Samsung, SK Hynix Now Financing Korea's Power Grid — And That Should Worry Everyone

The Deal That Sounds Fair but Isn’t

Korea Electric Power Corporation has asked Samsung Electronics and SK Hynix to prepay five years of electricity bills—20 trillion won and 5 trillion won respectively, 25 trillion in total. On the surface this is a modest financial arrangement. The interest rate offered is the government bond yield plus 0.15 percentage points, roughly mid-3 percent. Samsung and SK Hynix would not hand over cash outright; instead they would pay gradually over a year and then deduct each month’s electricity bill from the balance. KEPCO saves on borrowing costs. The two chipmakers earn a slightly better return than they would on government bonds. Everyone wins.

Except this is not really a commercial deal. It is a structural transfer of infrastructure risk from a crippled state utility onto the two companies that carry Korea’s semiconductor competitiveness on their backs.

210 Trillion Won and Growing

The numbers behind KEPCO tell a different story. As of June 2026, consolidated debt including affiliated generators stood at 210.7 trillion won. Borrowings alone reached 133.3 trillion won. Interest costs in the first half of the year totaled 2.1 trillion won—approximately 11.5 billion won per day. The special exemption that allowed KEPCO to issue bonds up to five times its capital and reserves expires at the end of next year. The window is closing.

Meanwhile, demand is exploding. The Yongin semiconductor cluster will require up to 15 gigawatts of new power. Four fabs in the Honam cluster need another 6.3 gigawatts. AI data centers add further pressure. KEPCO’s eleventh long-term transmission and distribution plan already raises investment through 2038 to 72.8 trillion won, up 16.3 trillion from two years ago. The real number will be higher once mega-cluster and data-center demand is fully reflected.

There is simply no money to build the grid. So KEPCO is asking the customers who will use it to finance the build-out first.

Reverse Circular Financing

Analysts have compared the structure to NVIDIA’s circular financing model, where the chipmaker lends to its own customers so they can build data centers that then buy more GPUs. The direction is flipped here. Samsung and SK Hynix lend to KEPCO so KEPCO can build the grid that lets them keep producing chips. Money flows from customer to supplier rather than supplier to customer.

The analogy is useful because it reveals what is actually happening: the boundary between infrastructure provider and industrial customer is dissolving. In the AI era, whoever controls power grid access controls semiconductor output. KEPCO cannot afford to control it. Samsung and SK Hynix are being asked to fill the gap.

This is not unprecedented in global history. Industrial firms have long subsidized local infrastructure where state capacity was insufficient. But what makes this moment distinct is the scale and the specificity. Korea has bet its economic future on semiconductors and AI. The government has declared an “AI powerhouse” strategy while its primary power utility carries 210 trillion won in debt and cannot issue bonds freely. The contradiction is not accidental—it is the direct result of prioritizing competitive industry over the infrastructure that underwrites it.

The Government Was in the Room

The proposal was discussed in July between officials from the Ministry of Economy and Finance, KEPCO, Samsung Electronics, and SK Hynix, and reportedly reported to the presidential office. That level of government involvement changes the character of the transaction. If Samsung and SK Hynix participate purely on commercial grounds, the deal is defensible. If refusal carries any implicit political cost, the voluntariness disappears.

There is also a longer-term danger. Once this model is accepted, it becomes a precedent. Other utilities in other sectors could replicate it. State-owned infrastructure providers facing fiscal constraints could routinely shift capital expenditure onto private corporate balance sheets. The boundary between public responsibility and private obligation erodes incrementally.

Who Wins, Who Loses

In the short term Samsung and SK Hynix gain something tangible: certainty that power grid construction will not delay their fab projects. A factory that cannot connect to the grid is an abandoned asset. The prepayment effectively buys insurance against that risk. KEPCO gains cheaper funding and reduced bond-market exposure. The government gains a narrative of public-private cooperation without spending fiscal resources.

The loser is the structural balance of risk. Samsung and SK Hynix are now financing not just their own operations but the national power grid that supports them. Their capital allocation decisions are no longer purely commercial. They are quietly subsidizing infrastructure that should be a public good. If the prepayment model expands—if KEPCO asks other heavy industrial consumers to participate—the effect compounds.

The other silent loser is policy credibility. Korea’s semiconductor strategy depends on keeping the world’s most advanced fabs in the country. But those fabs require enormous, stable, affordable electricity. A utility drowning in debt cannot deliver either stability or affordability indefinitely. The 25 trillion won prepayment is a bandage on a structural wound.

What Comes Next

The immediate question is whether Samsung and SK Hynix will accept the terms and how quickly. The broader question is whether this becomes the standard model for financing Korea’s energy transition alongside its industrial policy. If it does, the 25 trillion won figure matters less than the 210 trillion won already on KEPCO’s books. That is the number that determines whether the grid gets built at all—or whether the semiconductor industry eventually runs out of power it cannot afford to pay for.

Washington and Brussels watch Korea’s semiconductor push with interest. Few in those capitals will notice the power grid financing layer underneath it. That is precisely why they should. The bottleneck in AI and chip competition is no longer just manufacturing capability. It is who pays for the electricity that makes manufacturing possible.