U.S. Data Center Rules Won't Shake 'Samsung-SK hynix'? [Weekend Money]

Concerns over data center construction regulations after U.S. midterm elections
Big Tech responds with on-site power generation and flexible electricity use
Investment delays, not cancellations... Limited impact expected on HBM demand

Smart answers from chatbots are welcome, but that changes if the massive data centers behind them drive up my electricity bill. The opposition to data centers that has surfaced during the U.S. midterm elections is tied to the same concern: Should residents have to share the costs of bringing in power-hungry artificial intelligence (AI) data centers? The solutions proposed by big tech are surprisingly simple: generate the electricity they need themselves and use less when demand from others is high.

Mirae Asset Securities' Research Center forecast that regulations on big tech will increasingly require data center designs that place less strain on the power grid. Access to the grid was already a major obstacle to data center construction even before the elections. The analysis suggests that companies' preparations for on-site power generation and technologies to adjust electricity use could also help them respond to regulations.

The first approach is to build a dedicated power plant next to a data center. Instead of waiting for a turn to draw electricity from the local grid, the facility secures the power it needs separately. Microsoft signed a 20-year agreement with Chevron to receive power for a data center in Texas, pursuing a dedicated gas-fired power facility as part of the arrangement. The aim is to reduce the additional burden on the grid used by local residents.

The second approach is to change when electricity is used. Google has signed agreements with U.S. utilities to adjust demand by a total of 1 gigawatt (GW). When electricity is in short supply, it can scale back some AI training tasks or shift them to another time. Delaying non-urgent work eases pressure on the grid during its busiest periods. Google said this flexibility can help accelerate data centers' connection to the power grid.

Another option is to distribute facilities across different locations. Rather than concentrating GPUs in one place, the plan is to install them across multiple regions where power is available and connect them through high-speed networks. This would allow work to continue elsewhere even if a major construction project in one region is stalled. The report also suggested that investment in data centers outside the United States, including in South Korea and Japan, could accelerate as this trend takes hold.

The report estimated that up to 13 GW of capacity could be affected by regulations, or about 9% of the 150 GW in additional capacity expected by 2035. On this basis, it judged that AI investment is more likely to be delayed than abandoned altogether. Companies could recover free cash flow (FCF) and ease the burden of bond financing by postponing some capital expenditures. However, given cloud order backlogs and rapidly growing revenue, they cannot keep delaying capital investment indefinitely.

The report emphasized that the impact on South Korean semiconductor stocks should also be assessed by product category. "High-bandwidth memory (HBM), whose demand is secured through long-term agreements (LTAs), should remain relatively stable," said Kwon Youngbae and Kim Younggeon, researchers at Mirae Asset Securities. "General-purpose DRAM used alongside CPUs and SSDs may be more sensitive to construction delays. However, this is not a cause for concern, given that the shift to HBM is making the supply of general-purpose DRAM even tighter."

This content was produced with the assistance of AI translation services.