〔Day Digest〕HSBC Re-Frames the 1997 Parallel: Asia's Real Risk Is AI Demand Fragility, Not Financial - A-Share H1 Profits +16.8% vs 9月 Fed Hike Reflating on US-Iran Strikes

0. Daily Arc

HSBC's Frederick Neumann re-frames the 1997 parallel: high UST yields, weak yen, tech euphoria remain, but Asia's structural risk has migrated from 'financial fragility' to 'demand fragility' via US AI hardware dependence ^[1]. The A-share tape runs hot in the opposite direction: H1 net profits +16.8% with Q2 alone +25.9%, led by electronics, communications, resource cyclicals, and non-bank finance ^[2]; UBS's Meng Lei lifts 2026 profit growth to 15% ^[3]. Geopolitics has re-priced the dovish trade: US-Iran strikes lift oil, Warsh stays hawkish, and 9月 Fed hike odds rebuild ^[4][5]. Net: an AI capex supercycle in tension with a re-tightening policy path.

1. Macro & Geopolitics

• **[NEW] HSBC (Frederick Neumann, Aug 31 report):** current Asia 'highly similar' to 1997 (high UST yields, weak yen, tech optimism) but core risk has shifted to 'demand fragility' via AI hardware demand dependence; Korea, Japan, Singapore most exposed ^[1]. Cites 1997 parallels: 10Y UST from 5% (Oct 1993) to ~8% (Nov 1994), still ~7% in April 1997 ^[1].
• **[NEW] Song Xuetao (Citic):** rate hikes cannot save the long end - AI is a national investment with high financing needs that won't stop; oil and AI are supply/structural problems, not policy-fixable ^[5]. Fed 'mirror in the mirror' with market self-reinforcing; recommends Fed 'speak less, err less' ^[5].
• **[NEW] Everbright Futures (on US-Iran strikes):** oil spikes, gold pulls back; ISM manufacturing 54.6 (vs 55.2 expected, 2022 second-highest), new orders 53.7 (weakest since March), prices index flat at 71.1; JOLTS 7.27m (vs 7.31m expected), mfg vacancies 2023-12 high, layoffs Jan low ^[4]. Warsh hawkish + oil = '9月 rate hike can't be avoided'; gold short-term correction but medium-term uptrend intact ^[4].
• **[NEW] Macro Sept outlook (vReport):** US labor-market deterioration the tell; 'K-shaped divergence' with rate-hike-odds retreat + dollar asset premium decline as the through-line; Warsh's hawkish Jackson Hole protects Fed credibility but volatility rises ^[6]. 'Once the market has tried rate-hike-odds retreat, it cannot return to one-sided tightening' ^[6].
• **[NEW] Global research August monthbook:** global equities still positive; ranking Japan > US > MSCI EM Asia > Europe; software, materials, autos, pharma outperform; retail, utilities lag ^[7]. Overweight泛 AI capex; defensive preference for pharma; cautious on consumer ^[7].

2. China A-Shares: Earnings Inflection & Allocation

• **[NEW] Kaiyuan Securities (H1 panorama):** 5,521 of A-share companies reported by Sep 1 (99.95% disclosure); H1 revenue +7.2%, net profit +16.8% (vs Q1: +2.5pct, +9.3pct acceleration); Q2 single quarter: revenue +9.2%, net profit +25.9% (+19.1pct acceleration) ^[2]. Ex-financials-and-real-estate: H1 +15.8%, Q2 +19.0% ^[2]. High-growth concentration: non-ferrous, petrochemicals, basic chemicals, coal, electronics, communications (ex-3 telcos), non-bank finance ^[2].
• **[NEW] UBS Securities Meng Lei:** raises 2026 A-share profit growth to 15%; sees steady rebound on earnings recovery, fund inflows, global tech narrative ^[3]. Two risks: geopolitics/UST/oil vol on rate-sensitive tech; 'redemption pressure' risk from public funds' tech overweight ^[3].
• **[NEW] September broker top picks:** 26 brokers, 249 picks across 188 A-shares (main board 119, ChiNext 32, STAR 34, BSE 3); WuXi AppTec, CATL, Zhongji Innolight top popularity (7, 6, 5 recommendations) ^[8]. Fastest climbers vs Aug: Tianfu Communication, Roborock, Rongsheng Petrochemical (+5, +5, +3) ^[8].
• **[NEW] CITIC Construction Investment (broker morning call):** September balanced - retain AI compute, non-ferrous, innovative drugs, export manufacturing; rotate to under-owned bank/insurance/securities, batteries, industrial metals, energy metals, basic chemicals, steel, hog ^[9].
• **[NEW] CITIC Securities (since late July):** three factors behind DM sovereign yield surge - inflation expectations, sovereign credit concerns, DM central bank repricing; favors dividend sectors (banks, utilities, telecom, property mgmt); CN-US yield differential widening eases RMB pressure and supports HK dividend assets ^[10]. Also: REITs oversold opportunities from liquidity/sentiment swings; turnover not yet expanding on Aug rebound ^[11].
• **[NEW] A-share static framework (Sept):** all-A PE 21.9; most broad-index PE percentiles above 50%; ChiNext at 30%; 'sentiment continues to cool' with turnover and trading-vs-high ratios both falling; fund flow mixed (margin balance up, international broker indicator up, 3-body flow down) ^[12].

Single-Name Notes

• **[NEW] Orient Securities on Aerospace Electric (航天电器):** Q2 net profit 0.076bn yuan +88.98% YoY, H1 0.128bn +47.73%; high-speed and liquid-cooling interconnect orders strong; military + strategic emerging dual-driver from 2027; target 78.96 yuan (56x 2027 PE), 'Buy' ^[13].
• **[NEW] Kaiyuan Securities on Aiko Solar (爱旭股份):** H1 net loss 0.754bn yuan, Q2 loss 0.314bn (narrowing 0.126bn q/q); H1 GPM 7.82%, Q2 8.4%; module revenue 7.276bn +15.7%; downgrades 2026-28 earnings on price pressure, 'Accumulate' ^[14].
• **[NEW] Founder Securities on Sirio Pharma (仙乐健康):** H1 net profit 0.15bn -5.1% YoY; Q2 0.10bn +11.4% YoY, +106.4% QoQ; China soft, Americas deep adjustment, Europe growth, other regions high-growth; 'Recommend' ^[15].
• **[NEW] Pacific Securities on Hengdian Magnetic (横店东磁):** H1 net profit 0.948bn -7.03% YoY, Q2 0.562bn +0.09%/+45.76%; magnetic + PV + lithium multi-driver; 2026-28 revenue 24.40/28.20/32.20bn yuan; net profit 1.94/2.35/2.72bn; EPS 1.19/1.44/1.67 yuan; 'Buy' ^[16].
• **[NEW] Citi on Kingboard Laminates (建滔积层板, 01888.HK):** expects another CCL price hike in early Oct; Sep glass cloth avg selling price +15-20% (1080 15.8 yuan/m, 2116 15.2 yuan/m, 7628 11.6 yuan/m); absolute rise 1.55-2.5 yuan/m - new YTD high, above Aug's 1.5-2 yuan/m and well above H1 2026 monthly 0.5-1 yuan/m; target HK$95, 'Buy' ^[17].

3. AI Capex Chain: Optical, Memory, Software, Models

• **[NEW] Deutsche Bank:** AI-era optical networking hardware will be biggest winners; Coherent and Lumentum 'highest conviction' picks ^[18]. (Single source - flag.)
• **[NEW] CITIC CLSA (Aug 28, Nanya / TechInsights):** Nvidia's view - storage, not compute, is the AI bottleneck; $280B supply/capacity commitments +134% q/q; Samsung, SK Hynix, Micron positive ^[19]. Nanya: 1% global DRAM share, DDR4 66% of revenue, DDR5 ~10%; DDR4 shortage seen through 2028 ^[19].
• **[NEW] Huatai Securities:** AI shifts software competition from 'standard product' to 'product + industry know-how + deep delivery' (FDE model); closed-source narrative fade pulls US software into valuation repair; China AI applications see H1 revenue and profit acceleration ^[20].
• **[NEW] CITIC CLSA (Aug 27, Ox Alpha):** Z.AI's GLM5.3 Flash ('Ox Alpha'/'牛来') - native multimodal at 320B / 18B active params; 1/10 cost of GLM5.2, 1/4 cost of Opus 4.8 on coding; priced 0.8/2.8 yuan per million input/output tokens, ~40% cheaper than DeepSeek V4 Flash ^[21]. ARR milestones: Z.AI $1B in July, MiniMax $0.8B and MoonShot $1B in early Aug ^[21].
• **[NEW] CITIC CLSA (Aug 28, Singtel strategic day):** NCS repositioning from system integrator to AI orchestrator; Sunshine Core AI orchestration platform +30% YoY; Digital InfrCo (DIC) shifting from GPUaaS to MaaS, layering open-weight models on raw compute ^[22].

4. Tail Risks & Source Quality

• Three-way tension: UBS Meng Lei calls A-share rebound on earnings + inflows + tech narrative ^[3]; HSBC Neumann warns Asia is structurally exposed to US AI hardware demand fragility ^[1]; Song Xuetao argues the Fed cannot fix the long end because AI capex is structural and won't stop ^[5]. Bullish-AI + demand-fragile + Fed-powerless can coexist only if AI capex holds - and the +25.9% Q2 A-share net profit print is the test ^[2].
• Source quality control: Deutsche Bank's Coherent / Lumentum top-pick is single-sourced ^[18]; Ox Alpha / Z.AI ARR data is single-sourced CLSA relay ^[21]; the Nanya / DDR4-to-2028 call is single-sourced CLSA ^[19]; Kingboard's October CCL hike is single-sourced Citi relay ^[17]; the HSBC 1997 parallel relies on 1993-1997 UST data without counter-source ^[1].
• Contradictions to flag: Citi's Kingboard buy is commodity-cyclical bullish on glass cloth ^[17] while Pacific on Hengdian Magnetic sees PV 'bottoming' but H1 net profit still -7.03% ^[16] - both PV-adjacent, different read on supply tightness. Aerospace Electric's Q2 +88.98% YoY is the standout single-name print ^[13]; if Q3 H1 reports sustain that pace, the H1 +16.8% headline ^[2] likely re-rates higher.