〔Overnight Brief〕5% Yield Test, 69% September Hike Odds, and a $9B vs $163B Asymmetry — The Long End Forces the Tape Into the Weakest Month

0. Overnight Arc

The long end is forcing the question. JPMorgan's Grace Peters flagged the 10-year at 4.8% pushing into the 5% zone — historically an adverse threshold for equities — and the 30-year at a 19-year high as the binding constraint for global stocks entering a seasonally weak September ^[1][2]. Apollo's Torsten Slok pinned the latest yield leg on the Iran war and tariffs, not US fiscal worry, and the market now prices a 60-69% probability of a Fed hike at the September meeting ^[3][4]. Yet the equity side is more asymmetric than the rates side: BofA's systematic-strategy book has only ~$9B of residual upside-buy capacity against up to $163B of CTA and vol-control selling on any drawdown ^[5]. The arbitrage is in volatility, not direction ^[5].

1. The Long-End Pressure Test

• **[ESCALATED] JPMorgan — Grace Peters (chief global equity strategist):** rising yields are the "key risk" to global equities into the seasonally weak September; she flags a possible 5-8% pullback as healthy, not structural, ahead of the November US midterms ^[1][2]. The 10-year sits at 4.8%, the 30-year at a 19-year high ^[1].
• **[ESCALATED] Apollo — Torsten Slok (chief economist):** yields will push higher on the Iran war and tariffs, with the move having "almost nothing to do with US fiscal conditions" ^[4]. Treasury Secretary Bessent's expanded buyback program has failed to cap the long end ^[4]. Slok: "Rates are restrictive for housing, but not for AI" ^[4].
• **[NEW] TD Securities — "Election-year hikes and long-end rates risks":** the framing is now about cycle risk in the 10-year/30-year even as policy leans hawkish into the run-up ^[6].
• **[ONGOING] Bessent buyback failure mode:** long-end yields have re-elevated despite expanded Treasury buybacks aimed at compressing term premium ^[1][4]. Markets are increasingly guessing policymakers will be forced to hike, returning yields to pre-buyback levels ^[1].
• Disagreement to watch: the September hike-probability band is 60% (East Securities Hong Kong, post-Warsh Jackson Hole) to 69% (Apollo / market pricing) — quote as a band, not a point ^[3][4].

2. September Equity Setup — Asymmetric to the Downside

• **[NEW] BofA systematic flow model:** the upside scenario leaves only ~$9B of net systematic buying; a flat tape turns to ~$10B of net selling; a drawdown unlocks up to $163B of CTA + vol-control selling ^[5]. The asymmetry is the headline, not the point estimate.
• **[NEW] Nomura's McElligott (via BofA relay, single source — flag thin):** the current vol collapse is "epic" in scale; he flags the risk of outsized P&L losses as positioning unwinds ^[5].
• **[NEW] BofA — buyback blackout:** the September expansion of the corporate buyback blackout thins the buy-side cushion on any dip, amplifying CTA selling impact ^[5].
• **[NEW] JPMorgan — Dorothee Blessing (co-head of global investment banking):** dealmaking is on track for a record year as M&A and IPO volumes surge in the final months of 2026 despite macro concerns ^[7]. Constructive counterweight to the rates-and-positioning bear case.
• Falsifiable test: a 10-year print above 5% paired with a single weak tape day in the first half of September would mechanically trigger the $163B vol-control cascade; a soft non-farm payrolls print and a benign CPI pair is the offset ^[5].

3. AI Stack Reads: Hardware, Software, Meta, Nvidia

• **[NEW] Morgan Stanley on Meta:** with the social-media litigation overhang resolved, the product pipeline — MetaClaw/enhanced Meta AI, an SMB ad-agent suite, a new subscription, an API product, and neocloud options — can accelerate ^[8]. Re-rating mechanism is legal-cloud removal, not the news itself.
• **[NEW] New Street — Pierre Ferragu on Nvidia (single-analyst, single-house view — flag thin):** $400+ stock price "very likely" within 18 months ^[9]. Treat as a target, not a path.
• **[NEW] Goldman Sachs — semis conference preview:** ahead of Communacopia + Technology, five core topics — AI compute, wafer fab equipment (WFE), memory, analog, EDA — and the house stays constructive on AI infrastructure spend and the semiconductor cycle ^[10].
• **[ESCALATED] STAR-assets dot-com replay piece (Sina vReport broker index):** current AI positioning is closer to the 1998-1999 phase than the 2000-Q1 top; infrastructure earnings still deliver, software/application monetization still unproven ^[11]. Watch items: cloud capex / AI-revenue match, hardware order quality, application monetization, FCF, broad earnings-revision trends ^[11].

4. China A-Shares: Education, GPUs, STAR Framework, and Dealmaking Tailwind

• **[NEW] First Financial "Yi Rating" — Muxi (688802.SH):** 5-star initial rating, but DCF and DDM both show current market cap above intrinsic value; rare domestic third-party GPU designer, but cloud self-developed chips and peer competition are the long-term moat risks ^[12].
• **[NEW] First Financial "Yi Rating" — International Composites (301526.SZ):** 4.5-star initial rating, valuation above intrinsic; AI-server and advanced-packaging electronic-fab demand is the upside, but capacity additions and AI capex cyclicality are flagged ^[13].
• **[NEW] Strategy — September A-share allocation (Cailian Press commentary):** policy floor plus AI catalyst; top institutions' latest earnings show ~10x repair; fundamentals and new tech as dual support; framed as the "starting point of style rebalancing" ^[14].
• **[ONGOING] BOC International Securities September gold-stock basket (via Sina vReport index):** focused on AI infrastructure (PCB, CPO, liquid cooling, semis), non-ferrous/precious on weak-dollar and commodity reflation, and innovation drugs / AI pharma; specific names: HNA Holdings, Satellite Petrochemical, Anji Microelectronics, Shangtai Technology, Mindray Medical, Wedo, SDP Holdings, Feilihua ^[15].
• **[ONGOING] Bank of Korea:** second consecutive 25bp hike, to 3.00%, driven by AI-led semiconductor boom and Q2 GDP +3.7% y/y ^[3]. Regional read-through for Asia tech and the A-share semis basket.
• **[ONGOING] STAR framework (Sina vReport index):** the rates-vs-tech-cycle relationship is not linear; the current AI cycle is closer to 1998-1999; the watch items are cloud capex / AI-revenue match, hardware order quality, application monetization, FCF, and broad earnings-revision trends ^[11].