Samsung and SK Hynix Reject a Power Prepayment as Korea’s Grid Funding Gap Widens
Samsung and SK Hynix rejected a proposed 25 trillion won power prepayment, exposing a conflict between Korea’s chip ambitions and its grid finances.
The state-run Korea Electric Power Corporation, or KEPCO, wanted the manufacturers to advance roughly five years of electricity payments. The utility planned to use that cash to accelerate infrastructure for semiconductor clusters in Yongin and the Honam region.
The proposal would have turned two industrial customers into long-term financiers for the power network serving their future factories. Their refusal does not cancel those factories, but it leaves the underlying funding problem unresolved.
That distinction matters. Samsung and SK Hynix continue to announce large semiconductor investments as demand for artificial intelligence hardware drives memory sales. However, neither company accepted the idea that today’s favorable market conditions justify locking away five years of operating cash.
The episode is not a simple dispute over an electricity bill. It is a test of who should finance infrastructure when national industrial strategy moves faster than a heavily indebted utility can invest.
The Proposed SK Hynix Power Prepayment Was Really a Loan
KEPCO’s proposal asked Samsung and SK Hynix to finance grid construction before consuming the electricity tied to those payments.
According to a document provided to Reuters by lawmaker Lee Chul-gyu’s office, both companies conducted internal reviews and said the proposal was difficult to accept. Samsung and SK Hynix declined Reuters’ requests for comment.
The reported proposal covered a combined 25 trillion won, equivalent to about $18.7 billion at the reported exchange rate. Samsung’s requested contribution was approximately 20 trillion won. SK Hynix was asked for approximately 5 trillion won.
Those amounts reflected about five years of electricity consumption based on the companies’ previous annual payments. Samsung reportedly paid 4.1 trillion won during the reference year, while SK Hynix paid about 900 billion won.
KEPCO did not propose recognizing the entire payment as immediate electricity revenue. The utility would record revenue as it supplied power and deducted charges from the prepaid balances.
Economically, the arrangement looked more like financing than an ordinary customer payment. The chipmakers would provide cash before receiving electricity, while KEPCO would use that money for transmission lines and substations.
The utility reportedly considered paying interest above the yield on two-year Korean government bonds. Another possible structure involved reducing future electricity bills by the principal and accumulated interest.
KEPCO expected this approach to cost less than issuing more utility bonds. Samsung and SK Hynix would gain a return on their money and reduce the risk that grid construction delayed their factories.
Yet the scale made the plan unusual. Existing electricity prepayments totaled about 390 billion won during the previous year, according to the initial proposal details. State agencies made most of those payments.
The proposed 25 trillion won contribution was more than 60 times that existing annual prepayment total. It would have transformed a customer convenience into a major infrastructure funding instrument.
That transformation explains the resistance. The companies were not simply deciding whether to pay bills early. They were evaluating a concentrated credit exposure to their electricity supplier.
A company official told Reuters that uncertainty about the durability of semiconductor demand influenced the decision. The official spoke anonymously because of the issue’s sensitivity.
The answer from both manufacturers was ultimately the same. They support timely power infrastructure, but they would not accept KEPCO’s proposed financing mechanism.
Korea’s Chip Plans Are Moving Faster Than Its Power Network
A semiconductor fab cannot operate on investment announcements alone, because its electricity supply must arrive before production begins.
The power problem has become more urgent as South Korea expands several manufacturing regions at once. Yongin, south of Seoul, is the centerpiece, but planned development now extends into the country’s southwest.
The Yongin projects include separate manufacturing complexes associated with Samsung and SK Hynix. Together, their planned fabs, suppliers, utility facilities, and supporting operations create an industrial load measured in gigawatts.
Gigawatt-scale demand refers to continuous power capacity, not a factory’s annual electricity consumption. Supplying that load requires generation, high-voltage transmission, substations, and multiple layers of redundancy.
A delayed transmission line can therefore strand an otherwise completed factory. Cleanrooms, chipmaking equipment, and trained workers cannot compensate for an inadequate grid connection.
South Korea’s government recognized that dependency before the current financing dispute. In November 2024, Samsung, SK Hynix, KEPCO, K-water, and the Korea Land and Housing Corporation signed an infrastructure agreement covering power and water for Yongin.
The agreement reflected a shared understanding that private factory construction and public infrastructure must remain synchronized. Commercial operations at part of the cluster were scheduled to begin in 2027.
That deadline is now close enough to turn long-range planning into an immediate execution problem. Transmission projects require permits, land access, equipment, construction crews, and public acceptance.
SK Hynix has accelerated its side of the schedule. In February 2026, the company said it would move the first Yongin cleanroom’s planned opening from May 2027 to February 2027.
The company also approved 21.6 trillion won in additional facility investment through 2030. Combined with an earlier commitment, its stated investment in the first fab reached approximately 31 trillion won.
Its Yongin investment plan expanded the project to six cleanrooms across two building shells. Equipment installation costs were not included in those figures.
This creates a striking mismatch. SK Hynix is accelerating manufacturing capacity because it sees strong customer demand, yet it rejected a five-year electricity prepayment because long-term demand remains uncertain.
Those positions are not necessarily contradictory. A fab investment can be staged, modified, or aligned with customer commitments. A large cash transfer to KEPCO would immediately reduce financial flexibility.
The utility’s broader projections make that tension harder to postpone. Early completion of Yongin and development of the Honam cluster were estimated to add 20.6 gigawatts of electricity demand.
Regional AI data centers were expected to add another 7.9 gigawatts. These figures describe planned demand, so actual timing will depend on which projects reach operation.
Even so, the direction is clear. Korea is planning a much larger concentration of electricity-intensive computing and semiconductor production.
KEPCO’s grid must be ready before that demand materializes. Waiting until every factory’s commercial outlook becomes certain would leave too little time to build the required infrastructure.
KEPCO’s Debt Shifts the Pressure Back to the State
The rejection transfers the financing problem from two chipmakers back to KEPCO and South Korea’s industrial policy makers.
KEPCO originally planned to invest 72.8 trillion won in grid infrastructure through 2038. Expanded semiconductor and data-center projects increased the pressure on that program.
The utility has recently returned to operating profit, but its balance sheet remains heavily burdened. Its consolidated debt reportedly reached 210.7 trillion won at the end of June 2026.
That total had increased from 205.6 trillion won at the end of 2025. KEPCO also recorded 2.1 trillion won in interest expenses during the first half of 2026.
Another report placed the utility’s daily interest burden at approximately 11.5 billion won. Every additional bond therefore competes with grid construction for future cash.
KEPCO’s financing room also depends on a temporary legal measure. The government raised its bond issuance ceiling from twice to five times the combined value of capital and reserves.
That exception is scheduled to expire at the end of 2027. Unless lawmakers extend or replace it, the lower ceiling will return in 2028.
The power prepayment would have offered a bridge around this constraint. KEPCO could pull future customer payments forward without immediately relying on an equivalent amount of new debt.
The money would not solve KEPCO’s operating economics. It would improve the timing of its cash flows and give the utility a cheaper source of construction capital.
That distinction is central to the dispute. Samsung and SK Hynix were being asked to absorb part of KEPCO’s financing challenge, although the grid would remain a regulated national asset.
The companies would benefit directly from timely construction. However, transmission investments can also serve other customers and support regional development for decades.
This raises a public-policy question about allocating costs. Should major industrial users pay early because their projects create new demand, or should the state finance infrastructure promised under national policy?
South Korea previously moved toward greater public support. In April 2025, the government expanded its semiconductor assistance package to 33 trillion won amid trade and tariff uncertainty.
Officials said the program would include greater support for underground transmission infrastructure in Yongin and Pyeongtaek. The semiconductor package also included loans, subsidies, and research assistance.
KEPCO’s proposal moved in the opposite financial direction. Instead of drawing more state support, it asked the principal industrial beneficiaries to supply capital themselves.
The rejection leaves several alternatives. KEPCO can issue more debt, seek direct government funding, revise the prepayment terms, or share costs across a wider group.
Each option has consequences. More utility debt raises interest expenses, while taxpayer funding shifts the burden beyond electricity customers.
A smaller prepayment could become acceptable if it offered stronger security, a shorter duration, or clearer links to specific projects. However, neither company has publicly endorsed such a compromise.
The government can also revisit electricity tariffs for large industrial users. That choice would spread infrastructure costs over time, but it could weaken Korea’s manufacturing competitiveness.
No alternative removes the underlying expense. The grid must still be financed, permitted, built, and operated before the new factories reach full production.
Strong AI Demand Does Not Remove the Semiconductor Cycle
Samsung and SK Hynix are separating confidence in near-term AI demand from certainty about the next five years.
High-bandwidth memory, or HBM, combines multiple memory layers to deliver the data throughput needed by AI accelerators. Demand for these chips has supported earnings and manufacturing investment.
SK Hynix has benefited from a strong position in HBM. Samsung has also pushed to expand its presence as AI infrastructure spending reshapes the memory market.
Yet neither company treats current growth as a guarantee of permanently rising demand. Memory manufacturing has historically moved through periods of shortage, aggressive expansion, excess inventory, and price declines.
The SK Hynix power prepayment decision makes that caution visible. A five-year advance assumes that future electricity consumption and associated production remain sufficiently predictable.
Factories will still consume power during a market downturn. However, utilization rates, expansion schedules, and the timing of equipment installations can change quickly.
Chipmakers preserve cash because the sector demands continuing investment. They must fund new process technologies, fabrication equipment, packaging capacity, research, and overseas projects.
A 25 trillion won combined prepayment would compete with those commitments. Even an interest-bearing arrangement would restrict how easily the companies could redirect that capital.
The timing also complicates KEPCO’s pitch. SK Hynix is spending on Yongin while developing other Korean capacity and an overseas manufacturing footprint.
Samsung likewise faces capital decisions across memory, logic chips, foundry operations, displays, consumer devices, and global production sites. The two companies have different portfolios, but both value flexibility.
This is where the proposed power deal met its core limitation. KEPCO valued predictable infrastructure funding, while the manufacturers valued the option to adjust spending.
Both sides can support the same factories while disagreeing about who carries the demand risk. KEPCO needs investment years before electricity consumption becomes certain.
Samsung and SK Hynix want the grid available when their fabs need it. They do not want to fund the entire schedule based on forecasts they cannot control.
The anonymously reported explanation about long-term demand should also be read carefully. Neither company issued a public warning that AI memory demand was collapsing.
Samsung and SK Hynix declined to comment on Reuters’ report. Their refusal therefore reveals financial caution, not a verified change in production plans.
SK Hynix’s own statements still describe structural growth in demand from AI, data centers, and high-performance computing. The company also says future investments depend on customer forecasts and technological progress.
That qualification is important. Semiconductor companies routinely distinguish between a favorable market outlook and an unconditional commitment to every proposed expense.
KEPCO’s request demanded a particularly strong commitment. It asked the manufacturers to convert an uncertain stream of future operating payments into immediate financing.
Investors should not interpret the refusal as proof that the AI infrastructure cycle has peaked. The available evidence does not support that conclusion.
It is equally risky to dismiss the decision as routine negotiation. Two leading memory manufacturers independently rejected a mechanism designed around five years of expected consumption.
The most defensible interpretation lies between those extremes. AI demand remains strong enough to support new fabs, but not certain enough to eliminate disciplined capital allocation.
The Real Conflict Is National Ambition Versus Financial Flexibility
Korea wants infrastructure built for future chip demand, while the companies financing new fabs want room to respond if that demand changes.
This is the central tradeoff behind the story. Industrial policy rewards early construction because countries are competing to secure semiconductor capacity and supply chains.
Corporate planning rewards flexibility because a fab takes years to build and markets can change before it enters full production.
KEPCO’s proposed structure attempted to connect those competing time horizons. It would receive money early, while Samsung and SK Hynix recovered the balance through later electricity use.
The mechanism still concentrated risk on the chipmakers. They would provide funds before construction outcomes, future tariffs, power consumption, and semiconductor conditions were fully known.
KEPCO also has reasons to resist carrying all that uncertainty. It must construct shared networks before private demand becomes measurable through ordinary monthly bills.
If it waits for certainty, grid delays can undermine the government’s manufacturing strategy. If it builds aggressively with debt, its already large interest burden grows.
This tension appears across electricity-intensive industries. Data centers, battery factories, hydrogen projects, and chip plants often require infrastructure before their revenue outlook becomes certain.
The scale of semiconductor campuses makes the issue more acute. A single modern complex can require power comparable with the demand of a substantial urban area.
Manufacturing also needs reliability beyond ordinary commercial service. A short interruption can disrupt tightly controlled production and damage work already moving through a fab.
That need gives Samsung and SK Hynix a direct interest in KEPCO’s success. It does not automatically determine the appropriate financing structure.
A project-linked arrangement might distribute risk more precisely. Payments could follow completed substations, approved transmission milestones, or confirmed equipment installation schedules.
Government guarantees could also reduce counterparty concerns. Public funding could cover infrastructure with broad national benefits, while private users finance dedicated connections.
These possibilities remain hypothetical because the reported proposal did not produce a public counteroffer. KEPCO confirmed through Korean reporting that both companies communicated their refusal.
The rejection details suggest the utility must now identify other financing sources. They do not show that construction has stopped.
That uncertainty limits stronger conclusions. Reuters reviewed a document submitted to a lawmaker, but the full internal evaluations by Samsung and SK Hynix remain unavailable.
The public also lacks a detailed project schedule connecting each proposed payment with individual transmission assets. Without that information, outsiders cannot calculate the manufacturers’ potential return or exposure.
The political response remains another unknown. Lawmakers can change KEPCO’s borrowing limits, approve fiscal support, or demand that large users contribute more.
Local opposition and permitting can create delays even after financing is secured. More money cannot instantly resolve every route, land, or community dispute.
For technology customers, the issue sits several steps upstream. Developers and enterprise buyers will not receive an immediate change in chip availability because this proposal failed.
However, persistent grid delays would eventually constrain where Korea can add manufacturing capacity. That could affect the supply of memory used in AI servers, consumer electronics, and data centers.
The episode therefore offers a practical lesson about AI infrastructure. Chip availability depends on utilities, financing, construction, and public policy as much as semiconductor design.
Three Signals Will Show Whether the Grid Gap Is Closing
The next meaningful evidence will come from financing decisions, construction milestones, and changes to the manufacturers’ schedules.
The first signal is KEPCO’s replacement financing plan. The utility must explain whether it will rely on bonds, government support, revised prepayments, or another cost-sharing model.
A credible replacement announced soon would weaken concerns that the rejection threatens the semiconductor clusters. Continued uncertainty would strengthen the case that financing has become a schedule risk.
The details matter more than the headline amount. Investors should examine the cost of funding, repayment structure, participating institutions, and connection to specific grid assets.
The second signal is government action before KEPCO’s higher bond ceiling expires at the end of 2027. Lawmakers have time to extend the exception, create another financing vehicle, or allocate direct support.
Early legislation would give the utility more planning certainty. A prolonged debate would increase pressure as construction commitments approach their operational deadlines.
The government’s decision will also clarify its preferred division of responsibility. Public funding would treat the grid as national industrial infrastructure, while mandatory user contributions would shift more risk to manufacturers.
The third signal is any change to Samsung’s or SK Hynix’s factory schedule. SK Hynix currently expects its first Yongin cleanroom to open in February 2027.
Meeting that target with confirmed power access would show that the immediate phase remains manageable. Delayed equipment installation or revised production timing would indicate a more serious infrastructure constraint.
Readers should distinguish construction milestones from demand forecasts. A completed building does not prove full utilization, and strong chip orders do not guarantee that every planned fab opens on schedule.
Company financial reports can provide supporting evidence. Capital expenditure, free cash flow, HBM orders, and fab utilization will show whether corporate caution is increasing.
Those indicators should remain secondary to the three direct signals. The dispute centers on grid financing, legal borrowing capacity, and factory execution.
The SK Hynix power prepayment rejection is therefore best understood as an unresolved allocation problem. Korea has decided that it wants enormous semiconductor capacity, but it has not finalized who advances the money for the supporting grid.
For developers, enterprise technology buyers, and AI product teams, the immediate response is observation rather than alarm. Watch whether KEPCO names a funded alternative and whether Yongin’s first production milestones stay intact.
If both happen, the rejected proposal will look like a difficult financing negotiation. If either slips, the story becomes a warning that physical infrastructure is limiting Korea’s AI chip strategy.
The crucial question is no longer whether new fabs need electricity. It is whether government, utility, and industry can agree on financing before construction schedules outrun the grid.
