Is the Data Center Investing Trend in Trouble?

2026年9月4日，由乔恩·夸斯主持，Fool撰稿人马特·弗兰克尔和卢·怀特曼共同参与的Motley Fool Hidden Gems投资播客，探讨了两个重要话题：Snowflake的近期表现以及数据中心投资的演变趋势。 讨论首先从Snowflake开始，该公司股价飙升超过20%，达到52周新高，并接近历史最高点。马特·弗兰克尔强调，Snowflake轻松超出了预期，营收增长达到37%（预期为35%），并连续第五个季度盈利超出预期。该公司还发布了大幅上调的业绩指引，并拥有126%的净收入留存率，这表明客户正在增加支出。卢·怀特曼指出该股“奇怪”的历史，包括伯克希尔哈撒韦公司的早期投资和随后的出售，以及其股价在多年表现不佳后才勉强回到2021年的高点。尽管过去超出预期和上调指引并未总是显著推动股价，但这次情况有所不同。 马特将近期股价飙升归因于增长的加速，过去三个季度营收增长从30%提升到34%再到37%，同时利润率也有所改善。调整后的营业利润率达到15%以上，高于一年前的11%。卢解释说，AI是这一加速增长背后的驱动力，因为AI模型依赖数据而蓬勃发展，而数据正是Snowflake的核心业务。Snowflake的按使用量付费模式（客户只为他们使用的部分付费）现在极具优势，因为AI工作负载极大地增加了数据消耗。管理层证实，AI工作负载约占增长加速的一半。 然而，主持人也讨论了Snowflake的潜在担忧。估值仍然很高，远期销售额的市销率约为20倍，自由现金流的市现率约为80倍。股权激励是另一个问题，占营收的近30%，尽管有回购，仍导致超过4%的股权稀释。卢还提到，预计毛利率将下降100个基点，尽管目前毛利率仍高达74%。一个更大的问题是，当前的AI消耗是否“不理性”，以及如果转向“AI效率”，最终是否会抑制对Snowflake按使用量付费模式的需求。 第二个主要话题探讨了数据中心投资趋势，询问其是否陷入困境。乔恩·夸斯指出，纽约成为第一个对新建数据中心（特指50兆瓦及以上规模的）实施暂停的州，德克萨斯州也效仿暂停了审批。这种来自两党政治派别的反对，源于对电力消耗和社区影响的担忧。马特·弗兰克尔透露，70%的美国人不希望数据中心建在自己家附近，上半年有833个反对团体成功阻止或推迟了他们目标的三分之二的数据中心项目。一个主要担忧是不断上涨的电费，过去一年平均上涨了5%，而数据中心是重要的助推因素。伯克希尔哈撒韦公司首席执行官格雷格·阿贝尔甚至建议超大规模企业应承担这些电力成本。 卢·怀特曼认为这些暂停是暂时的，源于大型科技公司与小型市政当局谈判时存在的不平衡权力动态。他建议，州级谈判可以为社区提供更大的透明度和影响力，最终解决目前的“暂停”状态。马特·弗兰克尔承认了合理的积极方面，例如创造就业机会（如Meta在路易斯安那州的项目带来了7500个建筑工作岗位和1000个永久性工作岗位），通过保持技术领先来维护国家安全，以及为当地带来可观的房产税收入。 尽管有暂时的暂停，许多AI基础设施股票仍出现显著下跌（Marvell下跌超过30%，Celestica下跌近40%，Sterling Infrastructure从高点下跌超过50%）。卢认为，尽管数据中心建设将复苏，但这些“卖铲人”供应商的“爆炸性增长”可能已成为过去，市场将转变为一个“高水平运营环境”。马特表示同意，并列举了制造能力、员工可用性、电力基础设施、芯片短缺和资金等各种限制，暗示爆炸性增长的可能性较小，尽管有些股票仍可能超过此前的历史最高点。 最后，播客回答了听众本的一个问题，本在一家小型AI基础设施公司持有大量未实现亏损，但他仍然相信该公司到2030年有10-15倍的潜力。本想知道如何权衡这一点，与一家更安全、更成熟的公司（同期有2-3倍的潜力）相比。卢建议避免“沉没成本谬误”，敦促投资者根据当前的判断和风险承受能力做出决定。他建议采取平衡的方法，或许将60%分配给成熟公司，其余分配给更具投机性的公司。马特同意本通过了“沉没成本”测试，因为本确认他会以今天的价格追加投资。他还强调，对“更安全”的股票而言，到2030年实现2-3倍的回报（年化17-29%）仍然会是强劲的跑赢市场回报，他强调，当跑赢市场的回报仍然可能时，“慢而稳”不一定意味着“无聊”。

On September 4, 2026, the Motley Fool Hidden Gems Investing podcast, hosted by Jon Quas, joined by Fool contributors Matt Frankel and Lou Whiteman, addressed two significant topics: the recent performance of Snowflake and the evolving trend of data center investments. The discussion began with Snowflake, which saw its stock jump over 20% to 52-week highs, approaching all-time highs. Matt Frankel highlighted that Snowflake beat expectations handily, with revenue growth at 37% (compared to 35% expected) and a fifth consecutive quarter of beating the bottom line. The company also issued a massive guidance raise and boasted a net revenue retention rate of 126%, indicating customers are increasing their spending. Lou Whiteman noted the stock's "weird" history, including Berkshire Hathaway's early investment and subsequent sale, and its performance barely returning to 2021 highs after years of underperformance. Despite past beats and raises not always moving the stock significantly, this time was different. Matt attributed the recent surge to an acceleration in growth, with top-line growth moving from 30% to 34% to 37% over the last three quarters, alongside improved margins. The adjusted operating margin reached over 15%, up from 11% a year ago. Lou explained that AI is the driving force behind this acceleration, as AI models thrive on data, which is Snowflake's core purpose. Snowflake's consumption-based model, where clients pay only for what they use, is now highly beneficial as AI workloads dramatically increase data consumption. Management confirmed that AI workloads accounted for roughly half of the growth acceleration. However, the hosts also discussed potential concerns for Snowflake. Valuation remains high, trading at around 20 times forward sales and 80 times free cash flow. Stock-based compensation is another issue, making up almost 30% of revenue, leading to over 4% stock dilution despite buybacks. Lou also mentioned a forecasted 100-basis-point gross margin depression, although margins remain high at 74%. A larger question raised was whether the current AI consumption is "irrational" and if a shift towards "AI efficiency" could eventually temper demand for Snowflake's consumption-based model. The second main topic addressed the data center investing trend, asking if it was in trouble. Jon Quas pointed out that New York became the first state to implement a moratorium on new data centers (specifically those 50 megawatts and larger), with Texas following suit by pausing approvals. This pushback, across both political spectra, stems from concerns over power consumption and community impact. Matt Frankel revealed that 70% of Americans don't want data centers built near their homes, and 833 opposition groups successfully blocked or delayed two out of every three data center projects they targeted in the first half of the year. A major concern is rising power bills, which have increased 5% on average over the past year, with data centers being a significant contributor. Greg Abel, Berkshire Hathaway's CEO, even suggested hyperscalers should absorb these power costs. Lou Whiteman believes these moratoriums are temporary, resulting from an uneven power dynamic where large tech companies negotiate with small municipalities. He suggested that state-level negotiations could provide more transparency and leverage for communities, eventually resolving the current "pause." Matt Frankel acknowledged legitimate positives, such as job creation (e.g., Meta's Louisiana project bringing 7,500 construction and 1,000 permanent jobs), national security through maintaining a tech lead, and significant property tax revenue for local areas. Despite the temporary pause, many AI infrastructure stocks have seen significant declines (Marvell down over 30%, Celestica nearly 40%, Sterling Infrastructure over 50% from its high). Lou believes that while data center construction will recover, the "blockbuster gains" for these "picks and shovels" suppliers might be largely behind us, shifting to an "elevated operating environment." Matt agreed, citing various constraints like manufacturing capacity, employee availability, power infrastructure, chip shortages, and capital, suggesting explosive gains are less likely, though some stocks might still surpass previous all-time highs. Finally, the podcast addressed a listener question from Ben, who held a substantial unrealized loss in a small cap AI infrastructure company but still believed in its 10-15x potential by 2030. Ben wondered how to weigh this against a safer, more established company with 2-3x potential over the same period. Lou advised against the "sunk cost fallacy," urging investors to make decisions based on current judgment and risk tolerance. He suggested a balanced approach, perhaps allocating 60% to the established company and the rest to the more speculative one. Matt agreed that Ben passed the "sunk cost" test by confirming he'd invest more at today's price. He also highlighted that a 2-3x return by 2030 (17-29% annualized) for the "safer" stock would still be a strong market-beating return, emphasizing that "slow and steady" doesn't necessarily mean "boring" when market-beating returns are still possible.