The South Korean government is weighing a plan to cut electricity bills for artificial intelligence data centers that keep their power demand stable, according to sources. Domestic data centers' power demand is projected to expand to 20 gigawatts by 2040, equivalent to the output of 13 nuclear reactors.
A senior government official said the government plans to overhaul the system by offering discounts to AI data centers that maintain steady load patterns, securing both industrial competitiveness and grid stability at the same time.
Kim Sung-hwan, Minister of Climate, Energy and Environment, had earlier said the government would introduce a dedicated tariff within this year to strengthen competitiveness in attracting AI data centers, speaking at a public briefing on three mega projects in late June and at a work report early this month. The plan now is to also factor in how much AI data centers contribute to grid stability.
The government is considering conditional discounts for AI data centers because the burden they place on the national power grid is substantial. Semiconductor fabs and AI data centers both require large amounts of stable power around the clock, but AI data centers have the problem of erratic power demand. During "checkpoints," when AI saves intermediate results of its training, the load can drop to nearly zero for milliseconds, and then power demand surges when operations resume immediately afterward.
Park Chan-guk, a professor in the Division of Climate Change Convergence at Hankuk University of Foreign Studies, said AI data centers show extreme power fluctuations ranging from milliseconds to minutes during checkpoints, synchronization delays and the end of training runs, owing to the synchronized nature of hundreds of thousands of graphics processing units (GPUs). He noted that this not only poses a direct threat to power system stability but, if the sharp load swings combine with the problem of declining system inertia caused by wider adoption of renewable energy, could also lead to the risk of cascading blackouts. The government's goal of building GW-scale large AI data centers in each region is, from the standpoint of the domestic grid, a major challenge in itself.
In response, the government also plans to create a grid code — technical standards — for data centers, defining the minimum equipment needed to safeguard grid stability when a data center is built. An official at the Ministry of Climate, Energy and Environment explained that because AI data centers place a substantial burden on the grid, setting equipment conditions is unavoidable.
Meanwhile, major countries abroad are also moving to secure grid stability through data center grid codes and dedicated tariffs. Germany is pursuing a system that varies grid usage fees according to how much large loads contribute to system flexibility, while U.S. states such as Ohio require new data centers of 25 megawatts or more to pay for at least 85% of their contracted capacity regardless of actual power use.
