Stargate project, led by SoftBank and OpenAI, was launched with plans to invest up to $500 billion over the next four years to develop power and data centers together, demonstrating this trend toward integrated infrastructure.
SK and GS combine unused industrial land with on-site power plants
Against this backdrop, the asset value of South Korean refining and petrochemical companies that secured large industrial sites and combined heat and power or liquefied natural gas (LNG) power plants in advance is being reassessed. Physical infrastructure capable of providing both extensive land and large amounts of electricity has become a prerequisite for attracting major AIDCs.
SK Group, for example, has stepped up its business through the SK Hyper platform, which combines telecommunications and power. SK has begun construction of the 100 MW first phase of an AIDC on unused SK Chemicals land of approximately 20,000 pyeong in the Ulsan Mipo National Industrial Complex.
The site offers more than space, with substantial opportunities to link infrastructure. It is regarded as a location that could bypass grid shortages because electricity can be reliably procured through connections to nearby Ulsan GPS, SK Gas's 1.2 GW LNG power plant, and SK Multi Utility's combined heat and power plant.
GS Group has also established GS AI Infrastructure, a large infrastructure development platform, and begun putting its land to use. It is pursuing a large AIDC with total capacity of 2.4 GW by 2029 on approximately 180,000 pyeong of its own land, whose preparation was completed in 2022, in the Bukpyeong No. 2 General Industrial Complex in Donghae, Gangwon Province.
Power and cable industries anticipate extra-high-voltage equipment and BESS demand
As power-intensive infrastructure expands around privately owned sites, expectations are also growing in supporting industries, including power equipment and cables. This reflects government efforts to introduce differentiated electricity rates by region to disperse data center demand concentrated in the Seoul metropolitan area and accelerate physical grid expansion, including the west coast energy highway.
In particular, as the introduction of a system allowing private participation in national backbone grid construction approaches, more private companies are expected to build substations and connection lines directly. Demand for orders for very large transformers linking AIDCs and power plants, as well as extra-high-voltage transmission cables rated at 345 kV or above, is therefore expected to increase substantially over the medium to long term.
The unstable power loads characteristic of AI data centers are also opening a new market. Very large battery energy storage systems (BESS) are essential to buffer the rapid power fluctuations that occur when tens of thousands of GPUs start or stop simultaneously.
A refining industry representative said, "Success in the global competition for AI leadership depends on how promptly large amounts of electricity, ranging from hundreds of megawatts (MW) to gigawatts (GW), can be supplied. The industrial land and on-site generation infrastructure that the refining and petrochemical industry built in advance will provide an important foundation for easing power constraints."
Original Korean article: ebn.co.kr
English brief by Korea Wind Intelligence, koreawindintel.com
