The rate channel goes cross-border, then reverses at home

What happened

Japan's own version of Monday's story: a 10-year JGB push toward its highest yield since 1996 broke Tokyo's AI-proxy stocks together, with a sell-side analyst naming the mechanism explicitly. Nikkei 225 closed −2.54% to −2.56% (~67,460); SoftBank Group −8%+, Kioxia −9-13%, Arm −6% in US trading, Tokyo Electron −4.4%. A Daiwa Asset Management strategist: "higher yields will push up borrowing costs for hyperscalers, triggering market doubts over capital expenditure prospects." This is the same transmission channel this thread opened to watch, appearing a second time in five days on a different continent. (cryptobriefing.com, TradingKey)
Hours later, the US leg reversed via policy intervention rather than market stabilization. Treasury at least doubled its long-bond buyback size (10-20yr and 20-30yr sectors, $2B → "at least" $4B per operation), an unscheduled escalation. The 30-year fell from Tuesday's 5.32-5.34% intraday high to 5.196%, the 10-year to 4.647%, and US equity futures/stocks rallied. This directly answers the "what to watch next" note logged here 08-18 — the market-test of the next investment-grade order book never had to happen, because Treasury acted first. (CNBC) Watch: Whether AI-buildout debt is becoming its own underwritten, rated, publicly-watched risk category rather than a footnote inside each borrower's story. Track: whether rating agencies publish an explicit methodology for compute-collateralized paper, whether any of it prices wide of comparable real-asset debt, the first covenant breach or restructuring, and whether Nvidia's CDS moves with issuance volume. The open question is whether GPU compute behaves like an infrastructure asset with a 20-year life or like equipment on a 3-4 year depreciation clock — the two produce very different recovery rates in a default.

Sources 3

cryptobriefing.com cryptobriefing.com unrated
TradingKey tradingkey.com practices 4/7
CNBC cnbc.com high 0.85

Credibility ratings combine two public datasets: Lin et al. 2023, PNAS Nexus 2(9):pgad286 (CC BY 4.0) for the domain-quality score, and Wikipedia's perennial-sources list (CC BY-SA 4.0) for the reliability tier. A rating describes an outlet's general record, never this particular article. Unrated means absent from those datasets — not a judgement.