The headline numbers coming out of north-east Asia in 2026 are the best in a generation. Taiwan grew 8.63% in 2025 and expanded at an annualized 13.69% in the first quarter of 2026. South Korean chip exports rose 180% year on year in the first twenty days of July, to $22.1 billion, with computer-related shipments up roughly 232%. Japanese exports are surging and the Nikkei has spent the year near records.
Strip out one sector and all three economies are contracting.
This is the most important structural fact about the region, and it is almost entirely obscured by aggregate data. Chips and AI-related equipment now account for more than 40% of South Korean exports, more than double their share two years ago. In Taiwan the figure is 80%, against roughly half before the pandemic. Once firms that official statistics fail to classify as AI-linked are properly counted — Japan’s Advantest, which makes chip test equipment, or Taiwan’s Foxconn, which assembles data-center servers — the entire 15% rise in regional industrial output since 2019 is attributable to AI. Output at factories making anything else has fallen. Excluding semiconductors and AI servers, Taiwanese exports are down 40% since 2022.
Three of the most sophisticated manufacturing economies ever built have, in four years, become single-product exporters.
This is displacement, not decline
The temptation is to read the shrinking non-AI base as decay — aging plant, lost discipline, complacency. That reading is wrong and it leads to the wrong forecast.
What happened is that China became simultaneously competitive across the entire middle of these countries’ export baskets. South Korea’s export mix overlaps more closely with China’s than almost any other economy in exactly those product categories where Chinese market share is expanding fastest. Japan ranks third on the same measure, behind Vietnam. The middle of the north-east Asian industrial stack was not outcompeted gradually by one rival in one category. It was removed by one rival in every category at once.
Korean petrochemicals is the cleanest illustration because it is being liquidated in public. Ethylene that once approached $1,400 a tonne now trades below $1,000, after Chinese capacity expansion and import substitution destroyed the cost position that kept aging naphtha crackers viable. Seoul has responded with two rounds of state-brokered restructuring: a program to cut 3.7 million tonnes of cracking capacity, the closure of two Yeochun NCC ethylene plants, and a three-way merger of Yeosu assets. Once the Ulsan complex is dealt with, roughly one-third of South Korea’s total ethylene capacity will have been eliminated. Japan is in its third restructuring cycle since 2000, with capacity down from a 7.5 million tonne peak to just over 6.1 million. Neither country lost the ability to crack naphtha. Both lost the right to charge for it.
The same pattern runs through displays, where Korean LCD is gone and OLED is under sustained attack; through steel, squeezed by Chinese overcapacity; through batteries, where the shift to LFP chemistry and Chinese scale has hollowed out the Korean cell makers’ cost advantage; and through Japanese autos, where the EV transition arrived faster than the incumbents’ product cycles and where a run of certification and quality-falsification scandals across Daihatsu, Toyota, Hino and IHI did real damage to the one asset Japanese manufacturing could not afford to spend.
None of this is visible in national export data, because a single sector grew fast enough to cover all of it.
The boom does not reach the household
The second structural fact is that the AI sector is too small, in employment terms, to distribute what it earns.
Taiwan’s semiconductor industry employs roughly 300,000 people out of a workforce of 11 million. The broader electronics and IT manufacturing sector employs about a million. Services employ around seven million. Yet semiconductors are more than 20% of GDP, TSMC alone is more than 40% of the stock market, and technology pay running near double the national average sets housing prices and salary expectations for everyone. Taiwan’s central bank governor has a term for the result: a K-shaped economy. Real average wages grew 1.4% in 2025 and median wages 1.35%. Roughly 70% of Taiwanese earn below the average, which is what happens when a small, extremely well-paid cohort drags the mean away from the median.
The regional version of this is starker. Private consumption is 53% of output in Japan, and closer to 40% than to the 60% rich-country average in South Korea and Taiwan. These are aging societies, where the ratio of consumption to output should be rising mechanically as retirees stop producing and keep spending. It is not. The surplus is recycled into export capacity and outbound investment instead of into domestic demand.
Which means the household in these economies has no wage channel through which to participate in the boom its own country is manufacturing. It has only one channel: the equity market.
Korea already ran the experiment
What happens when the only available channel is leveraged is now a matter of record.
Since single-stock leveraged ETFs were approved in Korea in May, domestic retail bought a net 14 trillion won of them, against roughly 2 trillion won from foreign investors. Assets in the 25 largest leveraged Korea-focused ETFs rose from about 15% of the category in January to around 30% by June. Retail accounted for roughly 70% of trading volume in a $4.3 trillion market, on record margin debt.
Then the tape turned. Samsung fell 24.3% and SK Hynix 19.5% in the month to July 16. The KODEX SK Hynix Single Stock Leverage ETF fell about 70% from its June peak and roughly 50% from its May 27 debut, which is what daily rebalancing does to a two-times position in a volatile market and precisely what the product is engineered to do. Forced liquidations ran to 2.3 trillion won over about two and a half months. The exchange tripped seven market-wide circuit breakers in 2026, more than half of all such events since the mechanism was created. The KOSPI fell 25% from its June high. The Bank of Korea flagged record leveraged retail positioning concentrated in semiconductors. Oxford Economics downgraded Korean equities to neutral. The head of the Financial Supervisory Service publicly regretted approving the products, and regulators then raised the cash requirement to trade them tenfold.
The standard reading is a retail speculation story. It is not. It is what an economy looks like when 40% of exports are one sector, wages do not transmit the gains, and households are handed a two-times daily-reset instrument as their only way in.
The July deals make the concentration permanent
On July 24 and 25, at a state-attended AI summit in San Francisco, the region signed away the next five years.
Nvidia and SK Group announced letters of intent covering more than $500 billion, centered on securing SK Hynix memory supply for Nvidia and co-developing next-generation HBM. SK Telecom will build a two-gigawatt AI factory on Nvidia’s Vera Rubin systems and SK Hynix HBM4, first facility due 2027. Nvidia is putting $1 billion into Naver and expanding its data centers with Brookfield. SK Hynix intends to double wafer capacity by 2030. Samsung signed a memorandum with Broadcom worth more than $200 billion through 2030, covering HBM4 and HBM4E, 2-nanometer and below foundry at Pyeongtaek, and 2.3D and 2.5D advanced packaging.
Read against the narrowing data, these are not a rebuttal. They are the confirmation. Korea has just committed a larger share of its industrial base to a single end-market, for five years, with the demand curve owned by two American counterparties, at a moment when a third of its ethylene capacity is being scrapped. Korean foreign direct investment into the United States rose more than 100% year on year in the first quarter, to $10.2 billion — the physical plant increasingly follows the customer. And both agreements are letters of intent and memoranda, not committed volume at fixed price. They are frameworks announced during live trade negotiations, and they are the first thing to be repriced if the political context moves.
What actually breaks it
Three things, in ascending order of importance.
The comparison base goes first. Contract DRAM prices rose 90 to 95% quarter on quarter in early 2026 and NAND more than 50%. A 180% export figure built on that is mostly price, not units. By the second half of 2027 the year-over-year comparison cannot be cleared without further escalation, which is exactly what the announced capacity expansions are designed to prevent.
Then the cushion problem. Any deceleration in AI capital spending lands on economies that no longer have a second sector to fall back on. In 2022, a chip downturn in Taiwan was absorbed by the rest of the export base. That base is now 40% smaller. The same shock arriving in 2028 has nothing underneath it.
Last, and least priced: China is still moving. Its exports are shifting away from consumer goods toward intermediate inputs and capital goods, categories that have risen 26% and 32% over three years. The displacement that removed the middle of the north-east Asian industrial stack is not a completed event. It is a process working upward, and the next categories in its path are the ones these three countries still own — commodity DRAM through CXMT, NAND through YMTC, mature and eventually advanced logic through SMIC and its successors. The AI franchise is the last thing left, and it is not permanently defensible.
Japan, South Korea and Taiwan have not forgotten how to build things. They have lost the right to be paid for building anything except one. And in the last week of July, two of them signed five-year agreements confirming it.