〔Day Digest〕AI Narrative Flips Back to 'Physical Bottleneck' as UBS Doubles Down on 2026 Hikes and Hormuz Refined Flow Falls to 35% - Memory Squeeze Extends to 2027

0. Weekly Arc

The AI debate pivots from demand-side skepticism back to supply-side capacity. Morgan Stanley's Sept 7 report reframes the launch of GPT-6 Astra as a capability-breadth jump across reasoning, engineering, computer use, and physical-world tasks, shifting the question from "how much infrastructure for known demand" to "how many new workloads become economically viable" ^[1]. UBS doubles down on a 2026 Fed hike path ^[2] and extends its memory-shortage view into 2027 ^[3]. Geopolitics persists: Goldman notes Hormuz refined-product flow at just 35% and crude at 70% ^[4]. Yet Barclays lifts S&P 500 year-end 2026 target to 7,950 from 7,800 ^[5], and CITIC sees the shipbuilding industry in a decade-dimension up-cycle ^[6]. Net: an AI supply-side narrative fighting a still-hawkish Fed and a still-elevated oil curve.

1. Policy and the Fed Path

• **[ESCALATED] UBS:** doubles down on its 2026 rate-hike forecast ^[2].
• **[NEW] Danske Bank:** still expects the Fed to hike once in December and once in March ^[4].
• **[NEW] TD Securities:** flags market mispricing risk into the blackout ^[7].
• **[ESCALATED] CICC Wealth Futures:** the 10Y UST yield at 4.805% is the highest close since October 2023, driven by the oil-price shock; gold remains under pressure and patience is needed for the shock to fade and the U.S. CPI to land ^[8].
• **[NEW] JPMorgan:** the 10Y UST auction may require price concessions ^[4].

**What decides next:** the U.S. August CPI print - CMB International estimates 0.4% m/m headline and 0.2% m/m core, with y/y at 3.4% and 2.4% ^[9] - against UBS's hike conviction ^[2] and Danske's December/March path ^[4]. Secondary test: the 10Y UST auction outcome against JPMorgan's price-concession warning ^[4]. Source quality control: the UBS hike-doubling and TD blackout-risk notes are both single-source items worth flagging as thin ^[2][7].

2. AI Narrative: From Demand Debate to Physical Bottleneck

• **[NEW] Morgan Stanley (Sept 7 report):** reframes the launch of GPT-6 Astra as a shift of the AI question from "how much infrastructure for known demand" to "how many new workloads become economically viable" ^[1].
• **[ESCALATED] UBS:** the chip boom continues; July global chip sales fell 9.8% m/m, with memory down 16.3% and IC ex-memory up 0.9%; UBS now expects DRAM and NAND supply tightness to extend into 2027 ^[3].
• **[NEW] Citi (Atif Malik team):** NPO has replaced CPO as the near-term preferred trade thesis; lasers and fiber are being positioned as "the next HBM" for scarcity premium. Lightmatter management expects NPO to enter the market in 2027 and reach mass deployment in 2028; the CPO scale-out window remains 2028-2029 ^[10].
• **[ESCALATED] CICC (Miao Yanliang):** AI has not yet produced large-scale stock employment substitution; impact is concentrated in high-exposure industries at entry-level roles, with hiring demand visibly down in some sectors ^[11]. Source note: this is a single institutional view, not a labor survey.

3. Energy, Geopolitics, and Metals

• **[ESCALATED] Goldman:** Hormuz refined-product flow at just 35%, crude at 70% ^[4].
• **[NEW] UBS:** raises Marathon Oil's target to $450 from $321, maintains "Buy" ^[12].
• **[NEW] Morgan Stanley:** geopolitical risk is making oil traders increasingly cautious on long positions ^[4].
• **[NEW] Datong Securities:** Hormuz tension lifted crude more than 9% on the week; gold spiked then pulled back; the 10Y CGB yield fell 1.45bp to ~1.68% ^[13].
• **[ONGOING] Gold long-term view:** Goldman and JPM maintain constructive long-term gold views on central bank purchases, reserve diversification, and currency debasement; the September Fed and inflation data set the near-term breakout ^[14].

4. Equity Targets and Single-Stock Calls

• **[NEW] Barclays:** raises S&P 500 year-end 2026 target to 7,950 from 7,800 ^[5].
• **[NEW] Goldman Sachs:** Arista Networks "Buy" with a $225 12-month target on 36x NTM+1Y P/E, implying 16.1% upside from the Sept 4 close of $193.78; thesis rests on improved supply-chain visibility and demand from AI infrastructure and enterprise campus deployment ^[15].
• **[NEW] Scotiabank:** cuts Oracle target to $215 from $241 ^[16].
• **[NEW] HSBC:** U.S. equities are not expensive; valuation does not fully reflect AI potential ^[4].

5. China and Asia Strategy

• **[ESCALATED] CITIC Securities:** shipbuilding is in a decade-dimension upward cycle; the August new-ship price index came in at 186.34, +0.04% y/y and +0.46% m/m, the fifth consecutive monthly gain; bulk-carrier orders may be the next leg ^[6].
• **[NEW] Huatai Securities:** H1 2026 listed-bank revenue and net profit grew 7.4% and 3.0% y/y; state-owned big banks show "volume stable, price up, quality good, increased returns" ^[17].
• **[NEW] CMB International (Li Haoyang):** sees policy at a crossroads; H1 2026 exports rose 17.3% y/y; AI/semiconductor normalization and a higher base will weigh, but industrial machinery, autos, chemicals, and power equipment together contributed 8.44 percentage points ^[9][18].
• **[NEW] CITIC Securities (HK) on K-beauty:** export growth accelerated from ~20%+ in 2025 to 32% y/y in 2Q 2026 and 52% y/y in July-August; the boom continues ^[19].
• **[NEW] CITIC Securities (HK) on India tech:** SaaS being "AI-fied" rather than facing a "SaaSpocalypse" - most SaaS firms raised guidance with revenue per employee sharply up, while IT services cut guidance; SI workforce faces automation pressure ^[20].
• **[NEW] Morgan Stanley (James Lord):** yen appreciation is not enough to overturn EM carry trades; the trade's resilience depends more on global currency volatility, global growth, equities, and EM bottom-up dynamics than on JPY alone ^[21].