Won Past 1,500, How Dollar-Priced AI Infrastructure Breeds Hidden Currency Risk

The won's quarterly average has slipped past 1,500 to the dollar, a level last seen during the late-1990s financial crisis. That distinction matters. The spikes of 1997 and 2008 were convulsions of a breaking system, and when the panic passed the currency drifted back toward its old range. This weakness is different. It does not reduce to a single shock. It reflects a rate gap with the United States, a shifting trade structure, and above all the very investment program Korea has staked its future on. The AI buildout is itself a structural source of dollar demand, which means the exchange rate has stopped being a line item and become a precondition of industrial strategy.

The dollar invoice behind AI ambition

When a country declares that it intends to lead the AI era, the physical form of that ambition is data centers and semiconductor fabs. The trouble is that the components and equipment filling those facilities are billed almost entirely in dollars. Nvidia's accelerators settle in dollars. ASML's EUV scanners, the beating heart of any leading-edge fab, are priced in euros and dollars. Much of the specialized material and tooling that goes into high-bandwidth memory and advanced packaging is sourced abroad and paid for in dollars as well. Korea's world-class position in memory does not soften this exposure; if anything it exposes a timing mismatch. The capital to buy the equipment must go out now, in dollars, while the export revenue that justifies it arrives years later. The country earns dollars selling finished chips and spends dollars building the machines to make them, and the two flows do not line up.

What makes this acute is that the dollar-denominated cost was already swelling on its own, independent of the exchange rate. Wafer prices at the leading nodes climb steeply with each generation, and a single next-generation High-NA lithography tool now carries a price tag in the hundreds of millions. A reignited memory supercycle is pushing HBM and DRAM prices upward again. This is chipflation: the assets themselves are growing more expensive in dollar terms. Now Korea must buy those ever-pricier assets with a weakened currency.

A cost that compounds twice

Here the currency risk strikes back through simple multiplication. The real investment cost, expressed in won, is inflated by two factors stacked on top of each other: the rise in the dollar sticker price, and the rise in the exchange rate that converts it. If equipment costs twenty percent more in dollars and the won weakens another fifteen percent, the burden felt in domestic currency is not the thirty-five percent of naive addition but closer to thirty-eight percent, because the two effects multiply. When a nation is expanding not one machine but tens of trillions of won in fabs and data centers simultaneously, that multiplication rewrites the break-even point of every business plan.

The cruel timing is that this cost inflation lands exactly when investment must be most aggressive. AI infrastructure is a race in which whoever secures capacity first captures the market. Delay because the exchange rate stings, and you fall behind; push ahead, and the financial strain accumulates. Firms can hedge part of the exposure, but covering a multiyear, large-scale capital program entirely through hedges is impractical given cost and maturity structure. A substantial share has to be carried openly.

When macro risk and household leverage align

The most unsettling feature of this picture is concentration. One of the forces dragging the won lower is the dollar outflow tied to AI investment, and one of the stories propping up Korean equities and household expectations is the very same AI rally. If leveraged retail investors, those who borrowed to buy and scraped together every available won, are crowded into chip and AI names, then macro currency risk and micro household leverage are aligning toward the same underlying asset.

Conventional wisdom holds that a weak currency flatters exporters and supports their share prices. But when the weakness springs from capital flight and ballooning capital expenditure, the logic inverts. Should the AI cycle wobble, three pressures could converge in one direction at once: earnings squeezed by investment costs, foreign capital fleeing on currency anxiety, and leveraged households forced into simultaneous liquidation. Risk that ought to be diversified instead converges toward a correlation of one beneath a single narrative. That is why the figure 1,500 should be read not merely as a question of import prices but as a structural signal that a nation's growth strategy and its households' savings are now wagered on the same throw.