Samsung, SK Hynix reject KEPCO’s $17B prepayment proposal for power grids

1 hour ago 28

South Korea’s state-run power monopoly asked its two biggest corporate customers to essentially pre-fund their own electricity supply. Samsung Electronics and SK hynix said no.

Korea Electric Power Corp. (KEPCO) proposed that the two semiconductor giants prepay a combined 25 trillion won, roughly $17-18.4 billion, in future electricity bills. The money would have financed construction of power grid infrastructure needed to supply massive new chip fabrication clusters. Samsung’s share was 20 trillion won, with SK hynix on the hook for 5 trillion won, figures based on their projected 2025 electricity consumption.

Why KEPCO needs the money

KEPCO isn’t exactly flush with cash. The utility had accumulated debt of 210.7 trillion won as of June 2026. To put that in perspective, the company is burning through approximately 11.5 billion won per day just in interest payments.

That debt load makes traditional borrowing an increasingly painful option. The prepayment scheme was designed as a creative workaround: get the money without adding to KEPCO’s already staggering debt pile while giving chipmakers a clearer timeline on when their factories would actually have power.

The infrastructure in question isn’t optional. New semiconductor clusters in Yongin and the Honam region are projected to require multiple gigawatts of power capacity. Without the infrastructure, the fabs can’t operate. Without funding, the infrastructure doesn’t get built.

From KEPCO’s perspective, the math was straightforward. The interest chipmakers would forgo on their prepaid balances would sit somewhere between 3.4%, the two-year Korean Treasury bond yield, and 3.7%, the KEPCO bond yield. That’s the implicit cost the companies would bear for essentially lending money to their power provider.

Why the chipmakers walked away

Both Samsung and SK hynix cited the same core concern: locking up that much capital when the AI-driven semiconductor boom might not sustain its current trajectory feels like a risky bet.

That reasoning is telling. These are companies whose entire expansion strategies are built around AI demand. SK hynix recently announced plans to invest 54.3 trillion won in two new memory fabrication facilities specifically to capture the AI-driven surge in high-bandwidth memory and other advanced chips.

Yet both are drawing a line between investing in their own production capacity and fronting cash for someone else’s infrastructure. The distinction matters. Building a fab is an investment in your own competitive position. Prepaying electricity bills is, at best, an interest-free loan to a heavily indebted utility. At worst, it’s capital you can’t get back if demand softens.

The infrastructure dilemma remains

Saying no to the prepayment plan doesn’t eliminate the underlying problem. Samsung and SK hynix still need those power connections. The fabs they’re planning to build are useless without reliable, high-capacity electricity supply.

KEPCO, meanwhile, still needs to find the money somewhere. The utility’s debt-to-equity ratio has been a persistent concern for South Korean policymakers, and adding more borrowing to fund semiconductor infrastructure would only deepen that hole.

The situation highlights a tension playing out globally as AI-driven demand reshapes energy infrastructure requirements. In South Korea, the dynamic is different because KEPCO operates as a monopoly, meaning there’s no alternative provider the chipmakers can turn to. They need KEPCO, and KEPCO needs their money.

For investors watching the semiconductor space, the rejection sends a nuanced signal. On one hand, it suggests Samsung and SK hynix are being financially disciplined, protecting their balance sheets against a potential demand downturn. On the other, it introduces uncertainty about the timeline for bringing critical new fabrication capacity online. Samsung and SK hynix together dominate the global memory chip market, and their expansion plans are central to meeting worldwide demand for AI training and inference hardware.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.

Read Entire Article