AI and tariffs redraw South Korea’s trade map: Memory flows through Taiwan as Chinese cars flood the domestic market

The Bank of Korea’s 2026 report, “Map of Production and Supply Chains for South Korea’s Major Manufacturing Industries,” documents a pair of trade reconfigurations that are structurally distinct but share the same bilateral relationships. On one front, AI demand reroutes South Korean semiconductor exports through Taiwan. On the other, US policy severs the electric vehicle export channel while Chinese manufacturers fill the domestic market space left behind.

South Korea sits at the intersection of both shifts, adjusting its two largest manufacturing sectors on the same time scale but in opposite directions relative to the same trading partners.

The AI chip supply chain that routes South Korean memory through Taiwan

Taiwan-bound semiconductor exports rose from $12.78 billion in 2022 to $36.77 billion in 2025, nearly tripling. Taiwan’s share of South Korean chip exports climbed from 9% to 19.9%, moving from the fourth-largest export destination to the second. China remained the largest single destination, but its share fell from 53.1% to 40.3%.

The mechanism behind Taiwan’s rise is not buyer substitution—South Korean customers switching from Chinese to Taiwanese end-users—but a structural routing effect imposed by the AI chip supply chain’s physical geography. Nvidia designs graphics processing units. TSMC produces and packages them. Samsung Electronics and SK hynix supply the high-bandwidth memory that goes into those chips. Because TSMC is the sole advanced-packaging node for Nvidia’s AI chips, South Korean high-bandwidth memory must physically reach Taiwan for integration regardless of where the final chip ends up. The growth in Taiwan-bound chip volume is a consequence of AI architecture, not trade policy.

South Korea’s semiconductor production base reinforces this dependency. Domestic output rose from 74 trillion won ($50.4 billion) in 2014 to 210.8 trillion won ($143.6 billion) in 2024, doubling its share of manufacturing output from 5% to 10.1%. Production is concentrated 82.3% in the greater Seoul area—the Samsung Electronics plants in Hwaseong, Pyeongtaek, and Giheung, and the SK hynix facility in Icheon. These are sunk costs that cannot be cheaply repurposed to conventional memory for a China-only market. The facilities were built for the AI chain’s specifications.

Two non-obvious consequences follow. First, because the total semiconductor production base grew dramatically over the period, the 40.3% China share in 2025 sits on a much larger total export volume than the 53.1% share in 2022. At Taiwan’s 9% share and $12.78 billion in 2022, total semiconductor exports ran approximately $142 billion that year, putting China-bound volume at roughly $75 billion. At Taiwan’s 19.9% share and $36.77 billion in 2025, total exports ran approximately $185 billion, putting China-bound volume at roughly the same $75 billion. The share dropped 12.8 percentage points. The absolute volume barely moved. The shift was the denominator—a much larger total base being pulled toward Taiwan by the AI pipeline—not China substitution.

Second, South Korean HBM suppliers depend on a single packaging node—TSMC—that they do not control. The dependency is asymmetric. Nvidia has the widest exit options: it can dual-source high-bandwidth memory from Micron, though with a 12-to-18-month qualification delay that would cede market share to AMD or custom chips. Samsung and SK hynix sit in the middle: they could sell to Google, Amazon, or Microsoft as alternative AI-accelerator customers, but the sunk-cost concentration of their production in Gyeonggi locks them into the Nvidia-TSMC architecture for the duration of the AI demand cycle. TSMC has the fewest exit options—it cannot replace Korean high-bandwidth memory with in-house or third-party memory at the same node sophistication.

How US policy severed the EV export channel

The connection between US policy and South Korea’s EV export collapse is the most directly stated causal relationship in the Bank of Korea’s source data. The central bank identifies three mechanisms—US subsidies favoring domestically produced EVs, local production requirements, and tariff barriers—which “weakened the competitiveness of electric vehicles manufactured in South Korea and shipped directly to the United States.” The US share of South Korea’s electric vehicle exports collapsed from 33.6% to 5.2% between 2022 and 2025.

The United States moved first. The Inflation Reduction Act, signed August 16, 2022, with its local-content rules, constitutes a commitment device with institutional backing—legislation that Korean automakers answered with capital deployed in US factories. South Korea’s best response to the tariff-and-subsidy structure was to accept sunk-cost US factory investment rather than eat indefinite margin compression on direct exports. The equilibrium is self-reinforcing: once Korean firms invest billions in US production capacity, those sunk costs lock in continued US operations. Under a one-shot framing, Korean automakers would exit the US EV market rather than incur costly relocation. But the repeated-game nature of annual trade flows, combined with the long-lived asset structure of auto plants, sustains the onshoring equilibrium through trigger-strategy logic: the threat of future exclusion from the US market outweighs one-time exit savings. Korean automaker organizational inertia—the largest chaebol facing political sensitivity over abandoning domestic employment—constrains how aggressively this relocation can proceed, but the direction of motion is set.

South Korean automakers added a parallel hedge. Hybrid vehicle exports rose from 11.6% to 20.4% of finished-vehicle exports between 2022 and 2025. The Bank of Korea attributes this to two causes operating simultaneously—slowing electric vehicle demand and insufficient charging infrastructure—both pushing automakers to offer vehicles combining electric motors with internal combustion engines. The hybrid pivot is manufacturer adaptation to structural barriers on the supply and infrastructure sides, not a market-driven preference shift.

The United States remained South Korea’s largest auto export market overall at $30.15 billion, representing 41.9% of finished-vehicle exports. The EV segment was severed, but the broader export relationship survived—reoriented from direct EV shipments toward hybrid exports and US-based production.

Chinese vehicles captured more than a third of the import market

While South Korean automakers shifted capacity toward the United States, Chinese manufacturers filled the domestic market space left behind. The Bank of Korea reports that Chinese-made vehicles rose from 3.5% of South Korea’s automobile import market in 2022 to 37.2% in 2025, accounting for roughly 70% of electric vehicle imports by value.

The mechanism is an integrated supply chain covering vehicles, batteries, and components from a single ecosystem—a cost structure advantage that goes beyond simple subsidization. South Korea’s automakers left the domestic low-cost EV segment undefended while they pivoted toward the US market. China’s price competitiveness did the rest.

South Korea is the leading indicator of what happens when Chinese EV overcapacity meets an open import regime. The United States and European Union both face their own tariff barriers on Chinese EVs. The EU imposed definitive countervailing duties on Chinese battery electric vehicles effective 30 October 2024, with rates reaching as high as 35.3% above the 10% baseline tariff. South Korea has not yet deployed a comparable barrier. The 37.2% import share is not locked in by any policy commitment. Seoul could follow the EU precedent and impose countervailing duties at any time, collapsing the equilibrium in the import segment. This outside option is balanced against a welfare trade-off: Chinese EVs lower Korean consumer prices and accelerate domestic adoption, and any protectionist response would risk Chinese retaliation in other trade categories such as intermediate goods or rare earths. The absence of any observed Korean policy action, legislative proposal, or commitment device, and the absence of public debate in the National Assembly on Chinese auto import restrictions, makes inaction the political baseline.

What holds the two-front equilibrium together—and what breaks it

The semiconductor reconfiguration and the auto reconfiguration operate on different timescales and respond to different instruments, but they share the same bilateral trade relationships—South Korea with the United States and China. The combined equilibrium holds only while three conditions persist: the AI demand cycle continues, China remains at least one memory-process generation behind in high-bandwidth memory, and the US tariff environment stays predictable enough to justify multi-billion-dollar factory investments. Remove any one, and the architecture shifts.

China’s HBM gap is the most dynamic condition. ChangXin Memory Technologies (CXMT), the Chinese DRAM manufacturer, delivered HBM2E samples to customers in the first half of 2025 and targets mass production by mid-2026, with HBM3 production aimed at the end of 2026. If CXMT reaches qualification parity with Samsung and SK hynix at the HBM3 level, the division of labor that routes Korean memory through Taiwan faces its first structural stress test.

The auto import segment is the most vulnerable point in the structure. The 37.2% Chinese import share rests on no policy commitment and could be collapsed by Korean countervailing duties at any time. South Korea’s choice to deploy or not deploy that lever will determine whether the import surge is a temporary disequilibrium or a permanent structural shift.

The US policy environment carries its own fragility. If future administrations reverse course on EV subsidies or tariff enforcement, Korean firms that invested billions in US capacity face stranded assets. The sunk-cost commitment device cuts both ways: it locks in today’s cooperation under today’s policy, but the policy itself is reversible.

The Bank of Korea’s production data runs through 2024; its trade-flow data extends to 2025.

Analytical techniques used in this piece

This analysis applies the methods below. Each links to a short, plain-English explainer you can read and reuse.

Relationship Mapping

Extracts the network of ties among people, institutions, and entities.

Stakeholder Mapping

Charts the parties to a situation — their interests, power, and alignments.

Strategic Interaction (Game Theory)

Models a situation as a game — players, moves, payoffs, and likely equilibria.