Regulators and grid operators worldwide are launching audits into exaggerated power consumption claims submitted by data center developers.

With the advent of the artificial intelligence era, power authorities in major countries worldwide face a new challenge: distinguishing genuine power supply applications from those submitted by data center companies. Large data centers require astronomical amounts of electricity, with even a few facilities capable of consuming as much power as an entire city. The problem lies in taking corporate plans at face value. As the AI boom drains grid capacity, companies are rushing to secure spots before finalizing actual investments. This creates inflated “fake demand,” where a single operator applies to multiple candidate sites simultaneously, or multiple operators apply for the same location. Additionally, some centers fail to utilize all the power they initially requested after completion.

This dynamic places global power authorities in a difficult position. Accepting data center requests at face value and building trillions of won worth of power infrastructure risks leaving facilities severely underutilized if plans falter. Conversely, screening applications too strictly could cause nations to fall behind in the competition to attract data centers—a critical component of the AI era.

The South Korean government confronts the same dilemma. On August 20, during the demand reforecast for the 12th Power Supply and Demand Basic Plan, the government prioritized reflecting only the first phase of 10.8GW out of 20.8GW in AI data center investment plans, deferring the remaining 10GW to the next cycle. Consequently, the net maximum power demand for data centers in 2040 rose from 4GW to 11.9GW. Industry sources question the reliability of these figures, noting, “Since there could be a lot of fake demand, it needs to be scrutinized more carefully.”

Countries worldwide are implementing distinct strategies to filter out speculative power demand. In the UK, energy regulator Ofgem is advancing a plan that would require data centers to pay a refundable deposit when securing grid connection rights. The deposit is returned once operations begin but is forfeited if the project is abandoned midway. This follows a surge in grid connection applications from 41GW in November 2024 to 125GW in June of last year—a more than threefold increase in seven months. In Texas, a law enacted last year mandates that large power users disclose whether they have submitted similar applications for the same project in other regions. Applicants must also prove site usage rights, such as ownership or leases, and pay an initial review fee of at least $100,000 to deter multi-location spot securing. Ontario, Canada, employs a rigorous “grid impact review” that grades projects: fully confirmed projects count toward future demand at 100%, high-feasibility projects at 50%, and low-feasibility ones at 0%. Meanwhile, Australia saw connection applications spike to 44GW, prompting authorities to audit each project’s grid capacity and site permits, ultimately retaining only 7.9GW. For instance, when three developers each applied for 200MW for the same site, regulators calculated only the feasible portion rather than counting the full 600MW.

South Korea has also initiated efforts to filter out speculative data center demand. The government largely incorporated semiconductor investment plans, deeming them to carry high “policy intent,” but split data center projections into phases due to greater investment uncertainty. Professor Choi Yong-ok of Chung-Ang University, a member of the 12th Plan’s demand forecast committee, explained, “Projects with confirmed locations were reflected in phase one, while those without location info were deferred to phase two.” In July, the government announced plans for 7.4GW of data center capacity across several regions: Chungcheong (2GW), Southwest (1GW), Gangwon (2.4GW), and Yeongnam (2GW).

However, designating regions does not guarantee immediate construction, making it difficult to treat these allocations as firm power demand. Professor Jo Sang-min of Korea University of Technology and Education emphasized, “Even phase one needs verification if it will lead to actual investment. Like overseas, there should be a system to reassess feasibility based on project progress, such as site acquisition and grid connection procedures.”