AI's Dollar Appetite and Korea's Stubborn Exchange Rate Floor

When the Strait of Hormuz reopened earlier this year and ceasefire signals emerged from the Middle East, currency traders positioned for a won recovery. The logic was straightforward: remove geopolitical risk, reduce safe-haven dollar demand, let the won breathe. Yet the dollar-won rate has remained stubbornly anchored above 1,500 — a level that once triggered emergency policy responses, now treated almost as a new normal. The standard model is not wrong. Something new has overwritten its assumptions.

That something is the AI supercycle, and it is reshaping Korea's balance of payments in ways that existing exchange rate frameworks were never designed to capture.

How AI Infrastructure Became a Structural Dollar Drain

Korea's current account has historically been a stabilizer for the won. Decades of trade surpluses — built on semiconductors, automobiles, and petrochemicals — generated steady dollar inflows that supported the currency even during turbulent periods. That structural support has not disappeared, but it is eroding at the margins in a way that matters enormously for the exchange rate floor.

The culprit is capital equipment. As global investment in AI data centers has accelerated — driven by hyperscalers in the United States, sovereign AI programs across Asia, and insatiable demand for high-bandwidth memory — Korean chipmakers like Samsung and SK Hynix have responded by aggressively expanding HBM production capacity. Expanding that capacity requires buying the most advanced semiconductor manufacturing equipment in the world: ASML's extreme ultraviolet lithography machines, precision etch systems, advanced deposition tools. Almost all of these are priced and settled in dollars.

A single EUV system costs upward of $380 million. When a Korean chipmaker places orders for multiple units as part of a multi-year capacity expansion, the dollar outflows are not a one-quarter anomaly but a sustained structural drain. For years, Korea's semiconductor export surpluses comfortably absorbed these equipment import costs. The AI supercycle has changed the arithmetic. The pace at which Korean chipmakers are ramping HBM output now demands dollar expenditures running faster than dollar revenues from chip sales can offset — at least in specific quarters. The current account surplus remains positive, but its structural ability to anchor the won is weakening in real time.

Layered on top of this is the Federal Reserve's monetary posture. Stronger-than-expected US employment and inflation data in the first half of 2026 have pushed markets to reprice rate cut expectations sharply downward, with some positioning for additional hikes. A higher-for-longer dollar environment compounds Korea's structural pressure. The won faces a double bind: geopolitical risk may have faded, but structural dollar demand and rate differentials have stepped in as the dominant downward forces.

The Feedback Loop That AI Created

What makes this situation analytically interesting — and economically troubling — is that the same force driving the won lower is also making AI investment more expensive for Korean companies, potentially restraining the very supercycle that set this in motion.

Large export-oriented conglomerates can manage this duality. Samsung and SK Hynix earn dollars by selling HBM to Nvidia and hyperscalers, then spend those dollars on ASML equipment. Their hedging is partly structural — dollar revenues offset dollar expenditures. Their won-denominated earnings look stronger with every tick upward in the exchange rate, providing short-term tailwinds even as investment costs rise in absolute terms.

The picture is far more painful for Korea's AI-native ecosystem: the cloud startups, AI application companies, and mid-size enterprises attempting to build on top of foundational models. These companies raise capital in won, generate revenue predominantly in won, and must pay for AI compute in dollars — whether renting GPU clusters from AWS, purchasing inference capacity, or acquiring hardware for on-premises deployment. When the won weakens from 1,300 to 1,500 per dollar, effective infrastructure costs rise by more than 15 percent without any corresponding increase in domestic revenues. For capital-intensive AI infrastructure, this is not a rounding error. It is a meaningful barrier to entry that concentrates the AI buildout among large exporters who already have natural dollar hedges.

The paradox crystallizes here: the AI supercycle is weakening the won, and the weaker won is suppressing AI investment capacity for everyone except the largest players who are already generating the dollar demand that weakened the won in the first place. Korea's AI ecosystem risks bifurcating — with globally competitive semiconductor exporters on one side, and a chronically capital-constrained domestic AI software sector on the other.

A New Variable the Old Models Cannot Capture

Korea's Ministry of Economy and Finance has tools for currency management — foreign exchange reserve deployment, smoothing operations, coordination with the Bank of Korea on monetary signaling. But none of these tools address the underlying structural dollar demand, because that demand is not a market anomaly to be corrected. It is the rational behavior of companies investing in technology that defines competitive position for the next decade. Slowing Samsung's HBM expansion to protect the won would be strategically incoherent.

This is where the analytical challenge lies for policymakers and economists alike. Classical current account theory treats a surplus as a structural floor for a currency. But the quality of the surplus now matters as much as its size. When export revenues are immediately recycled into dollar-denominated capital imports — as they are in Korea's AI-driven investment cycle — the surplus's ability to anchor the currency is diminished. Korea is running what might be called a pass-through surplus: dollars come in from chip sales and flow straight back out for chip equipment, with limited net accumulation to support the won.

Taiwan, the Netherlands, Japan, and any other economy deeply embedded in the advanced semiconductor supply chain are navigating versions of the same dynamic. The AI supercycle is not just a demand shock for chips — it is a structural shift in how technology-exporting economies relate to the dollar. For Korea, the 1,500-won level may not be a crisis to resolve so much as a new equilibrium to understand. The Strait of Hormuz may be open, but the pipeline of dollars flowing out to build AI infrastructure is running at full pressure. Nothing in the near term is likely to slow it.