Record semiconductor exports from Korea, Taiwan and Singapore rest on AI buyers funding their build-out with debt
South Korean semiconductor shipments reached $44.82 billion in June, a 199.5% increase on a year earlier and the first month above $40 billion. Total exports cleared $100 billion for the first time. First-half chip exports of $192.4 billion already exceed the full-year record of $173.4 billion set in 2025. And what has changed alongside this increased volume is how the buyers are paying for it.
The largest customers behind those orders have moved from funding capital expenditure out of operating cash flow to funding a growing share of it with debt. That shift reaches suppliers in Seoul, Hsinchu and Singapore more directly than the equity volatility that drew attention when the Kospi fell almost 10% in a single session in late June and SK Hynix shed more than 12%.
U.S. semiconductor names came under pressure of their own in early July, with Intel down about 21% in a week on reports its 18A process would ramp later than expected and a broad AI-stock sell-off. A repricing in shares compresses valuations. That makes the durability of Asia’s export boom a question about credit markets it never sold into: if the debt funding the build-out reprices, the orders reprice with it.
Borrowed Capital Expenditure
The Bank for International Settlements documented the change in March 2026. Gross bond issuance by the large U.S. technology firms building AI infrastructure topped $100 billion in 2025, most of it at maturities beyond five years to lock in funding for multi-year projects. Credit default swap spreads on those issuers widened over the same period, with the largest moves among the lower-rated names.
Bond markets are only part of it. The BIS also traced a second channel, in which a dedicated joint venture or special purpose entity acquires data centre assets, raises debt through private placements, and services it from lease payments made by a technology firm that holds a minority stake and commits to long-term offtake. The BIS described these arrangements as “shadow borrowing”: obligations economically akin to debt that sit largely outside corporate balance sheets, held by private credit funds and insurers, with banks providing funding lines to the vehicles.
The scale is visible in ratings commentary. Moody’s Ratings estimated in July that six hyperscalers will spend roughly $785 billion in 2026 and close to $1 trillion in 2027, against $821 billion in leases and commitments that have not yet begun and therefore do not appear on their balance sheets. A “material shift in the structure of their balance sheets is becoming evident,” the agency wrote, and it flagged the web of cross-investment between chipmakers, model developers and cloud providers as a concentration and credit risk if demand for AI products fails to materialise. Morgan Stanley expects AI-related debt issuance to approach $570 billion this year, with a growing share moving into private credit as public bond demand softens.
Asia’s Own Commitment
Asian suppliers are matching that spending with fixed capital of their own. SEMI raised its 2026 forecast for global semiconductor equipment sales in July to $165.9 billion, a 23.2% annual increase, and projected a record $229.5 billion by 2028. Wafer fab equipment accounts for $143.9 billion of the 2026 total, with DRAM equipment alone rising 39% to $38.8 billion on high-bandwidth memory demand. First-quarter billings had already set a record at $36.55 billion. China, Taiwan and Korea remain the three largest destinations for that spending.
Equipment orders of that size convert into fabs, packaging lines and test capacity that take years to build and longer to depreciate. Suppliers are committing long-lived assets against an order book whose funding is increasingly conditional on the willingness of bond investors and private credit funds to keep refinancing someone else’s build-out. That is a different type of exposure from the one most of these companies have managed before, and it does not show up in a backlog figure.
The concentration runs deeper than chips, and it is not confined to Korea. Singapore’s non-oil domestic exports rose 38.4% year on year in May on a 94.8% jump in electronics, with integrated circuits up 80.9% and personal computers up 140.9%, and shipments concentrated on Taiwan, South Korea and the United States.
Korean shipments to the United States rose 78.6% to $20.02 billion in June and to China 92.1% to $20.03 billion, both attributed by the trade ministry to expanded AI server investment. Even steel exports turned positive for the first time in 14 months, up 9.6%, which the ministry linked partly to construction demand from data centre building. A single spending programme is now reflected across several unrelated lines of a national export account.
The Demand Test
While there remains obvious optimism about the economic potential of AI, end demand has not yet caught up with the capital being deployed against it. OpenAI, the sector’s reference point for AI revenue, posted a net loss of $38.5 billion for 2025. A 2026 survey of 1,550 enterprise AI decision-makers by Publicis Sapient found that 73% report using AI regularly or across most processes, while only 10% describe it as core to how their business operates. Adoption that broad and that shallow is consistent with continued spending, and equally consistent with a pause once boards begin testing returns.
For directors and long-term holders inside the supply chain, the practical question is the sequencing of their own funding. A concentrated holder of listed supplier shares can raise cash through financing secured against an existing equity position rather than against next quarter’s shipment numbers, which separates a balance sheet from a single customer’s capex decision. Because such financing is typically non-recourse and not subject to a margin-call cascade, it is a comparatively conservative way to raise liquidity in precisely this kind of environment. The holder keeps the long-term exposure and is not forced to sell into a repricing.
Spreads may be the more useful indicator here than share prices. Widening CDS on the lower-rated buyers of Asian components, or a slower quarter of private placements into data centre vehicles, would register in order books before it registered in an index. The June session in Seoul showed how quickly sentiment travels. For holders who can see the difference between a demand shock and a funding one, the useful move may not be to guess the timing, but to arrange funding that doesn’t depend on it.
