[Edaily Marketin JI YEONG-EUI Reporter] "It is essential to design and build power infrastructure and artificial intelligence (AI) infrastructure together, rather than as completely independent systems."

Amid growing strain on the power grid due to the rapid expansion of AI data centers, experts have suggested that the response should not only involve building new power plants and transmission lines but also enable data centers to flexibly adjust their power consumption. It is widely acknowledged that South Korea, where demand for AI is surging, must simultaneously expand investment in the power grid and improve the operational efficiency of existing infrastructure.

Mansi Shah, Head of Product at Emerald AI, emphasized this point during a presentation at the “Bitgaram International Power Technology Expo (BIXPO 2026)” held on the 5th at the Kim Dae-jung Convention Center in Gwangju, stating, “Given the scale and speed at which AI is growing, we cannot afford to wait five, seven, or eight years to build the necessary power infrastructure.”

Sha explained that not all tasks in an AI data center require the same level of power supply. While some tasks, such as fraud detection, must be processed immediately, others can be delayed slightly without significantly disrupting service.

“AI workloads inherently possess flexibility,” he said. “Even if power is restricted, the tasks will continue; they will simply be processed somewhat more slowly.”

As methods for regulating power consumption, he suggested limiting the power usage of graphics processing units (GPUs), shifting tasks to times when the power grid is less congested, and distributing tasks across data centers in different regions.

Sha emphasized, “By utilizing data centers flexibly, we can bring them online faster using the existing power grid, even before new transmission lines, substations, and power generation facilities are completed.”

He also presented real-world examples. Emerald AI conducted initial pilot projects in the United States, the United Kingdom, and the European Union (EU). He explained that at a data center in California, the company used software to adjust approximately 35,000 AI workloads, thereby reducing power consumption.

“We maintained a reduction in power consumption of about 15% and gradually increased usage after the power grid returned to normal,” he said, adding, “This enabled the data center to immediately secure about 25% more power.”

He emphasized the need to apply this approach to data center development in South Korea as well. Sha noted, “Most data centers in South Korea are located in the Seoul metropolitan area,” and diagnosed that “where to source power is the key constraint.”

He continued, “Power flexibility actually works and can be a powerful tool,” and expressed his intention to pursue demonstration projects with domestic power companies and other entities.

However, some pointed out that demand response alone cannot replace investment in new infrastructure.

Mark McGrath, a Fellow at the Electric Power Research Institute (EPRI), emphasized, “Flexibility alone cannot solve everything. We need to build transmission and distribution lines,” stressing the need to pursue expanded grid investment alongside digital transformation.

Citing an analysis by the European electricity industry association EurElectric, Researcher McGrahan explained, “Europe’s distribution grid requires approximately 67 billion euros annually,” adding that this is “at least double the amount invested each year over the past five years.” The expansion of data centers, the proliferation of electric vehicles, the growth of renewable energy, and efforts to ensure grid resilience in the face of climate change were cited as key drivers for increased investment.

He stressed the need to increase investment while simultaneously improving the utilization of existing infrastructure. To achieve this, he argued that a platform is needed to predict and coordinate electricity usage across various resources, including data centers, electric vehicles, and residential energy storage systems.

Researcher McGranahan stated, “Digitalization is the key to making all of this work,” adding, “Sensors and data installed in the power grid are essential for implementing AI.”

He also mentioned Korea Electric Power Corporation’s (KEPCO) Advanced Distribution Management System (ADMS). Noting that he had seen a demonstration at the KEPCO Human Resources Development Institute, he assessed it as “an excellent starting point for the technology needed in future control centers.” He continued, “We must optimize power grid operations by integrating management, forecasting, and coordination,” and urged that the benefits of new operational technologies be factored in from the investment planning stage.

During the panel discussion following the presentation, moderated by Professor Lee Jong-soo of Seoul National University, “trust” was emphasized as a prerequisite for collaboration between AI and the power industry. It was also suggested that safety and performance should be verified through virtual simulations before introducing AI into actual industrial settings, and that acceptance among on-site personnel should be increased.

Researcher McGranahan stated, “Making the work of power grid operations personnel easier and more accurate is the first step toward building trust before introducing AI into core infrastructure.”

Sha, the director, also emphasized, “Power companies and data centers must put safety measures in place and share sensitive data in order to find trustworthy solutions.” He added, “If data centers demonstrate flexibility and shoulder a greater share of the costs, local communities will be able to secure a more stable power grid at the same or lower cost.”