〔Day Digest〕AI Selloff Teeters Between 1998-Style Deleveraging and 2000-Style Bust: $27 Billion Tech Net Selling and 60% 2027 Capex Growth—Policy and Application Earnings Set the Test

0. Daily Research Arc and Source Quality ^{[1][2][3][4][5][6][7][8]}

• **[NEW] Source-quality flag:** The packet is research-heavy and several passages are truncated; ^[4] and ^[6] provide only headlines, while ^[7] is a single-column opinion piece, so those items carry the thinnest corroboration ^{[9][10][11][12][4][6][7]}.
• **[NEW] Central tension:** The AI debate is whether forced deleveraging and position adjustment have triggered the selloff or whether AI business models are being invalidated, even as the investment narrative rotates from infrastructure buildout toward applications with direct earnings leverage ^[9][10].
• **[NEW] China counterweight:** Foreign banks issued bullish views on Chinese AI equities, Tianhong maintained a conditional rebound call, and Morgan Stanley’s oversold indicator crossed its technical threshold, improving the tactical setup without removing the need for confirmation ^[2][3][8].
• **[NEW] Asset-allocation split:** Guotai Haitong proposes a balanced November mix with a 45.00% equity allocation—7.50% A-shares and 20.00% U.S. stocks—while overweighting Chinese government bonds, holding A-shares and commodities at neutral, and underweighting U.S. bonds ^[5].

1. AI Selloff: Liquidity Reset or Fundamental Bust? ^{[9][10][11][13][14]}

• **[ESCALATED] Goldman Sachs — Rich Privorotsky, FICC and equities head:** The Nasdaq fell 5.5%, the Philadelphia Semiconductor Index fell 12% and the S&P 500 fell 4.4% in one session ^[9].
• **[ESCALATED] Selling and volatility:** U.S. cash equities recorded $3 billion of net selling, including $2.2 billion from information technology and communication services, while cumulative technology net selling widened to $27 billion, above July’s $21 billion technical-liquidation episode ^[9].
• **[ESCALATED] Volatility signal:** The VVIX registered its largest one-day move since February 5, 2018, and Privorotsky said the episode looks more like a 1998 liquidity crisis than the start of a 2000-style technology-bubble collapse ^[9].
• **[ONGOING] Goldman’s working interpretation:** The AI growth engine may remain intact while financing availability and buildout timing—not underlying demand—are being repriced ^[9].
• **[ESCALATED] Morgan Stanley’s diagnosis:** The selloff reflects hedge-fund position adjustments and momentum unwinds spilling into broad indexes rather than a deterioration in economic fundamentals ^[10].
• **[ESCALATED] Technical damage:** The Nasdaq Composite was more than 10% below its October high and entered a technical correction, while the Philadelphia Semiconductor Index fell more than 12% in one session ^[10].
• **[ONGOING] Structural rotation:** Morgan Stanley’s strategy calls for reducing high-beta, order-driven supplier exposure and rebuilding beta through companies with earnings growth after the capitulation ends ^[10].
• **[ESCALATED] Capex deceleration:** AI capital-spending growth is projected to slow from about 100% in 2026 to about 60% in 2027 and 14% in 2028 ^[10].
• **[NEW] Jefferies — Christopher Wood:** Widening credit spreads and continued semiconductor weakness may be signaling that the AI capital-spending cycle is peaking ^[11].
• **[NEW] Macro exposure:** Over the four quarters through 2026 Q3, AI-related real capital spending contributed 1.39 percentage points to 2.1% real GDP growth, or 67%, while information-processing equipment and software investment reached 5.35% of nominal GDP, above the 2000 dot-com peak of 4.46% ^[11].
• **[NEW] Credit stress:** Amazon, Alphabet and Meta 10-year corporate-bond spreads widened by 16 basis points, 16 basis points and 35 basis points from recent lows, while Oracle and CoreWeave five-year CDS spreads rose to 329 basis points and 1,163 basis points from early-June lows ^[11].
• **[ONGOING] Strategy-research counterpoint:** Short-cycle measures do not look extreme because earnings have absorbed much of the valuation expansion, but the share of funds in stocks is above the 2000 dot-com peak, the Buffett indicator has entered a high-risk zone, and Nasdaq’s Shiller CAPE is overvalued rather than extreme ^[14].
• **[ONGOING] Industry split:** Upstream hardware is still converting capital spending into profit with revenue growth outpacing capex, while the gap between chip inventory growth and North American cloud revenue growth is widening, raising inventory and capacity risk for the cloud and model layer ^[14].
• **[NEW] Anthropic friction:** Anthropic wants to reduce both business taken from and commissions paid to cloud providers while also designing, developing and deploying systems with systems integrators, SaaS firms and IT-service providers, encroaching on revenue that has belonged to those partners ^[13].
• **[NEW] Ode uncertainty:** Anthropic’s minority seed investment in consulting firm Ode keeps Ode off Anthropic’s balance sheet, but Claude-directed work makes Ode’s profit motive unclear, and SaaS and IT-services sentiment may remain weak until that balance stabilizes ^[13].

2. Data Centers: Bigger Pipeline, Harder Policy ^[11][12][4]

• **[ESCALATED] Goldman Sachs forecast:** The 2026 U.S. data-center construction-spending estimate was raised to $80 billion from $67 billion, the 2027 estimate to $100 billion from $87 billion, and a 2028 forecast of $125 billion was introduced ^[12].
• **[ONGOING] Construction base:** Goldman Sachs cites 86GW of projects under construction and about $900 billion of high-confidence projects awaiting construction, with spending growth already at 90% year over year before the election cycle ^[12].
• **[ESCALATED] Texas risk:** Texas governor-elect Gina Hinojosa called for an immediate data-center pause, while Goldman Sachs separately identified Texas as a key policy-risk region ^[12][4].
• **[ESCALATED] Market reaction:** The Nasdaq fell 6% over the week, the AI momentum index posted its fifth-largest decline on record, and the data-center credit-default-swap premium set a record ^[12].
• **[ONGOING] Goldman’s industry view:** Adam Bubes and Michael Ng judge that policy uncertainty may delay some project activity but will not reduce the overall market size ^[12].
• **[ONGOING] Political overhang:** Democrats’ post-midterm control of Congress has raised concern about tighter regulation of data centers and AI models ^[11].
• **[NEW] Source-quality note:** The standalone ^[4] alert supplies only the Texas focus and no figures, so the numerical construction case rests on the fuller ^[12].

3. China Equities: Conditional Rebound Setup ^{[1][2][3][15][16][8]}

• **[NEW] Foreign-bank positioning:** Goldman Sachs’ trading desk favors Chinese domestic AI stocks after the valuation correction and continued policy support, while JPMorgan and UBS say risk-reward has improved across more AI subsectors ^[2].
• **[NEW] ETF flows:** Equity ETF net inflows over the past three weeks reached a stage high, led by the CSI 300 and STAR 50 ETFs, while regulators announced additional liquidity, buyback and ETF-deployment support ^[2].
• **[NEW] Corporate buyback:** ChangXin Technology’s chairman proposed raising the buyback ceiling to CNY5 billion, CNY3 billion above the prior proposal ^[2].
• **[ONGOING] Tianhong Fund — Huang Zihan:** The ownership overhang has eased enough to sustain a conditional November oversold-rebound call for A-shares ^[3].
• **[ONGOING] Supporting indicators:** Active equity funds reduced positions more than expected, the margin-financing balance fell more than 20% from its peak, redemptions from fixed-income-plus and pure-bond products were contained, ownership peaks in many growth stocks moved lower, and broad-based ETFs continued buying ^[3].
• **[ONGOING] Stress-test read:** Overseas-compute-chain stocks absorbed the day’s declines, which Huang treated as a useful stress test that strengthened confidence in the rebound thesis ^[3].
• **[NEW] Morgan Stanley — Laura Wang team:** The MSASI rose 18 percentage points to 31% through November 4 from October 28, crossing the 22% oversold threshold for the first time in more than a month ^[8].
• **[ONGOING] Historical signal:** Readings below 22% have historically preceded rebounds within one to two months, and the latest sub-threshold stretch lasted about one month, prompting a preference for high-quality oversold A-shares over Hong Kong ^[8].
• **[ONGOING] 2027 framework:** The longer-dated A-share strategy remains neutral-to-optimistic but says the market is bottoming, a right-side signal is still required, and a mid-2027 long-bull continuation phase would be driven by earnings rather than distant narratives or valuation expansion ^[16].
• **[ONGOING] Macro assumptions:** The 2027 outlook expects exports to remain firm, investment and consumption to recover moderately, CPI to stay stable and PPI to decline ^[16].
• **[NEW] State-owned enterprise reform:** A central-enterprise study frames a shift from factor-intensive growth toward technology-driven growth and proposes market-based decisions, clearer commercialization channels, classified industry-fusion mechanisms and forward-looking credit ratings ^[15].
• **[NEW] Chemical-cycle thesis:** China Securities Co. argues that chemical profitability has risen from the bottom while valuations remain low because of oil prices, that synchronized domestic and overseas capacity cycles are ready, and that an inventory-cycle start could create supply tightness ^[1].
• **[NEW] Repricing channel:** Faster urbanization and industrialization in Asia, Africa and Latin America could create an industrial-goods deficit and help reprice China’s core chemical assets ^[1].

4. Sector and Company Dispersion: Demand Holds, Margins Divide ^{[17][18][19][20][21][22][23]}

• **[NEW] DRAM cycle:** CLSA says DDR5 16Gb contract prices rose 10.6 times from 2024 Q3 to 2026 Q3, while consensus expects Samsung Electronics and SK Hynix to post 2026 DRAM operating margins above 80% ^[18].
• **[ONGOING] Manufacturing bottleneck:** Applied Materials, Lam Research and KLA-Tencor lead critical manufacturing stages, while clean-room delivery times, HBM wafer intensity and EUV constraints delay the supply response ^[18].
• **[NEW] Malaysian banks:** September domestic loans grew 6.1% year over year, led by 7.7% corporate-loan growth against 5.0% household-loan growth, while net interest margins were broadly stable quarter over quarter and the CASA ratio reached 32.5% ^[17].
• **[NEW] Bank guidance:** Public Bank targets 4%-5% loan growth against 5.5% in 2026 Q2, while Hong Leong Bank guides to 6%-7% for fiscal 2027 against 7.7% for fiscal 2026 ^[17].
• **[NEW] UOB:** Third-quarter net profit was S$1.25 billion, down 15% quarter over quarter and up 182% year over year but 5% below consensus, while net interest margin contracted 6 basis points to 1.68% versus 1.71% expected ^[21].
• **[NEW] UOB balance sheet:** Loans grew 3.0%, deposits grew 2.2%, the loan-to-deposit ratio rose to 82.5% from 81.7%, CASA fell to 54.6% from 55.6%, and the effective tax rate, CET1 ratio and nonperforming-loan coverage all moved adversely ^[21].
• **[NEW] Chinese automakers:** October sales at major automakers rose 5% month over month and 4% year over year, with domestic sales at BYD, Geely, Chery and Great Wall up 12% month over month but combined exports down 2% ^[20].
• **[NEW] Export details:** BYD exported 170,200 vehicles, down 6% month over month, with exports falling from 40% to 35% of passenger-car sales; Geely exports were flat at 107,200 vehicles with EVs at 77% of sales, while Great Wall exports rose 5% to 62,900 vehicles and reached 52% of deliveries ^[20].
• **[ONGOING] Export read:** CLSA views the export pause as a plateau rather than a reversal and identifies European Union tariffs as a fourth-quarter catalyst ^[20].
• **[NEW] Platform demand:** Meesho added 4.3 million weekly active users week over week, ahead of Amazon’s 1.2 million and Flipkart’s 500,000, while Blinkit and Zepto each added 1 million and Domino’s lost 500,000 ^[19].
• **[NEW] Varun Beverages:** Consolidated revenue and profit after tax beat consensus by 9% and 5%, while Indian volume growth reached 22.5% against 17% expected and international volume growth was 11% above consensus ^[22].
• **[NEW] Varun Beverages margins:** Gross margin fell 334 basis points year over year and landed 305 basis points below consensus because packaging-cost inflation exceeded the benefit of low-cost inventory and efficiencies, while volume-driven operating leverage limited the EBITDA shortfall ^[22].
• **[NEW] Cimory:** Management expects to exceed its 15%-20% sales guidance after nine-month sales rose 26%, with third-quarter dairy sales up 45% and consumer-food sales up 15% ^[23].
• **[NEW] Cimory margins:** Gross margin fell to 42.1% from 43.5% in the second quarter amid higher milk and packaging costs, while management plans dairy price increases and delayed a new consumer-food factory to April 2027 ^[23].

5. Falsification Map ^{[9][10][12][3][18][16][8][20][21][22][23]}

• **[ONGOING] AI tape:** The 1998-style thesis breaks if evidence points to end-demand destruction rather than forced financing, while the application rotation needs earnings growth to offset the projected capex slowdown ^[9][10].
• **[ONGOING] Data centers:** The construction thesis weakens if Texas-style pauses spread beyond delayed projects or if the 86GW under-construction base and roughly $900 billion planned pool fail to convert into actual spending ^[12][4].
• **[ONGOING] A-shares:** The rebound case weakens if ETF buying fades or the ownership overhang reappears, and confirmation requires the MSASI signal and improved trading conditions to persist ^[3][8].
• **[ONGOING] DRAM:** The pricing thesis weakens if clean-room, HBM and EUV constraints release supply faster than demand, while expected operating margins above 80% provide the near-term earnings check ^[18].
• **[ONGOING] Banks:** Bank earnings fail if net-interest-margin compression overwhelms loan growth, with UOB’s 1.68% margin versus 1.71% expected providing the current warning ^[17][21].
• **[ONGOING] Autos and consumer:** The export-plateau view fails if exports resume falling, while consumer resilience fails if milk and packaging inflation outpaces planned price increases ^[20][22][23].