When $150 Billion Looks Small

本期播客节目发布于2026年9月29日，主持人Tyler Crowe与Rachel Warren和Lou Whiteman一同，深入探讨了当前的市场事件和行业见解。 节目开篇讨论了NVIDIA最近的公告，包括一项1500亿美元的股票回购协议，被称为“史上最大”，以及一份关于AI安全的新闻稿。Crowe立即提出了他的“独到见解”，称这次回购“太小了”，尽管加上此前的回购总额已达2500亿美元。他认为，1500亿美元仅占NVIDIA市值的约2%，声称要真正“震惊投资者”，回购金额应该“至少是这个数字的三倍”，即约占其市值的5%，鉴于该公司拥有1340亿美元的自由现金流，他建议进行5000亿美元的回购。 Rachel Warren指出，尽管1500亿美元这一数字具有历史意义，但对于NVIDIA庞大的市值来说，它需要被“适度衡量”。她提到NVIDIA正在产生“巨大的现金流”，并且拥有的资本超出了其在研发方面的再投资能力，使得回购成为一个自然的结果。AI安全声明与回购同时发布被认为是故意的，旨在“帮助投资者更好地接受（这一举措）”。NVIDIA正通过一个名为OpenShell的开源框架和一个名为Sentry的硬件看门狗来解决AI安全问题，旨在建立一个“通用的安全行业标准”，以防止自主AI面临监管冻结。 Lou Whiteman认为NVIDIA已经“用尽了惊艳市场的方法”，而且尽管业绩强劲，其股价“自春季以来一直停滞不前”。他将此次回购视为“又一次可能失败的尝试，试图说：‘嘿，还记得我们吗？’”讨论随后转向NVIDIA潜在的隐藏风险。Rachel Warren强调了“循环融资”，指出NVIDIA“实质上扮演着许多AI行业的中央银行角色”，拥有13个公开股权投资、超过200个私人股权投资，以及200亿美元的数据中心租赁承诺，如果AI基础设施建设放缓，这可能令人担忧。她还质疑了持续数十亿美元GPU升级的假设，指出如果旧款GPU“完全足以满足80%或更多的企业推理工作负载”，那么如果“现有硬件表现过好”，NVIDIA可能会面临“增长放缓”。 Lou补充说，“AI周期的最佳时期已经过去”，他提到Google、Amazon、Microsoft、Meta和OpenAI等主要科技公司正在开发自己的芯片，加剧了竞争。他还质疑“前沿模型是否不再那么重要”，即最新、最昂贵的芯片在预期规模下是否真的有必要。他指出NVIDIA的历史周期性，提到每股收益在2023年还“低于每年1美元”，现在已达8美元，这引发了对当前盈利水平可持续性及其看似“便宜”的市盈率的质疑。 对话转向国防和安全行业，重点关注通过特殊目的收购公司（SPACs）上市的新一波公司，例如网络安全公司Red Lattice，预计到2026年底将以12.5亿美元与Bold Eagle Acquisition Corp.合并。Lou Whiteman凭借其经验解释说，SPACs允许早期公司比传统IPO更自由地讨论未来前景。他告诫投资者，新公司很难进入五角大楼市场，建议通过Lockheed Martin等“主要承包商”寻找“切入点”。他警告称，一些“新颖技术”可能已被老牌承包商归类并正在开发中，并强调由于分类问题，尽职调查具有挑战性。他强调，需要具备政府经验的领导层和拥有安全许可的员工来驾驭政府官僚机构。 Rachel Warren补充说，成功的原型产品不易转化为官方军事项目，漫长的资金等待期常常耗尽公司的资本。她还指出，国防公司最盈利的业务部门通常是保密的，这使得投资者在对收入进行建模时“无法获知确切的合同条款、客户身份和基本单位经济效益”。Anduril和Shield AI等特定公司被提及为潜在的IPO候选者，其中Anduril因其令人印象深刻的运营和近期获得的200亿美元美国陆军合同而受到关注，尽管Rachel认为其公开上市可能会被推迟。 最后，听众来信环节回答了Cam关于Klarna的问题，该公司股价下跌了20%，尽管其季度业绩看似“表现出色”。Rachel解释说，尽管Klarna报告了创纪录的收入（10亿美元，增长30%）和意外的净利润（900万美元），但该公司将其全年收入预期大幅下调至远低于市场普遍预期。管理层提到了“德国可自由支配消费者支出大幅放缓”（德国是Klarna最大且最盈利的地区），6亿美元的汇率逆风，以及其首席财务官和首席营销官的出人意料的离职。Lou补充说，“先买后付”行业尚未经历“真正的经济衰退”的考验，这使得其长期生存能力在完整的市场周期结束之前仍不确定。他还对Klarna是否是他首选的“先买后付”公司表示怀疑。Crowe总结说，来自传统银行和NewBank、Revolut等其他金融科技公司日益激烈的竞争将使Klarna的客户获取和增长变得更加困难。

This podcast episode, published on 2026-09-29, features host Tyler Crowe with Rachel Warren and Lou Whiteman, delving into current market events and industry insights. The episode opens with a discussion on NVIDIA's recent announcements, including a $150 billion share repurchase agreement, touted as the "biggest in history," alongside a press release on AI safety. Crowe immediately offers a "hot take," calling the repurchase "way too small," despite it adding to a total of $250 billion for buybacks. He argues that $150 billion represents only about 2% of NVIDIA's market cap, asserting that to truly "wow investors," it should be "at least triple this," or around 5% of its market cap, suggesting a $500 billion buyback given the company's $134 billion in free cash flow. Rachel Warren points out that the $150 billion figure, while historical, needs to be "right sized" for NVIDIA's immense market capitalization. She notes that NVIDIA is generating "tremendous amounts of cash flow" and has more capital than it can reinvest in R&D, making the buyback a natural outcome. The timing of the AI safety announcement alongside the buyback is seen as deliberate, "to help the medicine go down" for investors. NVIDIA is addressing AI safety through an open-source framework called OpenShell and a hardware watchdog named Sentry, aiming to establish a "universal industry standard for safety" to prevent regulatory freezes on autonomous AI. Lou Whiteman suggests that NVIDIA has "run out of ways to wow the market," and the stock has been stagnant "since the spring," despite strong earnings. He views the buyback as "another maybe failed attempt to say, hey, remember us?" The discussion then shifts to potential hidden risks for NVIDIA. Rachel Warren highlights "circular funding," where NVIDIA acts "essentially like a central bank for a lot of the AI industry," with 13 public stakes, over 200 private stakes, and $20 billion in data center lease commitments, which could be concerning if the AI build-out slows. She also questions the assumption of constant multi-billion-dollar GPU upgrades, noting that if older GPUs are "perfectly adequate for 80% or more of enterprise inference workloads," NVIDIA could face a "slowdown in growth" if "the existing hardware works too well." Lou adds that "the best is now behind them for the AI cycle," citing major tech companies like Google, Amazon, Microsoft, Meta, and OpenAI developing their own chips, increasing competition. He also questions if "frontier models don't matter as much," whether the latest, most expensive chips are truly necessary at the expected scale. He points out NVIDIA's historical cyclicality, noting earnings per share was "under a dollar per year" as recently as 2023, now at $8, raising questions about the sustainability of current earnings levels and its seemingly "cheap" price-to-earnings ratio. The conversation transitions to the defense and security industry, focusing on a new wave of companies going public via SPACs, such as Red Lattice, a cybersecurity firm expected to merge with Bold Eagle Acquisition Corp. for $1.25 billion by the end of 2026. Lou Whiteman, drawing on his experience, explains that SPACs allow early-stage companies to discuss future prospects more freely than traditional IPOs. He cautions investors that the Pentagon is difficult to access for new companies, recommending looking for "on-ramps" through "primes" like Lockheed Martin. He warns against "novel tech" that might already be classified and under development by established contractors, emphasizing that due diligence is challenging due to classification. He stresses the need for leadership with government experience and employees with security clearances to navigate government bureaucracy. Rachel Warren adds that successful prototypes don't easily convert to official military programs, with long waits for funding often burning through a firm's capital. She also notes that a defense company's most profitable segments are often classified, forcing investors to model revenue "blind to the exact contract terms, customer identities, underlying unit economics." Specific companies like Anduril and Shield AI are mentioned as potential IPO candidates, with Anduril highlighted for its impressive operations and recent $20 billion U.S. Army contract, though Rachel suggests a public debut might still be pushed back. Finally, the mailbag addresses Cam's question about Klarna, which saw its stock drop 20% despite a seemingly "blowout quarter." Rachel explains that while Klarna reported record revenue ($1 billion, up 30%) and a surprise net profit ($9 million), the company lowered its full-year revenue outlook significantly below consensus. Management cited a "major slowdown in discretionary consumer spending in Germany," Klarna's largest and most profitable region, a $600 million foreign currency headwind, and the unexpected departure of both its CFO and CMO. Lou adds that the "buy now, pay later" industry has not been tested by a "real downturn," making its long-term viability uncertain until a full market cycle plays out. He also expresses skepticism about Klarna being his first choice among BNPL companies. Crowe concludes that the increasing competition from traditional banks and other fintechs like NewBank and Revolut will make customer acquisition and growth harder for Klarna.