MU 暴涨引发估值大辩论；NVDA 沦为 funding short 遭热议

2026-05-26 周二 (ET) · 主帖 58 / 回复 235
以下部分引用 AI 总结，现阶段 AI 仍然有幻觉。请以内容的原文为准。
一些重点
  1. MU 暴涨逻辑Arcuri 将 MU 目标价上调至 $1625，引发多空双方就周期性资产能否享受世俗化溢价展开激烈辩论。
  2. NVDA 沦为提款机NVDA 表现疲软，被指沦为高 Beta 半导体股票的 funding short，但多头坚守其基本面。
  3. 太空数据中心Gavin Baker 预测 2028-2030 年实现轨道计算，引发对地面数据中心长期逻辑及利益相关方的探讨。
  4. ZS 资本开支激增ZS 下调 FCF 指引，主因是内存等组件价格上涨导致资本开支增加，凸显 AI 带来的短期通胀压力。
  5. Anthropic 营收超 OpenAIThe Information 报道 Anthropic 年化营收达 $45B，比 OpenAI 高出约 35%。
  6. FROG/GTLB 逻辑映射FROG 晨报指出安全/合规是主要变现载体，该逻辑同样适用于 GTLB Ultimate。
  7. 宏观与资金轮动通胀粘性担忧压制市场广度，资金从科技巨头流出，AI 半导体成为唯一确定性主线。

1 AI/半导体主线

MU 估值大辩论 争论·估值重估

MU 估值争议的核心 = 周期性资产能否享受世俗化溢价 —— Arcuri 研报引爆多空对决。 多头认为 HBM 带来持久盈利，空头警告无视周期性将导致崩盘。

  • 目标价 $1625 引群嘲: 卖方将 MU 目标价上调至 $1625，理由是 MU 的 P/E 不应与 NVDA 有太大差异[1]Zerocool · 05-26"..."[2]Ardo · 05-26"..."。群内空头嘲讽这是在“见顶”[3]ValueWithoutACatalyst · 05-26"..."。
  • 多头逻辑：世代计算转移 群内多头认为 HBM 三寡头格局 = 持久盈利 → MU 应获更高估值（世代计算转移）[4]Tae Kim · 05-26"..."[5]Tae Kim · 05-26"..."。另一位投资者补充，MU 增加了占需求 50%+ 的新客户群，降低了周期性[6]J Cap · 05-26"..."。
  • 空头反驳：警惕周期性 有空头指出 LTA（长期协议）占比不到 15%，不足以将周期性资产转化为世俗化资产，15x P/E 是合理的，但 50x 则过于夸张[7]Arun Dhar · 05-26"..."[7]Arun Dhar · 05-26"..."。群内空头警告，无视历史估值和周期性是危险的[8]ValueWithoutACatalyst · 05-26"..."。
  • GS 视角的估值真相: 高盛指出，尽管 MU 股价大涨，但远期 P/E 仍接近 10x，因为盈利预期也同步上涨了 10 倍。这意味着股价上涨是盈利驱动，而非单纯的估值扩张[9]H-Cap · 05-26"..."[9]H-Cap · 05-26"..."。

MU 的定价逻辑已生变 = 市场正将其从周期复苏转向 AI 驱动的长期盈利能力定价。

NVDA 沦为提款机 观察·资金流向

NVDA 沦为 funding short = 资金流向高 Beta 半导体，导致其表现疲软。 但多头认为其基本面依然强劲，估值便宜，等待资金回流。

  • Funding Short 现象: 群内讨论指出，NVDA 成为做空资金的来源，因为其他 AI 半导体股票每天都在大涨[10]TMT Breakout · 05-26"..."。有投资者补充，只要其他半导体有更大上涨潜力，资金就不会流入 NVDA[11]Ardo · 05-26"..."。
  • 多头坚守：基本面无虞 多头强调 NVDA 估值便宜，资本回报加速，且 2027 年共识预期仍低 20%[12]RI · 05-26"..."[12]RI · 05-26"..."。他认为最终资金会回流 NVDA[13]RI · 05-26"..."。
  • 估值上限担忧: 有观察者认为 NVDA 的上行空间受限，市场可能不会再给它超过 20 倍的远期市盈率[14]Ted · 05-26"..."。

太空数据中心叙事 观察·远期叙事

Gavin Baker 预测 2028-2030 年实现轨道计算，引发对地面数据中心长期逻辑的探讨。

  • 轨道计算时间表: Gavin Baker 预测 2028 下半年到 2030 上半年轨道计算将成为现实[15]Maverick · 05-26"..."。
  • 对地面的影响: 有投资者质疑这是否意味着应该做空地面数据中心和电力/冷却公司[15]Maverick · 05-26"..."。另一位投资者反驳称，轨道计算并不意味着地球瓶颈会立即消失[16]AGdoom22 · 05-26"..."。
  • 设备维护与利益相关: 针对太空设备的维护问题，讨论指出这属于“婴儿死亡率”问题，设备通常在初期故障，通过前期老化测试可剔除故障件[17]hyc · 05-26"..."。同时有人指出，鼓吹该叙事的人多在 SpaceX 有利益相关，意在为 IPO 抬价[18]ALC · 05-26"..."。

2 软件/应用

ZS 资本开支激增 空·成本通胀

AI 带来短期通胀压力 = ZS 因内存等组件涨价下调 FCF 指引。

  • FCF 指引下调: ZS 将全年 FCF 利润率指引从 ~26.75% 下调至 ~23%，并警告下财年资本开支占营收比重将上升 ~200bp[19]H-Cap · 05-27"..."[19]H-Cap · 05-27"..."。
  • 组件通胀是主因: 资本开支增加并非因为购买更多设备，而是因为组件价格（尤其是内存）飙升[19]H-Cap · 05-27"..."。SemiAnalysis 报告称，到 2026 年，内存将占超大规模企业资本开支的 ~30%[19]H-Cap · 05-27"..."。

Anthropic 营收超 OpenAI 观察·行业格局

The Information 报道 Anthropic 年化营收达 $45B，比 OpenAI 高出约 35%。

  • 营收数据对比: Anthropic 年化营收接近 $45B，而 OpenAI 最近突破 $30B，估计目前在 $33B 左右[20]Akh · 05-27"..."[20]Akh · 05-27"..."。

FROG/GTLB 逻辑映射 观察·逻辑映射

FROG 晨报指出安全/合规是主要变现载体，该逻辑同样适用于 GTLB Ultimate。

  • 安全附加逻辑: 安全/合规 = FROG 主要变现载体（52% 高级安全附加率）→ 同理适用于 GitLab Ultimate，渠道检查显示 Ultimate 占 Q4 ARR 的 ~57%，同比增长 ~35%[21]Randy Marsh · 05-27"..."[21]Randy Marsh · 05-27"..."。
  • AI 对冲效应: AI 生成更多代码，但去重（dedup）吸收了工件（artifact）的增加，而 AI 驱动的生产力吸收了席位（seat）的增加，导致顺风效应被结构性减弱[21]Randy Marsh · 05-27"..."。

3 宏观/其他

宏观与资金轮动 观察·宏观情绪

通胀粘性压制市场广度 = 资金从 Mega 科技股流出，AI 成为唯一确定性主线。

  • 通胀担忧持续: 尽管有停战传闻，但市场广度欠佳，群内指出通胀粘性担忧持续，压制了 AI 瓶颈之外的多数资产表现[22]kkhajavi7 · 05-26"..."[23]Bubbles · 05-26"..."。
  • 资金轮动与 AI 确定性: 科技巨头（Mega 7）成为资金轮动的提款机，而 AI 半导体因基本面确定性极高，持续吸纳资金，市场更害怕错过半导体上行而非其他风险[24]RI · 05-26"..."[25]RI · 05-26"..."。
数据新鲜度群聊 08-18 ✓卖方 08-18 ✓Wrap 断档50天 ✕News 断档None天 ✕X 断档47天 ✕页面生成 08-19
📊 今日群聊 ATOMS
44 atoms 🟦 fact 16 🟥 take 28 ▲ 16 / ▼ 20 / ◆ 8
实体 (今日): MU (Micron) (14), NVDA (5), Zscaler (ZS) (3), RI (2), Hassabis (Demis) (2)
主题 (本周): 估值 (12), 营收预测 (8), 产能 (6), 地缘政治 (6), 定价权 (5)
Δ 今日变化
  • 🔄 MU 今日转 bearish (7 条, 历史主导 bullish)
  • 🔄 NVDA 今日转 bearish (3 条, 历史主导 bullish)
  • ⏳ AWS forecast 本月到期: 「Q2 营业利润指引需达 $14-15bn 高端才能推动股价」(2026-04-27 由 chat 提出)
  • ⏳ Meta forecast 本月到期: 「META 指引需达 $61bn (28% yoy) 以支撑 AI 投资」(2026-04-27 由 chat 提出)
  • 🌱 新叙事: 轨道计算 (今日 3 条, 库内首见)
  • 🌱 新叙事: 年化营收 (今日 2 条, 库内首见)
→ 全部横切视图 (Screens)
📜 今日原 atoms (按 entity 折叠展开)
MU · 14 atoms
  • 🟦 ▲ 目标价 · 目标价/估值
    Arcuri 将 MU 目标价上调至 $1625
  • 🟥 ▲ 估值逻辑 · 估值/PE比较
    Arcuri: "we see no reason why MU should trade a whole lot differently than NVDA in terms of P/E"
  • 🟥 ▼ 市场情绪 · 市场情绪/顶部信号
    ValueWithoutACatalyst: He's putting the top in
  • 🟥 ▲ 估值重评逻辑 · 估值重评/HBM
    Tae Kim: HBM - 3 company oligopoly is here to stay = durable multi-year earnings power = higher multiple. this time really is different.
  • 🟦 ▲ 客户集中度与周期性 · 客户结构/周期性降低
    J Cap: MU added a brand new customer base that now makes up 50%+ of their annual demand and not tied to consumer end markets so a much less cyclical end user
  • 🟥 ▼ LTA 与周期性转化 · LTA合同/周期性
    Arun Dhar: LTAs being a semi-permanent panacea (which are < 15% of contracts signed and _all_ of them, according to Joe at MS, have fixed+floating structure). But to say LTAs now transform a cyclical
  • 🟥 ▼ 估值风险 · 估值风险
    ValueWithoutACatalyst: Ignoring multiples/historicals/generalists coming in, is how you blow up
  • 🟦 ◆ 远期 P/E 与盈利驱动关系 · 远期PE/盈利驱动
    高盛: despite MU being up 21% on the day, the forward P/E multiple is approaching only ~10x consensus — because earnings have also risen roughly 10x over the same 12-month period, leaving the multiple b
  • 🟦 ▼ 周期性底部估值 · 周期底部估值
    J Cap: last 3 cycle peaks have been 8-10x for MU
  • 🟥 ▲ 多空分歧本质 · 多空分歧/周期性质
    Tae Kim: the main difference is you think this is a typical historical cycle versus what I believe (generational computing shift)
  • 🟥 ▲ 25x25 目标 · 估值目标
    Tae Kim: 25 x 25. Michael Dell. Look it up.
  • 🟥 ▼ 供应风险 · 供应风险/产能
    Arun Dhar: There is no divine rule, other than supply discipline why DRAM and NAND supply can't ramp (a lot) in the coming years.
  • 🟦 ▼ 期权市场 signal · 期权信号/看跌情绪
    H-Cap: Another day of massive MU put buying. Been like this for 4-6 weeks. Yet sister SNDK massively bought via calls. Strange day in markets
  • 🟦 ▼ 股指影响力与再平衡 · 股指权重/再平衡
    ALC: MU is currently 3.8% of the Qs despite mkt cap below eg AVGO and META, which makes it responsible for ~70bp of the move today
NVDA · 5 atoms
  • 🟥 ▼ 资金流向角色 · 资金流向
    TMT Breakout: NVDA is the go to funding short given everything else in AI semis goes up 3%+ a day
  • 🟥 ▼ 相对吸引力 · 相对吸引力/资金轮动
    Ardo: there is just very little reason to put new $$ into NVDA as long as all other semis / dc plays have so much more potential to move, everyone is performance chasing
  • 🟥 ▲ 基本面与估值 · 估值/基本面
    RI: NVDA is up 16.3% YTD over 5 months with a very cheap valuation, accelerated capital return, and ridiculous fundamentals. Consensus is still 20% too low for ‘27.
  • 🟥 ▲ 资金回流预期 · 资金回流预期
    RI: Eventually the upside on everything else gets limited and you rotate back into NVDA.
  • 🟥 ▼ 远期市盈率预估 · 远期市盈率
    Ted: I think market has basically said we are never getting a >20x fwd mult on this thing again
Zscaler (ZS) · 3 atoms
  • 🟦 ▼ FCF 利润率指引 · FCF利润率/业绩指引
    ZS: reduction in free cash flow margin guidance for the year (from ~26.75% to ~23%)
  • 🟦 ▼ 下财年资本开支预期 · 资本开支预期
    ZS warned that capex as a percentage of revenue will rise ~200bp next fiscal year
  • 🟥 ◆ 资本开支上涨原因 · 资本开支/成本构成
    VK: ZS 的 capex outlook isn't so much because ZS is buying a lot of extra equipment, but instead due to the spike in component prices, especially memory
RI · 2 atoms
  • 🟥 ▲ AI 半导体确定性 · 市场情绪/AI半导体
    RI: this mkt is more fearful of missing out on semis upside than anything else.
  • 🟥 ▲ 资金轮动方向 · 资金轮动/半导体
    RI: the megas are the flow donors for a rotation trade. Any confirmation of more compute just results in marginal dollars going to higher beta semis.
Hassabis (Demis) · 2 atoms
  • 🟥 ▲ AGI 时间表修正 · AGI时间表/技术路线
    Hassabis still sees 2030 as his central estimate for AGI, but now says 2029 is possible.
  • 🟦 ◆ 递归自我改进系统的时间点 · 递归自我改进/AI发展
    Hassabis: "We're not yet at the point where the systems are getting better on their own, but the pace of development is clearly accelerating."
Gavin Baker · 1 atoms
  • 🟥 ◆ 轨道计算时间表预测 · 轨道计算/时间表预测
    Gavin Baker: "Second-half of 28 to first half of 2030 would be my point prediction [for orbital compute being a reality]."
Maverick · 1 atoms
  • 🟥 ▼ 对地面数据中心的短期影响 · 轨道计算/数据中心需求
    Maverick: Doesn't that mean we want to start shorting all the datacenter, neoclouds & power/cooling names soon?
AGdoom22 · 1 atoms
  • 🟥 ▼ 轨道计算与地面瓶颈 · 轨道计算/地面瓶颈
    AGdoom22: orbital compute doesn't mean the Earth bottleneck goes away instantly
SpaceX / 太空数据中心 · 1 atoms
  • 🟦 ◆ 设备维护理论 · 设备可靠性/测试方法论
    hyc: Elon said it was an infant mortality problem. Equipment tends to fail at the beginning so if you burn them in you'll weed out the failures
All-In Podcast 嘉宾 (Gavin Baker 等) · 1 atoms
  • 🟥 ▼ 利益冲突 · 利益冲突/IPO动机
    ALC: I have a sense that all those folks in the all-in pod are in spaceX one way or another and are hyping up the story so they can exit at the highest possible bid in the IPO.
超大规模企业 (Hyperscalers) · 1 atoms
  • 🟦 ▲ 内存占 capex 比例预测 · 内存/资本开支占比
    SemiAnalysis: memory will account for ~30% of all hyperscaler capex in C26, up from ~8% in C23 and C24
Anthropic · 1 atoms
  • 🟦 ◆ 年化营收 · 年化营收
    The Information: Anthropic was recently generating revenue at an annualized rate of close to $45 billion.
OpenAI · 1 atoms
  • 🟦 ◆ 年化营收 · 年化营收
    The Information: estimate their annualized rate now is around $33 billion
JFrog (FROG) · 1 atoms
  • 🟦 ▲ 安全/合规变现能力 · 安全变现/附加率
    Stifel: security/compliance is the dominant monetization vector for FROG (52% premium security attach, near-universal expansion intent)
GitLab (GTLB) · 1 atoms
  • 🟦 ▲ Ultimate 版本的 ARR 占比与增速 · 产品渗透率/营收占比
    Randy Marsh: Our channel checks have flagged Ultimate at ~57% of Q4 ARR growing ~35% YoY
AI 对软件股的影响 · 1 atoms
  • 🟥 ▼ AI 对冲效应 · AI对冲效应/增长估算
    AI is generating more artifacts and more code, but dedup absorbs the artifact lift and AI-driven productivity absorbs the seat lift. Tailwind exists at both layers but is structurally muted vs. the bu
Bubbles · 1 atoms
  • 🟥 ▼ 通胀对非 AI 资产的压制 · 通胀/非AI资产
    Bubbles: inflation isnt going away and is crushing everything outside of AI bottlenecks
AppLovin (APP) · 1 atoms
  • 🟥 ▼ META 对非 IDFA 流量竞价时间表预期 · iOS流量/竞价时间表
    Edgewater note: 'META is unlikely to begin bidding on non-IDFA traffic on iOS anytime soon.' At the absolute earliest, non-IDFA bidding might be a 2027 event
Amazon (AMZN) · 1 atoms
  • 🟥 ▼ Trainium 芯片对 NVDA 的威胁评估 · 竞争威胁/芯片替代
    Jim Cramer: "Amazon's Trainium chips were underestimated by me. The new ones have more long-term value than I thought and are more of a threat to Nvidia than I thought."
Reddit (RDDT) · 1 atoms
  • 🟥 ▲ 对基础模型公司的数据价值定价能力 · 数据定价/议价能力
    Mike on RDDT: then why don't they charge more? they should be getting paid 10x what they are getting paid from these foundation model companies
Semco · 1 atoms
  • 🟥 ◆ 2027 年业绩预期分歧 · 业绩预期/分歧
    MomoCap on SEMCO ests: JPM and generalists seem to circle 40k 27 but wondering if specialists have grounds for higher
DraftKings (DKNG) · 1 atoms
  • 🟥 ▲ 股价下跌与 Kuzma 推文关系 · 股价驱动因素/社交媒体影响
    kkhajavi7: that kuz tweet like saying cigarettes cause cancer - zero shot that would drive it down 5 percent
GFS (GlobalFoundries) · 1 atoms
  • 🟦 ▼ 大宗交易 · 大宗交易
    GLOBALFOUNDRIES HOLDER SAID TO OFFER BLOCK FOR $86.30-$86.80/EACH
SUMMARY COVERAGE

Market chatter · 总结覆盖率

历史口径未留档
该日生成时尚未记录覆盖率；未知不记为 0。