大型科技股10月抛售：AI侵蚀传统业务与负自由现金流担忧，Nomura认为存储板块反应过度

彭博社：【纳斯达克100指数进入回调，A 彭博社：【纳斯达克100指数进入回调，AI担忧加剧-彭博社】 纳斯达克100指数周一跌入回调区间，因投资者对大型科技股及人工智能交易日益悲观。该科技股为主的指数下跌1.4%，收于27,989.82点，较10月6日峰值下跌逾10%，达到技术性回调门槛。尽管该指数2026年仍上涨11%，但这是年内第三次回调，交易波动剧烈，投资者对AI的看法似乎每日都在变化。 当前抛售源于华尔街对AI建设成本及数千亿美元支出何时显现回报的担忧。Alphabet Inc.上周公布广告和搜索收入不及预期后，股价遭遇史上最差单日表现。Meta Platforms Inc.和Amazon.com Inc.在近期业绩公布后也大幅下挫。 “人们对这些公司不断投入、投入、再投入感到不安，”James Investment Research投资组合经理Brian Culpepper表示。“我不认为AI交易已死，但它确实超前了。今年市场相当强劲，但这导致许多股票过度延伸，尤其是AI涉及众多不同股票。” 本财报季除“七巨头”外，还出现多起高调爆雷事件，包括芯片制造商Intel Corp.和软件公司ServiceNow Inc.，两者均因令人失望的预测而遭遇历史性抛售。 https://www.bloomberg.com/news/articles/2026-11-02/nasdaq-100-enters-correction-as-megacap-results-amplify-ai-fears AI 解读 一句话结论 性质：中期逻辑的"验证性修正"，而非单纯情绪扰动。 触发点是Q3财报季现金牛业务（Alphabet搜索/广告）失速叠加巨额资本开支回报期被拉长，市场定价逻辑从"capex=利好"切换为"capex需ROIC验证"，因此短期交易层面偏利空纳指100、七巨头与AI基础设施链，且回调大概率未走完（年内第三次、速度快于前两次）。中期看，AI产业趋势（算力需求、token调用量、云订单）并未被证伪，受损的是"重资本+收入减速"组合，而非整个板块。受影响资产分层明显：利空Alphabet/Meta/Amazon/软件股（ServiceNow、Workday、Atlassian）与高beta半导体；相对受益公用事业、必需消费、医疗、长债及铜等"泛AI实物资产"。 我的倾向：短期谨慎偏空、不抄底指数；中期结构分化——轻资本强现金流者优于烧钱者。 事件背景与上下文 11月2日纳指100跌1.4%至27,989.82点，较10月6日峰值跌逾10%，为 2026年第三次技术性回调 （3月：伊朗战争+关税+AI回报质疑；7月：38个交易日即跌10%，SOX较高点跌25%入技术性熊市，叠加美联储加息预期）。本轮导火索是财报：Alphabet搜索与广告收入不及预期，创其史上最差单日表现；Meta、Amazon业绩后下挫；Intel（单日跌29%，四十余年最大）、ServiceNow因指引爆雷。关键变化在于 被质疑的不是烧钱部门，而是为烧钱提供弹药的现金牛部门 。研报背景：国金证券认为当前AI所处位置类似1997年互联网节点（上游盈利已兑现、中游capex扩张、下游应用未明）；广发证券指出巨头云与广告现金流是capex的财务基石，基石松动即引发估值修正；东吴证券测算纳指100 PE-TTM仍在31—34倍、历史分位70%—77%。 影响路径 传导链： 广告/搜索/电商收入减速 → 经营性现金流覆盖capex能力下降 → 巨头由"现金流机器"转为净借款者（发债、信用利差敏感）→ 市场重设AI收入兑现时点 → 长久期成长资产估值压缩 → 指数回调+资金向防御与实物资产轮动。 一阶影响： Alphabet、Meta、Amazon（现金牛失速+支出未减）；微软（云超预期但capex同样被审视）；企业软件（ServiceNow、Workday、Atlassian，同时承受"AI颠覆"与"客户预算被AI基建挤占"双重压力）。 二阶影响： 上游算力与设备——英伟达、博通、美光/SK海力士、KLA、服务器ODM、光模块/PCB/电源/散热、卡特彼勒等数据中心建设链。其订单不再由"叙事"线性外推，而取决于巨头capex增速是否边际放缓；存储与电力环节景气仍强，但预期修正幅度更大。软件被颠覆叙事反向利好AI原生应用与数据底座。 市场预期差： 市场可能 高估 了"AI需求崩塌"——OpenRouter周token调用量仍创新高（75.3万亿tokens）、头部云收入超预期、四巨头2026年capex指引约6,500—7,250亿美元仍在上调；同时可能 低估 了货币化延迟的时长，以及利率端（加息预期、油价与地缘推升实际利率）对高久期资产的持续压制。 核心分歧 看空方： capex/OCF比重显著抬升、自由现金流承诺"更遥远"、杠杆ETF与拥挤持仓待清算、一年三次10%回调（上次为2022年）、估值仍在历史70%+分位、抄底资金已连续受损。 看多方： 七巨头2026年利润预计+19%（标普其余+16%），预期PE已从28倍降至约21倍、低于十年均值；东北证券复盘显示历次回调后均较快创新高；产业需求数据未恶化。 最关键分歧变量： 巨头下一季capex指引是否首次下调，以及AI收入（云/广告/订阅）能否在2027年前给出可验证的规模化变现。 我的判断： 站在"修正未完、趋势未死"一侧。理由是本轮下跌由 收入端 而非纯支出端触发，属于基本面证据链的补全；叠加利率与地缘的风险溢价抬升，指数层面难有V型反转，但产业内部会走向"分化定价"，这与3月、7月的普跌不同。 受影响资产分层 层级 方向 标的/行业/商品 影响逻辑 确定性 时效 直接受损 利空 Alphabet、Meta、Amazon、ServiceNow、Workday、Atlassian、Intel 现金牛减速+指引爆雷+capex回报存疑 高 1—2周 直接受损 利空 纳指100/QQQ、高beta科技、3倍杠杆ETF（TQQQ/SOXL） 拥挤持仓去化、波动放大 高 日内—2周 间接受损 利空 博通、美光/SK海力士、KLA、光模块/PCB、超微电脑等ODM 订单增速依赖巨头capex边际变化 中 1—2周 间接受益 利多 公用事业、必需消费、医疗、电力设备（非AI主题） 防御轮动、低波动溢价 中 1—3个月 间接受益 利多 长端美债（TLT）、投资级债、黄金、铜/铝 避险+AI实物消耗双重支撑 中 1—3个月 情绪映射 中性偏多 苹果等轻资本AI合作者、AI原生应用与数据底座 不受自身capex拖累，颠覆叙事受益方 中 1—2周 商品/汇率/利率 分化 油价（伊朗）、日元（套息平仓）、10Y实际利率 通胀与紧缩预期主导科技估值分母 中 1—3个月 投资结论与观察点 短期（1—2周）： 不建议以"回调即买点"的方式抄底指数；财报爆雷余波、杠杆ETF清算与期权平仓可能带来惯性下冲。相对交易优于单边：减AI基建高beta、增防御与轻资本现金流标的；波动率高位下控制仓位而非预测底部。 中期（3—12个月）： 若capex与现金流匹配度未恶化、云与token需求维持，本轮更可能是"高景气长周期中的震荡校准"，届时优质AI基建与应用层具备更高性价比；反之若出现capex指引下调+云收入减速共振，则需下调板块评级。 最值得跟踪的3个信号： ① 巨头下一季capex指引、自由现金流与发债规模/信用利差；② SOX指数、HBM与DRAM合约价、美光/海力士/KLA订单（硬件需求真实读数）；③ AI变现证据——云收入增速、企业客户采购预算、OpenRouter token调用量，以及10Y实际利率与美联储路径。 结论增强条件： capex指引维持且AI收入指标继续上行→"趋势未破、逢低布局"增强；capex首次下调或信用利差明显走阔→升级为基本面下行，利空扩散至算力与设备链。 分类: AI人工智能相关新闻或者创新药相关新闻 原因: 核心内容围绕AI投资成本、回报预期及科技巨头业绩表现展开，是典型的AI主题新闻。 重要程度: 重要 温馨提示： 请将我们的邮箱地址 "mail.lxaa.top"、"lxkj.fintechcreater.top" 添加到您的邮箱白名单，以确保您能正常接收我们的邮件。 此邮件由系统自动发送，请勿回复。 如需调整推送设置，请 登录网站 进行修改。 如果您不想再收到此类邮件，请 点击这里退订 。

纳斯达克100指数已正式进入技术性回调区间，且本轮下跌伴随AI巨额建设支出回报时点受到质疑，以及Alphabet、Meta、Amazon、Intel和ServiceNow等公司的业绩或预测失误，显示市场正在加强对科技公司AI投入与实际回报之间匹配度的审视。

彭博社：【亚洲科技股承压，油价因伊朗局势 彭博社：【亚洲科技股承压，油价因伊朗局势下跌-彭博社】 亚洲科技股周二预计下跌，因美国芯片股抛售加剧，掩盖了市场对伊朗战争外交进展希望导致的油价暴跌。日本和韩国股指期货走低，澳大利亚合约上涨。美国芯片股指标连续三日下跌12%，纳斯达克100指数较峰值下跌10%，进入技术性修正。标普500指数下跌0.5%。 中国央行表示将鼓励金融机构加大股票支持工具使用并优化条款，以稳定市场，中国股市受关注。苹果公司销售增长放缓，中国收入未达预期。Anthropic PBC披露季度收入增长十二倍，计划未来数年支出超8950亿美元。 布伦特原油暴跌逾8%至每桶104美元上方，因交易员评估结束冲突的进展迹象，美国中期选举前一天。美元指数走弱，长期国债表现优异。地缘政治缓和降低能源价格和通胀风险，但原油下跌未能重振科技股需求，因AI支出和估值担忧持续。 Roundhill Financial首席执行官Dave Mazza表示，油价下跌应利好科技股，但股市未能守住开盘涨幅，显示市场信心不足。伊朗总统佩泽什基安称最高领袖已授权其推进和平谈判，特朗普暗示谈判有进展。特朗普称“他们想达成协议，这有可能”，并称油价下跌因“看起来进展顺利”。 AI支出担忧成为焦点，大型科技公司重返美国债券市场融资。甲骨文公司出售100亿美元债券和等值股票，Meta Platforms考虑至少200亿美元多币种发行，微软周四出售250亿美元债券，Alphabet周一第二次进入欧洲债券市场。 经济数据方面，美国10月制造业活动回升，生产增长近五年最强。周二职位空缺数据预计9月略有上升，周四初请失业金人数保持低位，周五非农报告预计10月就业增长翻倍至6.5万人，失业率维持4.2%。 Miller Tabak的Matt Maley表示，本周市场将极其有趣，需解读政治和地缘政治问题，AI投资回报的疑虑可能比历史模式更能决定市场方向。 https://www.bloomberg.com/news/articles/2026-11-02/stock-market-today-dow-s-p-live-updates AI 解读 一句话结论 这是一条典型的"利好失效"新闻：布伦特原油单日暴跌逾8%至104美元上方、伊朗最高领袖授权总统推进和谈，地缘缓和本应提振风险资产，但亚洲科技股仍预计下跌、美国芯片股三日重挫12%、纳斯达克100较峰值回撤10%进入技术性修正——说明压制科技股的核心矛盾已从"油价→通胀→利率"切换为"AI资本开支→债务融资依赖→回报验证"。性质上属短期交易偏利空、中期叙事重构的事件：利空美国芯片/AI算力纯支出方与油气开采板块，利多长久期利率资产、中国资产（央行表态加大股票支持工具）以及电力、铜、液冷等"泛AI"实物链条。我的明确倾向：不接科技股"地缘利好"的飞刀，把油价暴跌视为能源多头与通胀对冲的减仓信号；AI产业趋势未死（Anthropic季度收入12倍增长、未来数年8950亿美元支出计划），但选股逻辑必须从"谁花钱多"转向"谁能赚钱"。 事件背景与上下文 自2026年2月底美以发动打击、伊朗封锁霍尔木兹海峡以来，油价在78—116美元区间反复震荡，市场长期被"油价↑→通胀预期↑→美联储紧缩→科技股估值压缩"的链条主导（东吴、中信建投研报口径）。10月底至11月初，Alphabet、亚马逊、英特尔财报暴露广告/云等核心业务放缓，费城半导体单日暴跌7.2%、英特尔创四十多年最大单日跌幅，沪深30010月创纪录连跌12周。本轮表态发生在中期选举前一天，属于政治窗口期的"停火预期交易"，是冲突后期的缓和定价阶段。而芯片股与纳指在油价暴跌当天依然领跌，意味着杀跌动力已来自AI板块内部：估值、盈利兑现与融资结构，而非外部宏观。 影响路径 一阶影响 ：油价暴跌→油气开采、油服企业收入与资本开支预期下修（叠加国金化工周报提示的"供给洪峰"逻辑：海峡重开+伊拉克要求增产+对伊制裁豁免）；美国半导体抛售→亚洲存储与半导体制造链（三星、SK海力士、Kioxia、东京电子）跟随承压；苹果中国收入不及预期→果链消费电子订单与ASP预期下修。 二阶影响 ：科技巨头集体重返债市（甲骨文100亿美元债券+等值股票、Meta拟至少200亿多币种、微软250亿、Alphabet二度进入欧洲债市），意味着AI资本开支的融资结构从"经营现金流内生"转为"外部债务+股权稀释"，信用利差与CDS（此前已创历史新高）成为新约束，抬高高估值算力标的的实际折现率。反向看，Anthropic收入12倍增长与8950亿美元支出计划确认了上游实物需求——数据中心、电力/变压器、液冷、铜铝等"泛AI"链条的景气并未被破坏（国金证券：AI的宏观效应比AI本身更重要）。 三阶/宏观 ：油价下行→通胀尾部风险消退→美联储进一步加息概率下降→长债走强、美元走弱→非美与低估值资产相对受益；叠加中国央行明确"鼓励金融机构加大股票支持工具使用并优化条款"，A股政策底信号强化。 市场预期差 ：一是市场可能低估"利好不涨"的信号强度——油价跌8%+央行呵护仍挡不住科技抛售，说明筹码结构与叙事已受损；二是可能高估停火确定性，中期选举后政治约束解除，局势反复概率不低；三是可能误把"股票杀估值"读成"产业杀逻辑"，低估绝对支出量对实物链条的拉动。 核心分歧 多头认为：油价下跌+利率回落+Anthropic收入验证需求，科技股超跌反弹在即。空头认为：Meta已将资本开支指引上调至高于经营现金流，核心业务放缓使回报周期进一步推迟，叠加债务融资依赖，估值消化未完。最关键的分歧变量是"AI收入斜率 vs 资本开支斜率"的剪刀差，即下一轮财报（英伟达等）能否证明支出正在货币化。我更倾向偏空纯支出方的估值、偏多实物链条与中国资产——理由是市场"用盈利消化估值"的过程通常以季度为单位展开，而政治性停火随时可能反复，不足以扭转已坏的筹码结构。 受影响资产分层 层级 方向 标的/行业/商品 影响逻辑 确定性 时效 直接受损 利空 美国芯片/AI算力龙头（费城半导体、NVDA链）、油气开采与油服（中海油类、国际石油公司）、果链消费电子 芯片三日跌12%、纳指技术性修正；布伦特跌逾8%压缩上游利润；苹果中国收入不及预期 高 日内—1—3天 间接受益 利多 电力设备/变压器、液冷、铜铝等泛AI实物链；长久期国债；黄金 8950亿美元支出计划落地实物端；油价下行→通胀回落→久期资产与美元走弱受益 中—高 1—2周—中长期 情绪映射 分化 韩国存储（三星、SK海力士）、日本半导体设备（东京电子、Kioxia）、港股/恒生科技 跟随美股半导体beta；港股受央行股票支持工具+低估值支撑，相对抗跌 中 日内—1—2周 大宗/汇率/利率 明确 布伦特/WTI利空、美元指数利空、10—30年期美债利多、铜利多 地缘风险溢价回吐+通胀预期降温+美元走弱 高 日内—1周 投资结论与观察点 短期交易（1—3天） ：亚洲科技股大概率延续弱势，"利好不涨"后追多的风险回报比很差；可用油价反弹头寸对冲停火预期落空风险；A股/港股在央行表态下相对抗跌，关注股票支持工具条款优化细则落地。 中期（1—2周及财报季） ：定价锚转向英伟达财报、10月非农（预期翻倍至6.5万人、失业率4.2%）、中期选举结果与美伊谈判实质进展。中期产业判断不变：AI资本开支总量未收缩，但"支出方"股票与"受益实物方"股票将持续分化。 最值得跟踪的3个信号 ：①超大规模科技企业债/CDS利差与新发债成本——判断AI融资约束是否收紧；②布伦特能否站稳100美元下方及长端利率方向——判断"通胀—利率—估值"链条是否真正解除；③云/AI收入同比增速与资本开支指引的剪刀差——判断回报验证是否开始兑现。 结论增强条件 ：若油价继续下行、长端利率回落但纳指仍无法收复关键均线，则确认资金撤离成长风格，应加大防御+实物链条配置；若美伊谈判破裂、海峡再度受阻，地缘溢价回归，油气与军工回补，科技股估值将受二次压制。 分类: AI人工智能相关新闻或者创新药相关新闻 原因: 提及Anthropic收入增长、AI支出担忧及科技公司融资行为，直接关联AI领域发展与投资趋势。 重要程度: 重要 温馨提示： 请将我们的邮箱地址 "mail.lxaa.top"、"lxkj.fintechcreater.top" 添加到您的邮箱白名单，以确保您能正常接收我们的邮件。 此邮件由系统自动发送，请勿回复。 如需调整推送设置，请 登录网站 进行修改。 如果您不想再收到此类邮件，请 点击这里退订 。

暂无独立解读

芯片股拖累美股指，纳斯达克100指数陷入 芯片股拖累美股指，纳斯达克100指数陷入技术性回调区间，存储、光通信下挫，Marvell跌13%，Lumentum跌19%；油价大跌，黄金反弹。 伊朗最高领袖哈梅内伊授权政府推进和平谈判，特朗普称有可能达成协议，否认中期选举后将恢复对伊朗的打击。 王毅分别同伊朗最高国家安全委员会副秘书巴盖里、沙特国家安全顾问艾班会见会谈。 美国10月ISM制造业指数超预期扩张，创逾四年第二高读数，产出增长创近五年最强。 中国央行出手稳市场！将开展第三次证券、基金、保险公司互换便利操作，周二接受申报。央行等三部门优化资本市场支持工具，回购增持再贷款利率降至1%。 中国证监会：拟优化融资保证金比例等风控指标，适度释放融资空间，盘活存量资金。 苹果第四财季营收1144.8亿美元，高于预期。苹果预计第一财季营收将增长7%-10%。苹果盘后一度跌超4%。 Palantir三季度营收翻倍，美国商业收入首超政府收入，全年指引大幅上调。 Anthropic IPO前更新招股书凸显盈利压力：Q3营收近170亿美元、同比增11倍，净亏损扩大92%，算力承诺近9000亿。 时隔半年再筹200亿巨债，Meta拟首次发售欧元债券拓展AI资金池。 1150亿美元豪购Lumentum！Marvell打通“芯片+光学”，剑指AI互联超级帝国。Lumentum Q1营收同比飙升136%，斩获50亿美元订单承诺，Q2指引大幅超预期。 提醒：澳洲联储周二将公布利率决议，料加息25个基点。 AI 解读 芯片股拖累纳指100进入技术性回调：AI融资约束取代需求担忧成为主导变量 一句话结论 这是一次 由"AI资本开支融资能力"担忧触发的板块级去拥挤交易 ，而非需求证伪：费半三日跌12%、纳指100较峰值回撤10%进入技术性回调，Marvell、Lumentum同日双跌（-13%/-19%）说明市场惩罚的是"1150亿美元豪购"的稀释风险，而不是光模块订单本身（Lumentum Q1营收+136%、50亿美元订单承诺）。伊朗和谈进展使布伦特暴跌逾8%， 地缘缓和与油价下跌构成对美股估值的正面对冲，但未能提振科技股 ，这本身就是"信心不足"的信号。资产含义上：美股AI硬件（存储、光通信、网络芯片）短期偏空；应用层软件（Palantir）相对占优；A股/港股因央行互换便利、回购增持利率降至1%、融资保证金优化形成 独立政策底 ，是与美股AI链脱敏的相对收益方向。倾向： 短期回避高beta AI硬件，中期不认为资本开支周期结束，超跌后择优布局订单可验证环节 。 事件背景与上下文 AI交易自7月起反复被"融资约束"质疑：英特尔、博通CEO早在10月中旬即警告资金获取能力可能成为制约；此后甲骨文出售100亿美元债+等值股票、微软250亿美元债、Meta拟首次发行200亿欧元债、Alphabet二度进入欧洲债市，Anthropic Q3营收近170亿美元但亏损扩大92%、算力承诺近9000亿美元，黑石更预警其IPO或获"较高档垃圾级"评级—— 烧钱速度与信用质量剪刀差成为市场焦点 。叠加美联储官员接连放鹰（洛根称至少再加息两次）、9月CPI超预期、美元创周期新高，高估值成长股的双重压力（利率+融资）被同时打开。存储链另有独立利空：长鑫4F²架构进展引发中国DRAM替代威胁定价。中国央行此时推出第三次SFISF操作、回购增持再贷款利率降至1%、证监会拟优化融资保证金，是明确的对冲性政策信号。 影响路径 一阶（直接） ：AI巨头债务融资规模扩大 + 评级担忧 → 市场下调"资本开支可持续性"预期 → 高估值、高杠杆、高拥挤度的AI硬件股（Marvell、Lumentum、存储、光通信、服务器ODM）杀估值而非杀盈利 → 纳指100技术性回调、费半三日-12%。Marvell跌13%反映股权/债务稀释与整合风险；Lumentum跌19%（尽管业绩超预期）反映的是 板块贝塔与"利好出尽"式减仓 ，而非公司基本面恶化。 二阶（间接） ：若交易以现金+发债为主 → 光模块、DSP、CPO、光纤（康宁）、测试封装订单确定性提升 → 中长期利好上游互连与光电子；反之若融资成本继续跳升（Meta新债利率已现"跳涨"）→ 二线AI基建标的（CoreWeave类）信用风险外溢 → 拖累整个算力租赁与服务器链。油价-8% → 能源与通胀预期下行 → 美债收益率与美元承压 → 对长久期成长股估值是 正向但尚未被定价 的缓冲。 预期差 ：市场当前把"融资压力"错误定价为"需求崩塌"——Lumentum 50亿美元订单、美光FY27Q1指引615亿美元、AWS产能售罄至2028年，均指向需求仍强；而市场可能低估了 中国政策组合对港股/非银的边际弹性 ，以及 油价大跌对12月美联储路径的实质影响 。 核心分歧 看空方 ：AI支出靠债务而非现金流支撑，Anthropic亏损扩大92%、Meta/OpenAI融资目标调整，是2000年式"资本开支先行、回报滞后"的顶部信号；叠加中国存储替代，硬件毛利周期见顶。 看多方 ：本轮调整是权重集中（芯片占标普比重从8%升至更高）带来的指数级放大，订单与产能数据未变；油价下跌+ISM制造业超预期扩张，意味着"增长韧性+通胀缓和"的理想组合，回调即布局机会。 关键分歧变量 ：①本周大型科技公司新发债的 发行利率与超额认购倍数 ；②10月核心PCE与美联储12月加息概率；③长鑫4F²服务器DRAM量产节奏。 我的倾向 ：偏"流动性/估值冲击 > 基本面恶化"，但 不抄底高beta存储与光通信的贝塔 ——因为融资担忧的证伪需要1-2个季度，而拥挤度出清通常不会一步完成。相对收益在应用层与国内政策链。 受影响资产分层 层级 方向 标的/行业/商品 影响逻辑 确定性 时效 直接受损 利空 美股存储（美光、SK海力士、闪迪、西数、希捷）、光通信（Lumentum、Coherent、Credo、Ciena）、Marvell、服务器ODM 融资约束+估值拥挤+中国DRAM替代，杀估值为主 高 1-3天/1-2周 直接受益 利多 A股/港股券商、非银、高股息回购概念 SFISF第三次操作、回购增持利率降至1%、融资保证金优化释放杠杆空间 中高 1-2周 间接受益 利多 光模块上游（光纤、CPO、DSP、测试封装）、AI互连供应链 Marvell"芯片+光学"整合验证长期需求，订单承诺50亿美元 中 中长期 间接受损 利空 能源股、油价多头、通胀挂钩资产 布伦特-8%、霍尔木兹重开预期、哈梅内伊授权和谈 高 1-3天 情绪映射 分化 苹果（盘后-4%）、Palantir、Anthropic类AI应用 苹果中国收入未达预期=利好兑现；Palantir商业收入超政府=变现正例 中 1-3天 大宗/汇率/利率 混合 黄金反弹、美元走弱、长债走强、澳洲联储料加息25bp 避险+地缘缓和下黄金超跌反弹；美元走弱利于非美权益 中 日内/1周 投资结论与观察点 短期（1-2周） ：美股AI硬件以回避/减仓为主，技术性回调通常需要5-10个交易日消化，不宜抢反弹；A股/港股利用政策窗口做 非银+回购链 的相对收益，而非追高与美股同涨同跌的算力映射标的。油价大跌利好交运、化工中下游成本端。 中期（1-2季度） ：AI资本开支的"融资能力"取代"需求"成为核心定价变量，优选 订单可验证、现金流为正 的环节（设备龙头、软件应用层），回避依赖债务扩张的二线算力租赁商。 最值得跟踪的3个信号 ：①微软/Alphabet/Meta本周新发债的 票面利率与认购倍数 （利率再跳升=融资约束定价强化）；②周四初请+周五非农、10月核心PCE与12月加息概率、美元指数与10年期美债收益率；③DDR5/NAND合约价与长鑫4F²服务器DRAM进度，以及Lumentum交易对价结构（现金vs换股）。 结论增强条件 ：若伊朗协议落地使油价维持低位、核心PCE回落、且大型科技发债利率未再上行，则"估值冲击而非基本面恶化"判断增强，可转向超跌买入光通信与存储龙头；反之若Anthropic/甲骨文类信用利差走阔、订单出现下修，则需将本轮定性为 AI资本开支周期顶部 ，全面降低硬件敞口。 （以上为基于公开新闻与研报的分析推断，非投资建议；数据以公司公告与官方披露为准。） 分类: AI人工智能相关新闻或者创新药相关新闻 原因: Marvell收购Lumentum布局AI互联，Palantir AI商业收入超政府，Anthropic和Meta融资用于AI发展，高度关联AI产业进展。 重要程度: 重要 温馨提示： 请将我们的邮箱地址 "mail.lxaa.top"、"lxkj.fintechcreater.top" 添加到您的邮箱白名单，以确保您能正常接收我们的邮件。 此邮件由系统自动发送，请勿回复。 如需调整推送设置，请 登录网站 进行修改。 如果您不想再收到此类邮件，请 点击这里退订 。

原文将本轮下跌的核心新增信息界定为AI巨额资本开支在债务、股权及算力租赁环节面临的融资约束，而非AI需求崩塌；新的观察变量包括科技企业融资成本、AI应用层相对上游硬件的强弱、互换便利第三次操作及证监会融资风控指标优化。

彭博社：【AI投资对冲策略-彭博社】 彭博社：【AI投资对冲策略-彭博社】 投资专家表示，在人工智能相关股票近期下跌后，可通过多元化投资和对冲策略来应对市场波动。纳斯达克100指数本周早些时候进入回调，较10月峰值下跌10%。美国中期选举后，民主党控制参众两院，数据中心建设等AI相关股票也出现下跌。 三位投资专家建议，若手握100万美元，应遵循分散投资和长期思维原则。他们强调，尽管AI领域存在不确定性，但整体市场仍具韧性。 此外，彭博社独立分析显示，当前通胀压力主要来自能源和AI领域。伊朗战争推高油价，导致机票和汽油价格上涨；AI需求推升内存芯片价格，使笔记本电脑、电视和智能手机更贵。剔除这些异常因素后，今年截至10月通胀率为2.2%，接近美联储2%的目标，可能支持美联储今年暂缓调整利率。 彭博价格项目还发现，自去年以来，汽车保险价格下降5.5%，GLP-1药物下降36.1%，牛油果下降8.2%。住房和食品支出占美国家庭近半开销，同比分别上涨3.1%和2.9%。 https://www.bloomberg.com/news/newsletters/2026-11-05/how-investment-professionals-would-deploy-1-million-amid-ai-jitters AI 解读 【AI投资对冲策略-彭博社】 投资专家表示，在人工智能相关股票近期下跌后，可通过多元化投资和对冲策略来应对市场波动。纳斯达克100指数本周早些时候进入回调，较10月峰值下跌10%。美国中期选举后，民主党控制参众两院，数据中心建设等AI相关股票也出现下跌。 三位投资专家建议，若手握100万美元，应遵循分散投资和长期思维原则。他们强调，尽管AI领域存在不确定性，但整体市场仍具韧性。 此外，彭博社独立分析显示，当前通胀压力主要来自能源和AI领域。伊朗战争推高油价，导致机票和汽油价格上涨；AI需求推升内存芯片价格，使笔记本电脑、电视和智能手机更贵。剔除这些异常因素后，今年截至10月通胀率为2.2%，接近美联储2%的目标，可能支持美联储今年暂缓调整利率。 彭博价格项目还发现，自去年以来，汽车保险价格下降5.5%，GLP-1药物下降36.1%，牛油果下降8.2%。住房和食品支出占美国家庭近半开销，同比分别上涨3.1%和2.9%。 https://www.bloomberg.com/news/newsletters/2026-11-05/how-investment-professionals-would-deploy-1-million-amid-ai-jitters 分类: AI人工智能相关新闻或者创新药相关新闻 原因: 新闻核心聚焦AI投资策略、芯片价格上升及AI对通胀的影响，属于典型AI相关新闻。 重要程度: 重要 温馨提示： 请将我们的邮箱地址 "mail.lxaa.top"、"lxkj.fintechcreater.top" 添加到您的邮箱白名单，以确保您能正常接收我们的邮件。 此邮件由系统自动发送，请勿回复。 如需调整推送设置，请 登录网站 进行修改。 如果您不想再收到此类邮件，请 点击这里退订 。

纳斯达克100指数较10月峰值下跌10%并进入回调后，三位投资专家针对100万美元资金提出分散投资和长期配置建议。彭博社独立分析进一步指出，能源与AI相关因素正在推高通胀，而剔除这些异常因素后的通胀率接近美联储目标。

# Global Memory EQUITY: MEMORY ## Is data center unprofitable? No, it is very profitable OP surges first, FCF follows with a time lag, and big techs have options to reassure FCF-sensitive investors Big tech shares fell sharply last week due to two concerns: (1) that AI is eroding legacy businesses, and (2) FCF has turned negative (although this has been well anticipated) despite rapid profit growth and might get more negative. We think the first is in fact a reason why big tech companies cannot cut investment, which, in our view, is not bad for semiconductors. The second would be a problem for big tech CFOs, whether they could resolve the issue with a range of financing tools such as operating leases, joint ventures and securitization of future cash flows, in our view. It is not a variable that changes semiconductor demand or the value of semiconductor companies. In our opinion, the market's reaction to semiconductor stocks is overdone. ## Two reasons for the sell-off The October sell-off reflects two overlapping concerns, in our view: AI cannibalizing legacy business, and negative FCF. Alphabet (GOOGL US, Not rated) fell 12% on 29 October, which is its largest single-day drop since its public listing (link). \- AI cannibalizing legacy businesses: Alphabet's search advertising growth slowed to 12% in 3Q, a five-quarter low, which media attributed to AI Mode in Search (link). Meta's (META US, Not rated) AI agent Muse, launched on 8 September, handles shopping and bookings on users' behalf and carries no ads. Amazon's (AMZN US, Not rated) decision to block Muse shows that agents can bypass existing shopping and ad funnels. Each company's AI now threatens the others' legacy businesses. \- Negative FCF, borrowing and new share issuance: Four big techs and Oracle (ORCL US, Not rated) have raised USD326bn of bonds in 2026, about three times the amount of 2025, according to S&P Global Market Intelligence. In 3Q, Alphabet issued USD28.7bn of bonds and USD12.4bn of stocks through an at-the-market program. The first concern is why investment cannot stop. In our view, the only way to defend the legacy business is to build AI competitiveness, so slowing investment is hard. That is, if anything, good news for semiconductors. The second concern is understandable for short-term investors, but treating a timing gap as an AI profitability problem is an overreaction, in our view. ## Why big tech must invest early 1. Performance scales with inference compute, not just training. Reasoning models and agents use far more tokens per answer, and post-training reinforcement learning is compute-intensive. Without enough compute, even a good model cannot be sold in volume. 2. Revenue is capped by capacity, not demand. Backlogs stand at USD560.8bn for Google Cloud (end-September), about USD496bn for AWS (part of Amazon) (end-June) and USD677bn of commercial remaining performance obligations for Microsoft (MSFT US, Not rated) (end-September, +73%) (Stock Titan). Microsoft said 3Q capex fell short of its USD50bn plan because third-party capacity was delayed. 3. Customers are data. Customer usage feeds the next model, and once enterprises wire workflows, data and agents into one model, switching becomes hard. Customers might try several models at first but are likely to settle on one or two. ## Research Analysts Asia Pacific Technology CW Chung - NIHK cwchung@nomura.com +852 2252 6075 ## Korea Technology ## Eon Hwang - NFIK eon.hwang@nomura.com +822 3783 2318 YJ Kim, CFA - NFIK yj.kim1@nomura.com +82237832332 The next three years of compute build-out are therefore likely to set a market structure, in our view. For a company that slows investment, the cost shows up less in near-term FCF than in its later market position. Profit and cashflow diverge: common in growth industries, a classic J-curve Current investment is highly profitable and lifts future growth; we believe the negative FCF comes from the lag in payback. Cloud operating margins are rising even as capex surges: in 3Q AWS reached 38.7% (vs. 34.6% a year earlier) and Google Cloud 36.2% (vs. 23.7%) (link). Some media are misreading: reading negative FCF as 'not making money' confuses profit with cash. Big tech companies' operating income is surging. In our view, the negative FCF is simply a result of time gap from paying cash today for equipment that will generate future revenue. 3Q cloud revenue accelerated for all three providers listed below, and margins rose with it. The combined 3Q cloud revenue of the three providers was about USD110bn, up about $56\%$ y-y. Google's cloud revenue rose $103\%$ y-y in 3Q, and as surging investment converts into revenue, the others' revenue and operating income should surge as well. \- AWS: Revenue of USD46.0bn (+39%), its fastest growth since 2021, with a 38.7% operating margin. It is growing close to 40% again at nearly three times the size. \- Azure: +48% (Microsoft, FY1Q27). Microsoft said capex fell short of plan because of third-party capacity delays, showing growth is constrained by capacity rather than demand. \- Google Cloud: Revenue of USD30.7bn (+103%), operating margin 36.2% (23.7% a year earlier). It includes TPU system sales, so its growth is not directly comparable with the other two. Backlog is USD560.8bn. Fig. 1: Big tech's y-y cloud revenue growth trend ![](images/9b6d5b43d25292969de83a92de2332fd604abe579397cffb8762e8a99dbf57ae.jpg) Note: Azure data is based on "Azure and other cloud services" from the company's earnings data; Google Cloud revenue includes Workspace and TPU system sales. Source: Company data, Nomura research Fig. 2: Big techs' operating income keeps rising, while FCF traces a J-curve ![](images/847ee6efd38a2138a029a05ce361ae118da373817d174aa2cca49c9afeb877f2.jpg) Note: Based on combined earnings of four big techs (Alphabet, Meta, Amazon, Microsoft); 2025 figures are actual; 2026 figure is based on actual and company guidance; 2027 and onward is based on our estimates. Source: Company data, Nomura estimates \- Profit and cash timing: Capex leaves as cash in the year it is spent, while depreciation and revenue arrive 5-20 years after the assets go live. So while investment is accelerating, FCF can be negative even as operating income grows. \- Investment growth sets when FCF turns positive: If investment grows faster than cohorts pay back, negative FCF widens; once investment growth slows, installed capacity throws off cash and FCF rises quickly. If shares fall on negative FCF, we believe it offers a prime opportunity for long-term investors. \- Caution: The timing of the FCF trough and turn, and the slope afterwards, move with assumptions on investment growth, pricing and utilisation. Fig. 3: Earnings, depreciation, and operating cashflow For single cohort of investment ![](images/2821aa27a2ef60b316bda2f79f794dc680559b68114dadfa68464aacb88b4edd.jpg) Note: "Cohort" means the assets invested in a given year; above chart is for an illustrative purpose to describe cashflow trend following an investment of 100, and does not describe any specific company's investment or return. Source: Nomura estimates ![](images/01e9edd69370793cdf20b3b2943862e8dde6c0ce77be6e573447edca4cbd9b54.jpg) Note: Above chart is for an illustrative purpose to describe cashflows when investments for multiple cohorts happen across multiple periods, and does not desrcive any specific company's investment or return. Source: Nomura estimates ## Big tech's four options Big tech has four paths between investment and cash flow, in our view. We see a low probability of option 2 (a capex freeze, see the table below); it is more likely to keep the investment path of option 1 while adding the financing methods of options 3 or 4. Fig. 5: Big tech's four options between AI investments and cashflow Option (1) Accelerate investment (2) Freeze capex (3) (1) + JV / lease (4) (3) + forward sale of cash flows What it is Keep raising capex to meet demand Hold capex at 2026 level from 2027 A JV builds buildings/ power infrastructure; big tech leases them Sell the future cashflow of contracted enterprise customers now Near-term FCF More negative Improves sharply at once Less negative Can turn positive Long-term FCF/ revenue Highest Much lower Slightly below (1) (lease payments) Can turn positive Market position Secures lead At risk Same as in (1) Same as in (1) Balance sheet liabilities Rise for 2-3 years Little change Lower than (1) Depends on accounting Rating agencies are likely to interpret as... Higher debt Neutral Adds leases and guarantees to debt Depends on the financing structure Semiconductor demand Up Down Same as in (1) Same as in (1) Our view on probability Investment path maintained Low High(already underway) Medium-High Source: Nomura estimates Fig. 6: Big tech's annual FCF by scenario Scenario 2 and 4 lead in near-term; 1 and 3 lead later ![](images/e70b21e977fee061d3c6dbd014ae199b00215f179cabec261db1ad03220c1790.jpg) Note: Above chart is for illustrative purpose and does not reflect actual figures/estimates. Source: Nomura estimates During the heavy investment years of 2027-2030F, options 2 and 4 would generate the highest FCF. Once investment growth slows, options 1 and 3, which do not give up growth, would generate the most. \- Option 1, accelerate investment (current expectation): Near-term negative FCF widens, but FCF rises quickly once investment growth slows. FCF duration simply lengthens, so a share-price fall might be an opportunity for long-term investors. \- Option 2, freeze capex: FCF improves immediately, but future revenue becomes much smaller. If a compute shortage pushes prices up, the Jevons paradox (cheaper compute driving more usage) also weakens. Negative for semiconductors, but unlikely, in our view, for the reasons mentioned above. \- Option 3, JV and operating lease instead of finance lease: The approach Meta used at Hyperion (link 1, link 2; Meta 20%, Blue Owl (OWL US, Not rated)-managed funds 80%, USD27.3bn of SPV bonds rated A+ by S&P, four-year renewable operating leases with a residual value guarantee). Building investment drops out of cash capex, at the cost of lease payments over the following 20 years. Big tech companies improve cash flow, and institutional investors such as pension funds gain a stable, higher-yielding asset. \- Option 4, partial monetization and forward sale of operating cash flows: Big tech keeps the equipment, operations, customer relationships and mark-up, and sells only the guaranteed cash flows that enterprise customers must pay to long-term investors such as pension funds and insurers. Cash arrives when investment peaks, and the price is a discount cost paid later. Under both US GAAP and IFRS, it is likely to be recognized as a liability (link), but it raises cash without adding borrowings. Spinning off part of the assets and cash flows into a subsidiary and selling equity in it is another possibility. Oracle's plan is not based on a single scenario. It keeps the investment path of option 1 (no capex freeze) and combines option 3 (leases/SPV), a variant of option 4 (customer prepayments), and a structure in which customers or third parties buy the hardware. It is a real-world example of our call that big tech will solve the cash problem through financing methods rather than by cutting investment. Fig. 7: Oracle's AI investment and funding plan vs. our "four option"-based scenarios Item Details vs. four options Investment path - Capex peaks in FY27-28 (Jun 2026-May 2028), and steps down from FY29- FY27 gross capex of USD90-95bn (net cash capex no more than USD70bn) Option 1 /Option 2 for longer-term FCF turn - FCF positive in 2H FY28 (Dec 2027-May 2028)- FY29 FCF of USD30bn or more Option 1 (J-curve) Long-term leases - USD288bn of not-yet-commenced data center leases(15-19 years, mostly operating leases)- Developers build and Oracle leases Option 3 Project SPV - About USD18bn of bank loans + USD3bn of Blue Owl equity- Oracle reportedly bears the debt costs and power responsibility(Similar to Meta Hyperion, but with stronger guarantee-like support) Option 3 Customer prepayments - USD11.4bn received in 1Q FY27, with a significant financing component(future cash received upfront from customers, but not sold to investors) Option 4 variant Customer-provided HW - Part of the backlog is contracted with customers supplying their own GPUs- Customer buys the equipment(Same effect as our 20% retained / 80% sold idea) New type Vendor financing Reduces upfront cash needs A form of borrowing Equity and debt - USD20bn ATM equity program completed in 1Q FY27- New capital raises through FY30 capped at USD20bn(USD10bn equity + USD10bn gross debt, excluding refinancing) Equity funding Leverage target - Net debt (including operating/finance leases and vendor financing)- Adjusted EBITDA below 3.5x by FY29 Rating-agency view built in Source: Company data, Nomura estimates Big tech's financial capacity: borrowing rises, but it is not a burden, and many funding options exist Even if FCF stays negative for several years as assumed under option 1, leverage should not move far outside its historical range, while equity grows quickly. The big four tech companies' equity was USD2.0tn at end-September 2026, and the companies are cutting shareholder returns, and in some cases issuing stock (Alphabet's USD12.4bn ATM issuance in 3Q). \- (Point 1) historical range: Big four's “(financial debt + operating leases) / equity” ratio was about 31% at end-2025 and about 42% at end-2022. We estimate that the peak will likely be at 2022 levels even through option 1 assumes negative-FCF years and falls quickly once FCF turns positive. \- (Point 2) equity growth: Rising profits and lower shareholder returns grow equity fast enough to absorb the extra borrowing. If returns continue, the ratio rises further. \- (Point 3) funding cost: Big tech companies still issue bonds at high-grade credit ratings, and the Hyperion SPV bonds are rated A+. Memory risk if data centers are oversupplied? Unlikely for several years The biggest risk, in our view, is a compute glut that leaves revenue per GW below expectations. Building and equipment depreciation is a fixed cost, so lower pricing hits profit directly. \- Big Tech's response to partial data center oversupply: (1) long-term supply contracts cushion price declines (see the backlogs mentioned above); (2) spare compute is redirected to model training, accelerating performance gains; and (3) capex growth is fine-tuned modestly. Point (2) works to re-stimulate demand through lower token prices and better performance. Overall, the impact on semiconductor demand is limited, we think. \- Unexpected severe data-centre oversupply (very unlikely near term; possible in the long run): Data center customers appear to procure about 80% of their total memory needs under LTAs with purchase obligations, and the rest outside LTAs. We estimate memory makers have about 50-70% of total volume locked into LTAs with a guaranteed minimum margin (we estimate operating margin of 60% or more). If LTA customers buy only their obligated volumes and cut purchases outside of LTAs (10-15% of total memory volume), some of that supply would ease severe shortages at other non-LTA customers, but non-LTA prices would stop rising and in the worst case could fall below protected LTA prices. Memory makers would then adjust capex plans or cut utilisation to find a new supply-demand balance. Their response is more flexible than in the past, and for memory makers, short-term profit and the bonuses of the employees who make decisions now matter more than short-term market share. \- Signs of data-centre oversupply are still weak: 3Q cloud growth accelerated (AWS +39%, Azure +48%, Google Cloud +103%), and Microsoft's capex fell short of plan because of third-party capacity delays. Indicators to watch: slowing cloud growth, GPU rental prices and backlog growth. ## Appendix A-1 This report has been produced by Nomura International (Hong Kong) Ltd. (NIHK), Hong Kong. See Disclaimers for Nomura Group entity details. ## Analyst Certification We, CW Chung, Eon Hwang and YJ Kim, hereby certify (1) that the views expressed in this Research report accurately reflect our personal views about any or all of the subject securities or issuers referred to in this Research report, (2) no part of our compensation was, is or will be directly or indirectly related to the specific recommendations or views expressed in this Research report and (3) no part of our compensation is tied to any specific investment banking transactions performed by Nomura Securities International, Inc., Nomura International plc or any other Nomura Group company. ## Important Disclosures ## Online availability of research and conflict-of-interest disclosures Nomura Group research is available on www.nomuranow.com/research, Bloomberg, Capital IQ, Factset, LSEG. Important disclosures may be read at http://go.nomuranow.com/research/m/Disclosures or requested from Nomura Securities International, Inc. If you have any difficulties with the website, please email grpsupport@nomura.com for help. The analysts responsible for preparing this report have received compensation based upon various factors including the firm's total revenues, a portion of which is generated by Investment Banking activities. Unless otherwise noted, the non-US analysts listed at the front of this report are not registered/qualified as research analysts under FINRA rules, may not be associated persons of NSI, and may not be subject to FINRA Rule 2241 restrictions on communications with covered companies, public appearances, and trading securities held by a research analyst account. Nomura Global Financial Products Inc. (NGFP) Nomura Derivative Products Inc. (NDP) and Nomura International plc. (NIplc) are registered with the Commodities Futures Trading Commission and the National Futures Association (NFA) as swap dealers. NGFP, NDPI, and NIplc are generally engaged in the trading of swaps and other derivative products, any of which may be the subject of this report. ## Distribution of ratings (Nomura Group) The distribution of all ratings published by Nomura Group Global Equity Research is as follows: 60% have been assigned a Buy rating which, for purposes of mandatory disclosures, are classified as a Buy rating; 34% of companies with this rating are investment banking clients of the Nomura Group\*. 0% of companies (which are admitted to trading on a regulated market in the EEA) with this rating were supplied material services\*\* by the Nomura Group. 38% have been assigned a Neutral rating which, for purposes of mandatory disclosures, is classified as a Hold rating; 57% of companies with this rating are investment banking clients of the Nomura Group\*. 0% of companies (which are admitted to trading on a regulated market in the EEA) with this rating were supplied material services by the Nomura Group 2% have been assigned a Reduce rating which, for purposes of mandatory disclosures, are classified as a Sell rating; 17% of companies with this rating are investment banking clients of the Nomura Group\*. 0% of companies (which are admitted to trading on a regulated market in the EEA) with this rating were supplied material services by the Nomura Group. As at 30 September 2026. \*The Nomura Group as defined in the Disclaimer section at the end of this report. \*\* As defined by the EU Market Abuse Regulation ## Definition of Nomura Group's equity research rating system and sectors The rating system is a relative system, indicating expected performance against a specific benchmark identified for each individual stock, subject to limited management discretion. An analyst's target price is an assessment of the current intrinsic fair value of the stock based on an appropriate valuation methodology determined by the analyst. Valuation methodologies include, but are not limited to, discounted cash flow analysis, expected return on equity and…

报告将负自由现金流解释为资本开支前置与5至20年现金回报回收之间的“J曲线”时间错配，而非云业务已经失去盈利能力；同时提出维持并加速投资、放缓或延后投资、合资或租赁、未来现金流出售四种资本安排。报告还以Oracle的SPV、租赁和证券化融资方案说明大型科技公司可能继续投资，并以存储芯片长期协议覆盖率及利润保证估算支持其存储板块抛售过度的判断。