此次讨论由Telus Demos、金融分析师Kevin Koharkey以及《华尔街日报》的Jonathan Wild共同参与，深入探讨了微软、Meta和Alphabet等主要AI超大规模公司在财务报告中，尤其是在自由现金流（FCF）方面，存在的误导性问题。核心论点是，尽管这些公司预测的盈利和FCF数据令人印象深刻，但仔细审视股权激励（SBC）和相关股票回购的会计处理，就会发现其真实的现金生成能力截然不同，而且往往远低于预期。 Kevin Koharkey解释说，传统的FCF计算方法（通常是从经营现金流中减去资本支出（CapEx））未能充分考虑股权激励（SBC）的现金成本。这一成本主要来自两个方面：一是员工的税款代扣，二是（更关键的是）公司为抵消员工行使股票期权或限制性股票单位（RSU）造成的股权稀释而进行的大规模股票回购。尽管股权激励（SBC）在利润表中会随着时间推移被费用化，但与这些活动相关的实际现金流出往往并未在现金流量表的经营现金流部分得到充分体现。 Jonathan Wild引用了他关于这一主题的研究，强调Meta是一个典型的例子。对于2025年，Meta预测的460亿美元FCF在调整了股权激励（SBC）的现金成本（主要是为防止稀释而进行的股票回购）以及其他因素（如融资租赁）后，将大幅缩减至仅50亿美元——削减了89%。同样，Alphabet 2025年的FCF从730亿美元降至240亿美元，减少了67%。这些差异并非仅仅是理论上的；它们从根本上改变了公司财务健康状况及其通过自身运营为未来增长提供资金的能力的图景。 这种调整后FCF的影响是深远的。首先，它直接影响估值。依赖报告的FCF数据的投资者可能严重高估了这些公司的价值。正如Kevin所说，如果“买下整个公司”，实际产生并可供所有者使用的现金将远低于普遍认知。 其次，大规模的AI数据中心建设需要巨额资本支出（CapEx）。如果一家公司的真实FCF不足以覆盖这些投资，它就必须越来越多地依赖其资产负债表——通过举债或发行新股。像Meta这样历史上几乎没有债务的公司，现在也正在承担巨额负债。例如，Alphabet在一个季度内借款300亿美元并发行了850亿美元的股票。这种对外部融资的依赖，而非依靠强劲的经营现金流进行自我供血，标志着商业模式从“轻资产”向“重资产”的转变，可能会增加财务风险。 演讲者还批评了华尔街常用的EBITDA（息税折旧摊销前利润）等指标，因为它们通常会将股权激励（SBC）加回，进一步模糊了真实的现金成本，并可能导致净杠杆率被低估。这可能会影响公司的信用评级和资本成本。 此外，人们担心未来几年市场对FCF的预测过于乐观，显示出一种“不太可能”的V型复苏，即资本支出（CapEx）成本被假设会急剧减少。AI基础设施的长期成本——包括环境影响、设备的快速淘汰以及公用事业需求的增加——尚未被充分理解或纳入当前的财务模型中，这可能导致比预期更快的资本支出（CapEx）周期和更高的再投资需求。 讨论总结道，当前对AI公司的热情可能是在牛市中由“希望和梦想”所驱动，而非坚实的财务基本面。尽管像英伟达（NVIDIA）这样的公司正从销售AI硬件中获得即时收益，但它们的客户可以推迟相关费用的支出（例如，折旧只在数据中心投入运营后才开始），这创造了一个暂时的“黄金时刻”，即报告的盈利看起来强劲，但实际现金流出却滞后。这种差异造成了一种盈利和现金生成的“视觉假象”，对于那些不仅仅看表面数字的投资者来说，这值得高度关注。

The provided discussion, featuring Telus Demos, financial analyst Kevin Koharkey, and WSJ's Jonathan Wild, delves into the often-misleading financial reporting of major AI hyperscaler companies like Microsoft, Meta, and Alphabet, particularly concerning their free cash flow (FCF). The core argument is that while these companies project impressive earnings and FCF, a closer look at the accounting for stock-based compensation (SBC) and associated stock buybacks reveals a significantly different, and often much lower, true cash generation. Kevin Koharkey explains that traditional FCF calculations, which typically subtract capital expenditures (CapEx) from operating cash flow, fail to account for the cash cost of SBC. This cost arises from two main areas: tax withholding for employees and, crucially, massive stock buybacks undertaken by companies to offset the dilution caused by employees exercising stock options or restricted stock units (RSUs). While SBC is expensed on the income statement over time, the actual cash outflow related to these activities often isn't fully reflected in the operating cash flow section of the cash flow statement. Jonathan Wild, referencing his work on the subject, highlights Meta as a prime example. For 2025, Meta's projected FCF of $46 billion dramatically shrinks to just $5 billion once adjustments for the cash cost of SBC (primarily buybacks to prevent dilution) and other factors like finance leases are made—an 89% reduction. Similarly, Alphabet's 2025 FCF drops from $73 billion to $24 billion, a 67% reduction. These discrepancies are not merely academic; they fundamentally alter the picture of a company's financial health and its ability to fund future growth from its own operations. The implications of this adjusted FCF are profound. Firstly, it directly impacts valuation. Investors who rely on reported FCF numbers may be significantly overvaluing these companies. If one were to "buy the entire company," as Kevin puts it, the actual cash generated and available to the owner would be far less than commonly perceived. Secondly, the massive AI data center build-outs require enormous CapEx. If a company's true FCF is insufficient to cover these investments, it must increasingly rely on its balance sheet—taking on debt or issuing new shares. Companies like Meta, which historically operated with little to no debt, are now incurring substantial liabilities. Alphabet, for instance, borrowed $30 billion and issued $85 billion in shares in a quarter. This reliance on external financing, rather than self-funding from robust operating cash flow, signals a shift from "asset-light" to "asset-heavy" business models, potentially increasing financial risk. The speakers also criticize common Wall Street metrics like EBITDA (Earnings Before Interest, Taxes, Depreciation, and Amortization) for often adding back SBC, further obscuring the true cash costs and potentially leading to an understated net leverage ratio. This can impact a company's credit rating and cost of capital. Furthermore, there's concern that market forecasts for FCF in the coming years are overly optimistic, showing an "improbable" V-shaped recovery where CapEx costs are assumed to taper off sharply. The long-term costs of AI infrastructure—including environmental impacts, rapid obsolescence of equipment, and increased utility demands—are not yet fully understood or factored into current financial models, potentially leading to quicker CapEx cycles and higher reinvestment needs than anticipated. The discussion concludes by suggesting that the current enthusiasm for AI companies might be driven by "hopes and dreams" in a bull market, rather than solid financial fundamentals. While companies like NVIDIA are seeing immediate benefits from selling AI hardware, their customers can defer associated expenses (e.g., depreciation only starts when data centers become operational), creating a temporary "golden moment" where reported earnings appear robust but actual cash outflows are lagging. This disparity creates an "optical illusion" of profitability and cash generation that warrants significant concern for investors who look beyond headline numbers.