The Motley Fool《隐藏宝石投资》播客最近讨论了两家科技巨头——台湾积体电路制造公司（台积电，TSMC）和Meta Platforms——的加速增长，并指出这些趋势反驳了“AI需求正在见顶”的说法。尽管它们并非“隐藏公司”，但其重要的活动是全球AI市场的关键先行指标。
**台积电的加速增长：**
台积电（股票代码：TSM），全球最大的芯片制造商，生产了超过90%的最先进计算机芯片以及大约60%的所有计算机芯片。该公司最近公布的月度数据显示出强劲增长，六月份营收较去年同期猛增近68%。值得注意的是，通常受夏季季节性影响，六月份销售额会有所下降，但今年却环比五月份有所增长，这表明需求正在积极加速，而非放缓。根据播客文字记录，台积电2026年上半年（原文如此）的总营收达到约750亿美元（2.4万亿新台币），较2025年同期增长36%。凭借英伟达（NVIDIA）、苹果（Apple）和超微半导体（Advanced Micro Devices）等主要客户，台积电预计2026年AI芯片收入将超过400亿美元，约占其总营收的25%。该公司先进的生产线“订单排满”，凸显出主要科技公司庞大的资本支出。
**Meta的Hyperion数据中心扩建：**
Meta Platforms通过进一步扩建其位于路易斯安那州的Hyperion数据中心项目，进一步证实了强劲的AI趋势。该设施最初于2024年末宣布时预计投资100亿美元，去年年底增至270亿美元。现在，Meta计划对该设施本身直接投资约500亿美元。此次扩建包括再收购1400英亩土地，使总占地面积超过3200英亩——大约是纽约中央公园面积的四倍。该项目目前规划为5千兆瓦（GW）的设施（最初为2千兆瓦），总投资（包括电力基础设施）可能超过2500亿美元，预计到2036年才能完全建成。为满足巨大的电力需求，Meta正在资助建设10座新的天然气发电厂，并承诺向当地基础设施投入超过10亿美元。不断升级的成本反映了项目规模的扩大，以及现代数据中心所需的专用冷却、电力基础设施和先进芯片日益增长的费用，每吉瓦成本已比最初估计翻了一番。
**潜在受益者（隐藏宝石）：**
利用这些趋势，播客重点介绍了两家鲜为人知但有望从持续的AI基础设施繁荣中获益的公司：
1. **Comfort Systems (FIX)：** Matt Frankel介绍了Comfort Systems，一家领先的供暖、通风、空调（HVAC）、管道和电气承包商，专注于大型商业地产项目。该公司已成功转型至AI数据中心领域，这需要巨大且精确的冷却和电气工程。在最近一个季度，Comfort Systems实现了营收同比增长56%，每股收益翻倍，积压订单增长80%。随着数据中心日益采用复杂的液体冷却技术，Comfort Systems在处理此类项目规模和复杂性方面的专业知识，使其直接受益于台积电和Meta等公司大规模建设带来的资本支出增长。
2. **Celestica (CLS)：** Rachel Warren重点介绍了Celestica，一家电子制造服务公司，它充当着“AI基础设施的架构师”。Celestica接收英伟达和AMD等公司提供的先进芯片，将其安装到定制电路板上，并组装成大型液冷AI服务器机架。他们还制造高速网络交换机，使成千上万的芯片能够有效通信。Celestica在供应链中占据关键位置，将原始科技组件转化为超大规模数据中心的功能性超级计算机。随着“新芯片浪潮”的预期到来，以及科技巨头对定制设计的AI服务器机架以实现最佳效率的需求，Celestica在规模化构建复杂、定制系统方面的工程实力，使其在为期多年的硬件周期中占据有利地位，从而获得显著增长。
**听众提问：为读研卖股票？**
播客还回答了来自萨克拉门托的听众本（Ben）提出的问题，他纠结于是否要出售三分之一的投资组合来支付研究生学费，还是申请学生贷款。Jon Quast分享了一个关于他出售投资以资助购买汽车的个人经历。Matt Frankel建议，投资是“实现目标的一种手段”，为计划中的生活开支付款，特别是为了避免债务，是出售股票的合理理由。他将其视为将资本从一种生产性资产（股票）转移到另一种（通过教育获得的个人盈利能力），这可能带来更好的回报。他提醒听众考虑资本利得税和可用学生贷款的利率。Rachel Warren补充说，虽然对于长期投资者来说，出售股票感觉是违反直觉的，但投资组合应被视为“工具箱”，用于建设生活和实现财务自由。她总结道，投资于自己的教育是“押注于一种由你决定成功的资产”，从而赋能个人。
The Motley Fool Hidden Gems Investing podcast recently discussed the accelerating growth of two tech giants, Taiwan Semiconductor Manufacturing Company (TSMC) and Meta Platforms, arguing that these trends push back against the narrative that AI demand is peaking. While not "hidden companies," their significant activities serve as crucial leading indicators for the global AI market.
**TSMC's Accelerating Growth:**
TSMC (ticker TSM), the world's largest chip maker, fabricates over 90% of the most advanced computer chips and an estimated 60% of all computer chips. The company's recent monthly numbers demonstrate robust growth, with June revenue jumping nearly 68% compared to last year. Notably, June sales usually dip due to summer seasonality, but this year they rose from May, indicating actively accelerating demand rather than a slowdown. For the first half of 2026 (as stated in the transcript), TSMC's total revenue reached approximately $75 billion (2.4 trillion New Taiwan dollars), marking a 36% increase compared to the same period in 2025. With key clients including NVIDIA, Apple, and Advanced Micro Devices, TSMC is projected to generate over $40 billion from AI chip revenue in 2026, representing about 25% of its total revenue. The company's advanced manufacturing lines are "booked solid," underscoring the massive capital expenditure by major tech players.
**Meta's Hyperion Data Center Expansion:**
Meta Platforms further reinforced the strong AI trend with another expansion of its Hyperion data center project in Louisiana. Initially a $10 billion facility announced in late 2024, it was increased to $27 billion late last year. Now, Meta plans an estimated $50 billion direct investment for the facility itself. This expansion involves acquiring an additional 1,400 acres, bringing the total site to over 3,200 acres – roughly four times the size of New York's Central Park. The project, now envisioned as a 5-gigawatt facility (up from an original 2 gigawatts), could see total investment, including power infrastructure, potentially exceeding $250 billion, with the full build-out not expected until 2036. To meet the immense power demands, Meta is funding 10 new natural gas plants and committing over $1 billion towards local infrastructure. The escalating costs reflect both a larger scope and the rising expense of specialized cooling, power infrastructure, and advanced chips required for modern data centers, with the cost per gigawatt doubling from initial estimates.
**Under-the-Radar Beneficiaries (Hidden Gems):**
Leveraging these trends, the podcast highlighted two less-known companies poised to benefit from the ongoing AI infrastructure boom:
1. **Comfort Systems (FIX):** Matt Frankel introduced Comfort Systems, a leading heating, ventilation, air conditioning (HVAC), plumbing, and electrical contractor specializing in large-scale commercial properties. The company has successfully pivoted to AI data centers, which require enormous and precise cooling and electrical work. In the most recent quarter, Comfort Systems saw a 56% year-over-year revenue growth, a doubling of earnings per share, and an 80% increase in its backlog. As data centers increasingly adopt sophisticated liquid cooling, Comfort Systems' expertise in handling the scale and complexity of these projects makes it a direct beneficiary of increased CapEx from companies like TSMC and Meta's expansive build-outs.
2. **Celestica (CLS):** Rachel Warren highlighted Celestica, an electronics manufacturing services company that acts as an "architect for AI infrastructure." Celestica takes advanced chips from companies like NVIDIA and AMD, mounts them onto custom circuit boards, and assembles them into massive liquid-cooled AI server racks. They also build the high-speed networking switches that enable tens of thousands of these chips to communicate effectively. Celestica occupies a critical position in the supply chain, transforming raw tech components into functional supercomputers for hyperscalers. With a "tidal wave" of new silicon expected and tech giants demanding custom-designed AI server racks for optimal efficiency, Celestica's engineering prowess to build complex, bespoke systems at scale makes it well-positioned to capture significant growth from the multi-year hardware cycle.
**Mailbag Question: Selling Stocks for Grad School:**
The podcast also addressed a listener's question from Ben in Sacramento, who pondered whether to sell a third of his portfolio to cover grad school tuition or take out student loans. Jon Quast shared a personal anecdote about selling an investment to fund a vehicle. Matt Frankel advised that investing is a "means to an end," and paying for planned life expenses, especially to avoid debt, is a valid reason to sell. He framed it as transferring capital from one productive asset (stocks) to another (one's own earnings power through education), which could potentially yield a better return. He cautioned listeners to consider capital gains taxes and the interest rates of available student loans. Rachel Warren added that while it feels counterintuitive for long-term investors to sell, a portfolio should be viewed as a "tool belt" for building life and achieving financial freedom. She concluded that investing in one's education is a "bet on an asset where you dictate the success," empowering the individual.