Hot CPI Lit the Fuse on the First AI-Credit Rout

September CPI (released Wed): headline +0.6% m/m, 3.7% y/y (vs 3.6% expected), up from 3.4%; core +0.3% m/m, 2.6% y/y (vs 2.5%) — core accelerating after August's 2.4% was the best reading in years. Gasoline +9.1% in the month, energy +20.4% y/y — gasoline alone drove over half the headline jump (BLS).

What happened

Markets: Nasdaq −1.88% (worst day since July), KOSPI −7.57% with a circuit breaker, Nikkei −4.90%, Taiex −3.41%, SOX index −4.3% (sixth straight loss). TSMC posted a sixth consecutive record-profit quarter and raised capex guidance — and still fell ~2%. That's the tell: this isn't a demand scare. The transmission mechanism is credit, not demand. Oracle CDS spread hit a record 288 bp, SoftBank 424 bp, CoreWeave above 1,000 bp (it carries $22B bonds + $26B loans). Reuters' framing, which multiple outlets repeat: the market is no longer asking whether AI demand exists but whether debt-financed infrastructure can clear at rising rates. ~$80B of market cap erased from the most levered AI names in one day. Yesterday's check — "do hike odds hold above 30%?" — resolved: no, and that's the interesting part. Pricing spiked to ~50% for an October hike on the CPI print, then was talked back down to 27–40%

The caveats that matter

Bowman and Daly both signaled the Fed "can wait." Daly: more hikes "more likely than not," but "I don't feel urgent." Analysts (ANZ's Khoon Goh) openly attribute the paring to the Oct 27–28 FOMC landing days before the Nov 3 midterms — the market is pricing political reluctance, not economic analysis. A separate structural oddity flagged by Franklin Templeton/CPRAM: there will be no year-over-year CPI figure in November (the October 2025 print was skipped in the government shutdown), so the Dec 8–9 FOMC decides on a thinner data base than usual.

Also noted

Verified by. BLS figures consistent across CNBC, Reuters-sourced reports, Yahoo Finance, Moneycontrol, sedaily. CDS and index numbers via Reuters/economymiddleeast.

Discrepancies flagged. (1) Hike-odds numbers vary by snapshot time (50% → 27% → 40%) — cite ranges, not a point. (2) The "$80B erased" figure is Reuters' estimate over a defined basket; treat as indicative. (3) Core at 2.6% vs headline 3.7% is a genuine energy-shock-vs-broad-inflation dispute: Commerzbank reads it as "strength of the economy + energy pass-through"; the Fed's dovish camp reads it as a Hormuz/diesel artifact that fades if the strait settles. Both readings are defensible; the policy choice depends on which one wins.