South Korea’s AI Chip Boom Hits a Power Financing Wall

South Korea’s artificial-intelligence chip expansion has encountered an obstacle that has little to do with transistor design: who pays for the electricity infrastructure needed to run the factories.

Samsung Electronics and SK Hynix have rejected a proposal from state-run Korea Electric Power Corp., or KEPCO, to prepay roughly 25 trillion won ($18.7 billion) for electricity that would support planned semiconductor mega-clusters. The proposal would have required Samsung to provide about 20 trillion won and SK Hynix about 5 trillion won—roughly equivalent to five years of each company’s recent electricity bills. (investing.com)

The decision is consequential because it puts a financial boundary around South Korea’s ambition to build some of the world’s largest chip-production centers. It also offers an early warning that the AI infrastructure boom is not simply a story of unlimited demand meeting constrained supply. Even the companies best positioned to benefit from AI may hesitate when expansion requires them to underwrite the power grid years before the factories reach full production.

What changed

KEPCO’s proposal was designed to solve a practical problem. New semiconductor facilities require enormous and highly reliable electricity supplies, while the utility is carrying heavy debt and needs capital to expand transmission and generation infrastructure. By asking the chipmakers to pay in advance, KEPCO could obtain funding for the networks serving clusters under construction in Yongin, south of Seoul, and in South Korea’s southwestern Honam region. (en.yna.co.kr)

Samsung and SK Hynix reviewed the proposal internally and concluded that accepting it would be difficult. Reporting by Reuters, based on a document submitted to a lawmaker, said the companies questioned whether such large upfront payments were necessary and cited uncertainty over the long-term durability of semiconductor demand. Yonhap separately reported that the companies wanted to limit management uncertainty while the AI-driven chip cycle remains difficult to forecast. (investing.com)

Neither company publicly endorsed the arrangement, and neither offered a detailed explanation. That leaves open whether the rejection is a broad signal of weaker confidence in AI demand or a narrower objection to the structure of the financing. The distinction matters: a company can remain committed to building factories while refusing to finance a utility’s balance sheet through a five-year prepayment.

Why it matters

The dispute reveals that electricity has become a strategic input to semiconductors on the same level as advanced lithography, high-bandwidth memory and clean-room capacity. Governments can announce chip clusters, provide tax incentives and attract anchor manufacturers, but factories cannot operate without transmission lines, substations and dependable generation.

South Korea’s challenge is particularly sharp because its industrial policy is trying to concentrate production geographically. Clusters can improve logistics, supplier access and workforce efficiency, but they also create large, synchronized demands on local infrastructure. If public utilities lack the capital to expand quickly, policymakers must decide whether the state, the manufacturers, cloud companies or consumers should bear the cost.

The rejection also complicates a popular assumption about AI demand. The leading memory suppliers have benefited enormously from demand for high-bandwidth memory used in AI accelerators. That has encouraged governments and investors to treat the current cycle as a durable structural shift rather than a conventional semiconductor boom. But Samsung and SK Hynix’s refusal to commit $18.7 billion upfront suggests that even major beneficiaries are unwilling to treat future demand as risk-free.

This does not mean the companies expect AI investment to collapse. More likely, they are distinguishing between demand that is visible today and demand that is sufficiently certain to justify locking up capital for infrastructure serving factories that may take years to reach full utilization. That distinction could influence the pace and sequencing of new fabs across Asia, the United States and Europe.

A test for industrial policy

KEPCO’s proposal places South Korea’s industrial strategy in an uncomfortable position. The country wants to remain a central supplier of memory and advanced chips, but the financial burden of supporting that ambition is distributed across institutions with different incentives.

The utility needs long-term certainty to justify grid investment. Chipmakers need flexibility because technology cycles are volatile and product demand can shift rapidly. The government wants fast construction to preserve South Korea’s position against Taiwan, the United States and China. Local communities may face higher land, water and electricity costs even if the economic benefits arrive later.

Those incentives do not naturally align. A five-year prepayment would have transferred significant demand risk from KEPCO to Samsung and SK Hynix. The companies’ rejection therefore amounts to a negotiation over who should absorb uncertainty—not necessarily a rejection of the factories themselves.

The outcome may push Seoul toward a more conventional public-finance solution: government guarantees, regulated infrastructure bonds, targeted subsidies or cost-sharing agreements tied to actual electricity consumption. It could also encourage a slower, staged buildout in which transmission capacity is added alongside confirmed factory commitments rather than in anticipation of maximum expansion.

What remains uncertain

The immediate question is whether KEPCO can secure another funding mechanism without delaying the planned clusters. Public reporting does not establish that construction will stop, nor does it show that Samsung or SK Hynix have canceled major capacity plans. The companies’ refusal should not be treated as a forecast that AI memory demand is ending.

The larger uncertainty is how much of the semiconductor industry’s projected power demand will materialize on schedule. AI data centers, chip fabs and advanced packaging facilities are all competing for electricity, but their timelines and economics differ. If AI infrastructure spending remains strong, South Korea may eventually find that public investment in power capacity is indispensable. If demand normalizes, overbuilding the grid could leave KEPCO and taxpayers with stranded costs.

For investors, the episode is a reminder that the semiconductor cycle is becoming an infrastructure-finance cycle. The bottleneck is no longer only whether companies can secure equipment and engineering talent. It is whether utilities can fund the physical systems required to turn announced capacity into operating capacity—and whether manufacturers trust the demand outlook enough to pay for that system in advance.

South Korea’s chip strategy is still intact. But the KEPCO dispute shows that the next phase of the AI boom will be negotiated not just in boardrooms and laboratories, but across utility balance sheets, public budgets and the power grid.