The key battleground in the competition for artificial intelligence (AI) data centers is shifting from “land” to “power.” Simply securing large sites and stacking servers is no longer enough to take the lead in the AI infrastructure race. As the use of graphics processing units (GPUs) required for AI model training and inference increases, power consumption by data centers is also surging. The core of the AI data center (AIDC) business has now become how quickly, reliably, and cheaply operators can secure electricity.

● AIDC boom triggers big deals in the U.S. power sector

On 29 June (local time), global consulting group Deloitte reported that mergers and acquisitions (M&A) in the U.S. power and utilities sector reached USD 203.6 billion (approximately KRW 316 trillion) as of May this year, marking an all-time high. This is more than 40% higher than last year’s total transaction volume of USD 141.7 billion (KRW 220 trillion).

The surge in power company M&A is driven by the enormous electricity consumption of AIDCs. Training and operating generative AI services such as ChatGPT requires large-scale GPU servers, which in turn require power. As a result, big tech companies are not just building data centers but are also moving to secure power plants, transmission networks, and even power companies. Competition in data centers has become a contest over “who can secure the electricity to run the servers.”

The race to secure power has emerged as a “bottleneck” for the global AI industry. In the United States, power demand from data centers is lengthening grid connection waiting times in some regions. Debate is also intensifying over who should bear the burden of higher electricity costs.

According to Reuters on 1 July, PJM, the largest power grid operator in the United States, is discussing additional power procurement and demand management measures as rising data center demand threatens the balance of power supply and demand. PJM stated that over the past two years, the supply-demand balance on the grid across 13 U.S. states has been shaken. Capacity prices in this area have surged more than 1,000% since 2024.

● Korean government also elevates AIDC to a national strategic industry

The same problem is expected to emerge in Korea. For this reason, the government’s recently announced AIDC strategy also treats power supply feasibility as a key prerequisite for AIDC deployment. To mitigate potential burdens on transmission networks if data centers concentrate in the Greater Seoul area, the government plans to distribute AIDCs mainly across non-capital regions with available power capacity.

SK is promoting an AIDC project of 5GW centered on Ulsan, GS is pursuing 2.4GW in Donghae, Gangwon Province, and Naver is developing 1GW in Sejong. Construction is scheduled to begin in 2028, with phased operations starting in 2029. The capacity will then be expanded to 15GW by 2035, completing an infrastructure system totaling 18.4GW. The government’s strategy is to respond to the sharp increase in AI computing demand while securing competitiveness in domestic AI infrastructure.

Lee Yu-soo, professor of economics at Soongsil University, said, “As AI becomes more advanced, power demand is bound to increase in tandem,” adding, “AIDCs must be built in regions where electricity can be supplied in order to secure industrial competitiveness.”

Popular News