South Korea is again standing at a height it can jump down from

In the winter of 1998, a certain kind of death in South Korea belonged only to middle-class fathers: jumping off the rooftop of an apartment building they had paid for over ten years. It wasn’t out of despair—it was because they had done the math. Once a person dies, life insurance payouts can cover the remaining mortgage, and the wife and children can keep that home. That year, the suicide rate among middle-aged men in South Korea jumped by more than 40%.

Twenty-eight years later, in 2026—after an entire generation—Seoul’s home prices have climbed for 74 straight weeks. The KOSPI’s intrayear peak reached 9,385 points. A Samsung Electronics engineer can borrow 500 million Korean won from the company at a 1.5% interest rate to buy a home in the Dongtan area, with no meaningful constraints from government loan limits.

Meanwhile, South Korea’s foreign exchange reserves stand at $427.3 billion, and the current account surplus is at a record level. It looks like 1998 will never come again.

But if you lay this country’s balance sheet bare—household debt approaching 2000 trillion Korean won, and equity-market financing balances surging to the highest level since records began in 1999—you realize something uncomfortable: South Korean households are once again standing at the top of a cliff. Only this time, the ground beneath their feet isn’t external debt—it’s AI chips.

The outer wall

South Korea won’t replay the kind of crisis it saw in 1997–98.

Back then, the Thai baht collapsed first. Panic swept across Southeast Asia, and foreign banks refused to roll over short-term dollar loans to Korean companies. South Korea’s foreign exchange reserves were on paper about $20 billion, but the amount truly available at one point was only about $6 billion.

The won crashed from 800 to nearly 2,000. For companies that had borrowed in dollars, their debt obligations doubled overnight. Banks followed the collapse, and eventually the entire country accepted a $58.4 billion IMF bailout along with an entire set of humiliating conditions.

Every step along that route today is blocked.

Those $6 billion of reserves have grown to $427.3 billion—enough to cover short-term external debt at 2.6 times. A current account deficit flipped into a surplus. In 2025 alone, it recorded a $123 billion surplus, setting a record in South Korea’s history. With the exchange rate freely floating, nobody can target the won again.

Breaking in from the outside is no longer possible. But that’s exactly the problem: everyone is staring at the wall outside—rebuilt for 28 years—while very few people look down at what’s under their own feet.

The floor inside

South Korea’s central bank released the latest data on July 7. In the first quarter, the ratio of household debt to GDP fell to 85.3%, dropping by nearly 3 percentage points in a single quarter. It looks like deleveraging is happening—and not slowly.

But three weeks ago, at a press conference for the Financial Stability Report, Vice Governor Jang Jeong-soo of the Bank of Korea said one line that punctured that comfort: nominal GDP growth is highly concentrated in semiconductors. “Those households that are borrowing aren’t seeing income improvements across the board.”

Household credit outstanding already reached 1,993.1 trillion won in the first quarter, rising month after month at a pace of tens of trillions of won. The ratio is falling thanks to the denominator. Nominal GDP rose 17.1% year over year—the fastest in three decades—but nearly all of it came from chip sales by Samsung and SK hynix.

The Bank of Korea’s own breakdown shows that in the first quarter, export components in the GDP deflator rose 23.5%, while domestic consumption rose only 2.1%. Corporate operating surplus rose 17%, but labor compensation rose just 4%.

The money went into semiconductor companies’ profit and loss statements—it didn’t land in household bank accounts. Minister of National Income Kim Do-eun acknowledged at the press conference that the widening of the nominal GDP growth rate could be expected to reduce household debt and government debt as a share of GDP.

He was telling the truth—the ratio really is falling. But it’s not falling because households are repaying debts; it’s falling because chip sales are going well. Debt is hard; GDP is soft. The attractive-looking 85.3% number is standing on a single chip.

So where did those tens of trillions of won added to household debt every month actually go?

First, into the stock market.

In the second quarter, the KOSPI surged 59% over three months. In June it even touched 9,385 points intraday. Retail investors didn’t just use their own money to pile in—they borrowed too. In late June, the financing balance hit a historical high of 386 trillion won.

In the Financial Stability Report, the Bank of Korea calculated an even larger figure: as of end of May, total financing used to buy stocks directly plus credit-backed funding amounted to 394 trillion won, increasing by 112 trillion won within the year. The growth rate was the largest since records began in 1999.

And this leverage was concentrated heavily in just two stocks. By mid-June, the combined financing balance for Samsung Electronics and SK hynix was about 91 trillion won—nearly four times the start of the year—and accounted for more than one-third of all financing on the KOSPI.

On June 23, the KOSPI plunged 9.99% in a single day, falling by 910 points and setting the largest single-day decline in history, triggering the fourth circuit breaker of the year. That day, foreign investors outside the KOSPI sold 41 trillion won, and institutions sold 45 trillion won. Retail investors, however, turned around and bought 85 trillion won—also a historical record for a single day.

By the end of June, the KOSPI had pulled back about 10% from its high, but the financing balance only dropped from the peak of 386 trillion won to 373 trillion won—almost unchanged. They weren’t liquidating positions. They were using leverage to bottom-fish.

A repeat of a selloff at the same level would trigger reverse liquidation: if the stock price breaks through the margin line, brokerage systems automatically sell. The sold shares then push prices even lower, triggering the margin lines of the next batch of leveraged investors.

Then, into the real estate market.

After the June 23 crash, money in the stock market didn’t return to banks. The monthly growth in other loans slowed by about 1.6 trillion won compared with May—less borrowing to trade stocks—but the growth in housing mortgage loans expanded instead, from 40 trillion won to 45 trillion won. The total cooled, but the leverage was moved from one address to another.

And where did it move to? Not Gangnam, but the heartland of the chip fabs.

Dongtan in Hwaseong rose 1.29% over one week, while Yeongtong-dong in Suwon rose 1.19% over one week. The media calls this “semiconductor money.” On June 30, the government designated Dongtan, Uicheong, and Guri as speculation zones. The price gains didn’t slow—they accelerated. Hot money flowed into nearby places like Suwon, Namyangju, and Seongnam’s Bundang district.

Samsung Electronics has finalized, via labor-management agreements, an internal home-buying loan system for employees without homes: an annual interest rate of 1.5% with a maximum loan amount of 500 million won. It is not constrained by DSR or LTV rules and is preparing for implementation. SK hynix is seeking the same terms. The Financial Services Commission said the truth on July 9: legally, it can’t be done. For ordinary people, mortgage rates are 6%. For Samsung engineers, 1.5%. Regulators can’t rein it in.

At this point, you can see one thing clearly: whether it’s financing positions in the stock market or mortgage loans in real estate, every direction the money flows points to the same thing—semiconductors.

The KOSPI rises because of Samsung and SK hynix; the KOSDAQ in the same period falls 18%. The real estate boom is concentrated in the fab belt. The reason nominal GDP grows enough to pull down the debt ratio is also because of chips. Four seemingly independent things—stocks, real estate, GDP, leverage—are sitting on the same floor.

If chip sales slow down

The previous semiconductor downturn period was already not far off, from 2022 to 2023. Back then, chip exports had negative year-over-year growth for more than a dozen straight months. But at that time, household leverage wasn’t as high, financing balances weren’t as extreme, and home prices weren’t tied as tightly to fabs.

This time, if AI capital expenditure slows—without needing a crash, just a slowdown—the process will play out step by step.

Nvidia’s orders first decline. Then Samsung and SK hynix’s earnings are revised downward. The KOSPI, weighed by the 386 trillion won financing pile, drops. After that, apartments in Gyeonggi Province—lifted by “semiconductor money”—start to loosen. With the engine of nominal GDP gone, the household debt-to-GDP ratio bounces back from 85.3%. Then fragile borrowers will default en masse...

But that doesn’t necessarily mean it will collapse.

South Korea’s household loan delinquency rate is currently 1.00%, below the long-term average. Banks have sufficient capital. The IMF judges that South Korea’s “financial stability risk is manageable.” Mortgages are mostly with recourse—there is no subprime-style securitization contagion like in the United States.

The $123 billion current account surplus in 2025 and South Korea’s net creditor country status are hard-moat defenses. In 2022, the Bank of Korea raised rates from 1.00% to 3.25%. Home prices fell by about 8% over two years, without evolving into a systemic crisis. South Korea has the capacity to absorb one round of housing price adjustments.

If the AI infrastructure construction cycle lasts five to ten years, the sustained expansion of nominal GDP means real deleveraging.

What if it goes wrong?

Now the most important indicator is the year-over-year growth rate of South Korea’s semiconductor exports. If it turns negative for two or three consecutive months, the situation will start to enter a negative feedback loop.

In 1998, the last thing those fathers did before jumping was to do the math. They weren’t beaten by emotion—they were beaten by an arithmetic problem: mortgage balances, life insurance payouts, and the cost of living for their wife and children. After solving it, the answer pointed in one direction—and they went.

In 2026, South Korea is doing its own math. Only the variables have changed: not external debt and exchange rates, but AI capital expenditure and nominal GDP.

The entire country’s deleveraging, stock valuations, housing prices in Gyeonggi Province, and the sustainability of household debt of 2000 trillion won are all staked on one assumption: AI chips continue to sell at full speed.

If they get it right, everyone stays safe.

If they get it wrong—those fathers in 1998 knew what it feels like to be wrong.