AI Power Grid Costs Should Not Be Passed to the Public [Why It Matters]

South Korea's push for artificial intelligence as a future growth engine hinges on a critical question: who will pay for the vast electricity infrastructure required to power it? The government and corporations champion AI development, planning to expand data centers and secure more advanced hardware. However, the substantial energy demands of AI data centers, which operate 24/7 and require extensive cooling, pose a significant challenge.

Beyond sheer consumption, these large facilities necessitate upgrades to power generation, transmission lines, and substations. The costs associated with these reinforcements, along with the need for water, land, and backup power, could disproportionately burden local communities and general electricity consumers, while corporations reap the profits. The article argues that if AI and semiconductor clusters are national strategic projects, the supporting power infrastructure must be part of a national plan, not a burden shifted to the public and local areas.

To foster public acceptance, the article suggests that corporations must bear a reasonable share of the costs for power generation and grid reinforcement. Conversely, if data centers provide tangible benefits like tax revenue, jobs, and improved local infrastructure, public acceptance could increase. This requires a clear distinction between costs borne by businesses and those considered public infrastructure, alongside a reform of electricity pricing structures.

The piece advocates for a robust 'power system impact assessment' to transparently evaluate the appropriateness of requested power, the grid's capacity, the location of new transmission lines, and the cost allocation before approving large data centers. It also calls for preventing businesses from hoarding grid capacity or securing usage rights for projects with low feasibility. While not advocating for outright suppression of corporate power demand, the article stresses that businesses generating costs should be responsible for them, with swift power connections for feasible projects.

Furthermore, it suggests that AI facilities could manage their power consumption by shifting non-urgent tasks to off-peak hours and reducing usage during grid instability. While self-generation, energy storage, and renewable energy purchases can mitigate some impact, they cannot fully meet the 24/7 demands of AI operations. A balanced approach combining various power sources, storage, and distributed generation, based on clear criteria for cost, risk, and stability, is essential. The article concludes by emphasizing that true AI competitiveness lies not just in rapid construction but in fair cost-sharing policies that ensure stable power, preventing corporations from offloading electricity expenses onto the public.