Nikkei -4.9%, KOSPI Crashes Through Circuit Breakers: How US AI Debt Fears Repriced Asia's Memory Complex

The Asian session on October 15 opened into a US tape that had already broken. The Bureau of Labor Statistics reported September CPI at 3.7% year-over-year, with energy pushing the monthly print up 0.6%. Within hours, the 10-year Treasury yield spiked to 5.32% — its highest intraday level since July 2007 — before settling at 5.189%, per WSJ's October 14 live coverage. That single yield move repriced the discount rate applied to every far-future cash flow on the AI infrastructure stack, and Asia's most rate-sensitive proxies absorbed the hit first.

The credit leg of the story had been building for weeks. CoreWeave — the AI cloud provider that is essentially a leveraged Nvidia financing vehicle — saw its unsecured junk-bond yields push toward 15% on October 14, while its equity fell 15.46% to $65.78 in a single session, per BigGo Finance. The company carries net debt of roughly $46 billion against an enterprise value near $84 billion and an interest expense that has already outpaced gross profit on a TTM basis, according to its latest 10-Q. Per Goldman Sachs Research, roughly one-third of the ~$600 billion in 2026 hyperscaler AI capex will be debt-financed — a ~$250 billion wall of refinancing that markets are only now starting to price.

The selling in Asia wasn't uniform — it reflected three different exposures to the same credit shock. Japan's Nikkei 225 closed at 64,334.68, down 4.9% (a 3,318-point drop), per the Associated Press dispatch. The Taiex fell 3.4% as chip foundries sold off into the night before TSMC's Q3 release. Korea's KOSPI closed down 7.57% at 5,554.22 — a six-month low — and tripped circuit breakers for the tenth time this year when it hit an intraday 9.29% decline, per Reuters via Central Oregon Daily. Foreign investors pulled a net ₩3.6 trillion ($2.68 billion) out of Seoul.

The damage inside Japan's and Korea's tape was concentrated in NAND flash and HBM — the two product categories most exposed to AI accelerator buildouts, and most exposed to a capex pause. Kioxia, which listed on the Tokyo Stock Exchange Prime Market in December 2024 with an offer price of ¥1,455 and trades at roughly 13x earnings on a TTM basis, saw 11.2% of its market cap evaporate in a single session. SK Hynix, the HBM leader whose 76% gross margin and 1.0x ROE have driven its 2026 valuation higher, fell 11.1%. Samsung — the only conglomerate on the list with material non-memory earnings buffer — still dropped 9.5%.

The contrast inside Japan's tape told the story. Advantest, the dominant HBM/SoC test equipment supplier, opened lower but closed up 1.66% at ¥41,610. Tokyo Electron — the deposition and etch leader whose tools run inside every leading-edge fab — also closed up 1.98% at ¥10,570. The reason: order books for advanced-node equipment are already booked into 2027, with 2nm ramps and CoWoS packaging expansion keeping WFE (wafer fab equipment) spending above the $140 billion level per SEMI's Q2 2026 data. A rate spike hurts; a cancelled order book does not.

TSMC's own Q3 results, released October 16, reinforced the foundry divergence. Net profit hit NT$763.1 billion ($23.59 billion), up 69% year-over-year; revenue reached NT$1.494 trillion with gross margin at 66.4%. CEO C.C. Wei struck a more cautious tone on 2027 — \"prudent in our business planning\" — but the company raised its full-year 2026 revenue growth forecast to ~44% (from >40%) and guided Q4 revenue to $52.3-54.3 billion, per Reuters. Capex was maintained at $60-64 billion for the year. Foundry earnings are now the structural counter-anchor to the AI credit story.

SoftBank Group is the cleanest single read on the AI credit cycle in Asia — and the October 15 drop of 9.6% to ¥4,852 (the lowest close since April 2025) reflected exactly that. The company's FY2025 annual report, filed in July 2026, shows cumulative OpenAI investment at $34.6 billion against a fair value of $79.6 billion — a $45 billion unrealized gain that already partially reversed when AI equities sold off in October. S&P revised SoftBank's outlook to Negative in March 2026 following additional OpenAI commitments, per S&P Global Ratings, and SoftBank has been issuing ¥1.12 trillion in domestic bonds plus $2.2 billion and €1.7 billion in foreign bonds over fiscal 2025 to fund its position.

Per The Kobeissi Letter, SoftBank is in the process of pricing what would rank among the largest junk-bond deals ever to fund a follow-on OpenAI investment. A widening of high-yield credit spreads — even by 100 basis points — flows directly into SoftBank's funding cost and, by extension, into the mark-to-market value of its OpenAI stake.

The October 15 sell-off did not hit every name equally because the AI supply chain has clear bottlenecks where demand cannot easily be cancelled. The transmission map runs in two directions:

Memory and AI-proxy companies entered this sell-off with very different balance sheets. Kioxia carries net debt/EBITDA of 0.02x and trades at 15.5x TTM earnings — its sell-off is pure multiple compression. SK Hynix has negative net debt and trades at 5.5x earnings, but its 76% gross margin and 100% return on equity are the most exposed to a capex pause. SoftBank, by contrast, carries net debt/EBITDA of 3.55x and a debt/equity ratio of 1.54x, with interest coverage at -0.5x — meaning operating cash flow alone cannot cover interest expense, which is why a credit-spread widening hits the equity first.

The rebound in CoreWeave — up 3.6% on October 15 to $68.12 — came only after Nvidia CEO Jensen Huang told Bloomberg in Stockholm that the company's \"strength is improving every day\" and its \"cost of capital is actually decreasing.\" That single sentence from the largest shareholder — Nvidia holds a substantial economic interest in CoreWeave — was enough to drive a $1.5 billion intraday rebound in a stock that had lost 15.5% the prior session. The episode exposed the entire AI complex to a new fragility: it now trades on the verbal backstops of a handful of large players, not on operating fundamentals alone.

Three near-term catalysts will determine whether October 15 is a one-session shakeout or the start of a deeper Asian deleveraging cycle.

The IMF warned this week — per The Guardian's October 15 reporting — that an AI share-price shock would ripple globally and put financial stability at risk, with Asian markets most exposed. October 15 was a small test of that thesis: a US credit event translated into a -7.57% KOSPI close and a -4.9% Nikkei within 12 hours, with foreign outflows in Seoul alone at $2.68 billion.

The structural question over the next 12-36 months is whether the AI buildout decouples from the credit cycle. If the roughly one-third debt-financed share of the ~$600 billion 2026 hyperscaler capex cannot be refinanced at acceptable spreads, the upstream Asian order book compresses — hitting Tokyo Electron, Advantest, TSMC, and the memory complex in sequence. If it can, the same names re-rate back toward the order-book visibility they had entering October. Either way, Asia is now structurally a beta-2 trade to the US high-yield credit market — and that is a linkage investors will price into every session from here.