以下是内容的中文翻译： 《Motley Fool Hidden Gems Investing》播客节目，由 Tyler Crowe 主持，嘉宾 Matt Frankel 和 John Quest 共同参与，节目伊始深入分析了 **NVIDIA** 最新的财报。他们形容这份财报“不令人失望”，并且“远超盈利预期和业绩指引”。录制节目时，NVIDIA 股价上涨了 7%，此份财报被称为“最受期待的财报”。NVIDIA 公布了“令人难以置信的”106% 的同比增长收入，并预计来年营收将增长 70%，远高于华尔街 45% 的预期。该公司指出其“受制于供应”，暗示如果能物理上生产更多产品，增长率还会更高。超大规模数据中心（Hyperscaler）的资本支出（CapEx）预计将从今年的 8000 亿美元飙升至明年的 1.3 万亿美元。尽管该股票的远期市盈率为 25 倍，这已经计入了部分增长放缓的预期，但当前 70% 的增长指引仍然是一个强劲的利好因素。主要担忧是内存成本上升可能导致利润率受挤压，尽管目前这被巨大的营收增长所抵消。 讨论随后转向了 AI 基础设施生态系统中，那些可能从这波浪潮中受益或受损的“卖铲子”公司。 * **受损者：** John Quest 指出了 **惠普 (HP)**，理由是内存成本上升。内存公司正从消费市场转向服务 AI 领域，这导致消费设备内存成本增加和短缺，对像惠普这样缺乏 AI 服务器业务顺风的公司影响尤为严重。Matt Frankel 将 **Garmin (GRMN)** 加入到受损者名单中，因为该公司对第三方内存的依赖可能导致设备价格上涨或利润率受挤压。 * **受益者：** Quest 重点提到了 **Credo Technologies (CRDO)**，一家营收 13 亿美元的小公司，专门从事 GPU 互联技术。随着 NVIDIA 预计的巨额支出，Credo 预计 80% 的增长使其成为潜在的“隐藏赢家”。Frankel 指向了 **Onto Innovation (ONTO)**，该公司生产内存产品的检测工具，受益于内存需求的增长。他还提到了 **IES Holdings (IESC)**，该公司为数据中心设计电气系统，并强调随着预计 1.3 万亿美元的超大规模数据中心资本支出，电力是一个重要的制约因素。总体结论是，芯片制造商和内存公司亟需大幅扩充产能，以满足这种“贪得无厌”的需求。 播客随后讨论了 **CrowdStrike** 的财报，这也是一个“爆发式增长的季度”，股价上涨了 18%。Matt Frankel 详细指出，该公司营收同比增长 26%，净新增年度经常性收入（ARR）创纪录地增长 51%，远超管理层指引。这表明 AI 正在以超出预期的速度推动网络安全需求。剩余履约义务（积压订单）增长 49%，超过 100 亿美元。John Quest 引用了 CrowdStrike 首席执行官的话，强调“AI 正在网络安全公司中划出一条清晰的界线：一部分公司解决问题，另一部分公司则制造更多问题。”管理层提到“Mythos 时刻”（一个据称突破控制的 AI 模型）有助于他们的业务，因为公司寻求更强的保护。Quest 还指出 CrowdStrike 的新“灵活计费模式”，它允许客户更灵活地采用模块，并能推动 ARR 更快增长。 主持人随后讨论了 **CrowdStrike 的估值**，该公司的市销率高达“过高”的 37 倍。Tyler Crowe 指出，这意味着公司需要 37 年才能通过营收达到其市值，尽管他承认营收预计会增长。John Quest 尽管认可该公司业务，但坦言他曾因仅仅担忧估值而卖掉基本面强劲的股票，后来“后悔不已”。他认为 CrowdStrike 是一个“非凡的业务和行业”，估值可能不是主要考量。然而，Matt Frankel 反驳称“估值至少总会有一点影响。”他引用了“巨额”股权激励，这在 GAAP（通用会计准则）基础上有效地抹去了营业收入，并导致稀释后股份数量的增长，使得 CrowdStrike 成为“一家出色的公司，但价格令人不安。” * **替代品：** Quest 建议了 **Rubrik (RBRK)**，一家网络安全公司（并非直接竞争对手），市销率为 14 倍，增长更快，自由现金流为正。Frankel 推荐了 **Zscaler**，该公司与 CrowdStrike 合作，并提供“更容易接受的估值”。 最后，播客回答了一个听众提问：本地 AI 使用（例如在新 Apple Mac 上）是否对云 AI 公司构成威胁。 * John Quest 认为这并非威胁，而是“不同的使用场景”。他相信个人将越来越倾向于个性化的本地 AI，而医院、学校和政府等机构将出于数据隐私考虑选择本地 AI。然而，云 AI 将对创新至关重要。他认为“未来两者并存”。 * Matt Frankel 也表达了相同观点，称这是“水涨船高”的情景。数据中心保留了内存采购能力和规模等关键优势。他还认为本地 AI 可能是“对网络安全来说最好的事情之一”，因为在企业网络中保护数千个新的 AI 端点将推动 CrowdStrike 和 Zscaler 等公司的需求。Tyler Crowe 指出，科技行业可能从云迁移又回到本地硬件的这种讽刺。

The "Motley Fool Hidden Gems Investing" podcast, hosted by Tyler Crowe with guests Matt Frankel and John Quest, began by dissecting **NVIDIA's** latest earnings report, which they described as "not disappointing" and "smashed earnings expectations and guidance." Shares were up 7% as they taped, following what was called the "most anticipated earnings report." NVIDIA reported an "unbelievable" 106% year-over-year revenue growth and issued guidance for 70% revenue growth in the coming year, significantly above Wall Street's 45% expectation. The company noted it is "supply constrained," implying growth would be even higher if it could physically produce more. Hyperscaler CapEx is projected to soar from $800 billion this year to $1.3 trillion next year. While the stock trades at 25 times forward earnings, pricing in some deceleration, the current 70% guidance is a strong positive. The main concern raised was potential margin compression due to higher memory costs, though this is currently offset by massive revenue growth. The discussion then shifted to "picks and shovels" companies in the AI infrastructure ecosystem that stand to win or lose from this surge. * **Losers:** John Quest identified **HP**, citing higher memory costs. Memory companies are pivoting away from the consumer market to serve AI, leading to increased costs and shortages for consumer devices, which disproportionately impacts companies like HP that lack an AI server business tailwind. Matt Frankel added **Garmin (GRMN)** to this list, as its reliance on third-party memory could lead to higher device prices or margin compression. * **Winners:** Quest highlighted **Credo Technologies (CRDO)**, a small $1.3 billion revenue company specializing in GPU interconnectivity. With NVIDIA projecting massive spending, Credo's 80% expected growth makes it a potential "hidden winner." Frankel pointed to **Onto Innovation (ONTO)**, which makes inspection tools for memory products, benefiting from the increased demand for memory. He also named **IES Holdings (IESC)**, which designs electrical systems for data centers, emphasizing that power is a significant constraint with the projected $1.3 trillion hyperscaler CapEx. The overall takeaway was the critical need for significant capacity expansion from chip manufacturers and memory companies to meet this insatiable demand. The podcast then covered **CrowdStrike's** earnings, another "blowout quarter" with shares up 18%. Matt Frankel detailed a 26% year-over-year revenue growth and a record 51% increase in net new annual recurring revenue (ARR), significantly exceeding management's guidance. This suggests AI is fueling cybersecurity demand more strongly than anticipated. The remaining performance obligation (backlog) grew 49% to over $10 billion. John Quest quoted CrowdStrike's CEO, highlighting that "AI is driving a clear divide between the cybersecurity companies that solve problems and those that compound problems." The "Mythos moment" (an AI model reportedly breaking containment) was cited by management as benefiting their business, as companies sought enhanced protection. Quest also noted CrowdStrike's new "flex billing model," which allows customers more flexibility in adopting modules and can drive faster ARR uptake. The hosts then tackled **CrowdStrike's valuation**, which trades at an "exorbitant" 37 times sales. Tyler Crowe pointed out that this implies it would take 37 years for the company to generate its market cap in revenue, acknowledging that revenue is expected to grow. John Quest, despite valuing the business, confessed he'd "lived to regret" selling stocks solely based on valuation concerns when the underlying business is strong. He views CrowdStrike as a "phenomenal business and industry" where valuation might be less of a primary concern. Matt Frankel, however, pushed back, arguing that "valuation always matters at least a little bit." He cited the "massive" stock-based compensation, which effectively wipes out operating income on a GAAP basis and leads to diluted share growth, making it a "wonderful business at an uncomfortable price." * **Alternatives:** Quest suggested **Rubrik (RBRK)**, a cybersecurity company (not a direct competitor) trading at 14 times sales, with faster growth and positive free cash flow. Frankel recommended **Zscaler**, which partners with CrowdStrike and offers a "much more palatable valuation." Finally, the podcast addressed a listener question about whether local AI usage (e.g., on new Apple Macs) poses a threat to cloud AI companies. * John Quest argued it's not a threat but rather "different use cases." He believes individuals will increasingly want personalized local AI, and institutions like hospitals, schools, and governments will opt for local AI for data privacy. However, cloud AI will remain crucial for innovation. He sees "both in the future." * Matt Frankel echoed this, stating it's a "rising tide lifts all ships" scenario. Data centers retain key advantages like greater memory purchasing power and scale. He also suggested that local AI could be "one of the best things that ever happened to cybersecurity," as securing thousands of new AI endpoints within enterprise networks would drive demand for companies like CrowdStrike and Zscaler. Tyler Crowe observed the irony of the tech industry potentially moving from cloud migration back to local hardware.