Alpha decays. This is not a provocative claim; it is arithmetic. Every dollar of outperformance one investor earns must come from somewhere, and in aggregate, active managers as a group cannot beat the market they collectively constitute. After fees, the expected value of hunting alpha is negative for most participants. Sharpe made this point in 1991 and the data since then have only reinforced it. Fama and French, reviewing mutual fund returns across decades, found that the distribution of alpha is consistent with what you would expect from luck alone. A handful of managers outperform persistently, but identifying them in advance, rather than in hindsight, is its own zero-sum game.

None of this means alpha does not exist. It does, in pockets, for those with genuine informational or analytical edges. The question is not whether alpha exists but whether the expected value of pursuing it exceeds the expected value of the alternative. For most investors, especially those who are not doing this full-time, I think the answer is no.

The alternative is beta, and beta deserves more respect than it typically gets. If an industry is in structural growth, the sector as a whole tends to appreciate. You do not need to identify the winner within the sector. You only need to be in the sector. The return from being exposed to the right thematic direction is real, repeatable, and does not require you to be smarter than the collective intelligence of the market about any individual company.

Thematic investing, at its core, is the discipline of identifying good sources of beta. The AI supply chain is the clearest current example. You do not need a non-consensus view on NVIDIA's margin structure or Taiwan Semiconductor's next-node yield rates to participate in that return stream. You need a view on whether AI infrastructure buildout is a durable, multi-year trend. That is a structurally easier question to answer with conviction, because it depends on macro forces rather than firm-specific execution.

The operational approach follows from this. The first step is theme validation: is this sector in genuine structural growth, or is it a narrative without underlying demand? Once that question is answered affirmatively, the implementation is a basket of names with high thematic exposure. You are not trying to call which company wins. You are trying to ensure you are present for the sectoral appreciation. Position sizing within the basket matters less than theme selection. Risk management also becomes cleaner: if you lose conviction on the theme, you reduce the entire position, rather than making individual judgments about dozens of companies.

History offers several clear illustrations. In the 2010s, mobile internet was the dominant thematic beta. You did not need to predict in 2010 whether Apple or Samsung would capture more smartphone market share. You needed to believe that smartphones would penetrate mass markets globally and that the supply chain enabling that penetration would grow. A basket of display manufacturers, chip designers, baseband modem suppliers, and component assemblers would have captured the beta from that structural shift, even if individual company bets within that basket were wrong. The same logic applied to electric vehicles in the early 2020s: the relevant thematic question was whether EV adoption would accelerate, not which OEM would dominate. Lithium producers, battery component suppliers, and charging infrastructure companies all participated in that beta.

AI infrastructure today sits in the same category. The relevant analytical question is not whether NVIDIA's gross margins will compress in fiscal 2027. It is whether the structural demand for AI compute, storage, and interconnect will grow over the next several years. That is a question about industrial necessity, not about competitive dynamics at the firm level. Given the pace of AI model deployment, the expansion of inference use cases, and the enterprise capex commitments already announced, the directional answer seems clear enough to act on.

There is a deeper point here about where investors should direct their cognitive effort. Alpha generation requires you to have a view that is both correct and different from the market consensus. That is a high bar. It requires proprietary information, analytical superiority, or a willingness to be wrong in ways that are uncomfortable and career-threatening. Beta generation requires you to have a view on direction, which is a lower bar, and to accept market-level returns within the sector you have chosen. The expected value of the second strategy is higher for most people, not because it is easier in an intellectual sense, but because the base rates are better.

The goal of investing is not to prove that you are smarter than the market. It is to be exposed to the right structural trends. Getting that exposure right is a sufficient condition for building wealth over time. Most investors would do better if they spent more of their analytical energy on the first-order question, which is thematic direction, and less of it on the second-order question, which is stock selection within a theme they have already identified as attractive.

Alpha 会衰减。这不是什么惊人的论断，只是算术。每一个投资者多赚的一块钱，必然从某个地方来。从整体上看，主动管理者作为一个群体，不可能跑赢他们共同构成的市场。扣除费用后，追逐 alpha 的期望值对大多数参与者来说是负的。Sharpe 在 1991 年就指出了这一点，此后的数据只是不断印证它。Fama 和 French 梳理了几十年的共同基金回报，发现 alpha 的分布和纯靠运气的预期几乎完全一致。确实有少数基金经理持续跑赢，但在事前而非事后识别他们，本身又是一场零和博弈。

这并不是说 alpha 不存在。它确实存在，在某些角落里，属于那些真正拥有信息或分析优势的人。问题不是 alpha 存不存在，而是追逐 alpha 的期望值是否超过了其他策略的期望值。对大多数投资者来说，尤其是那些不把投资作为全职工作的人，我认为答案是否定的。

另一个选择是 beta，而 beta 通常没有得到它应有的重视。如果一个行业处于结构性增长中，整个板块往往会上涨。你不需要找出板块内的赢家，只需要身处这个板块。顺着正确主题方向获得的回报是真实的、可重复的，而且不需要你在任何单一公司上比市场的集体智慧更聪明。

主题投资的核心，就是识别好的 beta 来源。AI 供应链是当下最清晰的例子。你不需要对英伟达的利润率结构或台积电的先进制程良率有什么非共识的判断，就能参与那条回报主线。你需要的只是一个判断：AI 基础设施的建设是否是一个持久的、多年期的趋势。这个问题在结构上更容易给出有把握的答案，因为它依赖的是宏观力量，而不是某家公司的具体执行能力。

操作思路由此而来。第一步是验证主题：这个赛道是否处于真实的结构性增长，还是只有叙事没有底层需求？一旦这个问题得到肯定回答，实施方式就是一篮子具有高主题敞口的股票。你不是在押哪家公司赢，而是确保自己参与整个板块的升值。篮子内各仓位的大小，重要性远不如主题的选择本身。风控也因此更清晰：如果你对主题失去信心，整体减仓，而不是对几十只股票逐一做判断。

历史提供了几个清晰的案例。2010 年代，移动互联网是最主要的主题 beta。你不需要在 2010 年预测苹果还是三星会赢得更多智能手机市场份额，你只需要相信智能手机会在全球普及，支撑这一普及的供应链会随之增长。一篮子显示屏厂商、芯片设计公司、基带调制解调器供应商和组装厂，即使其中个别公司的押注是错的，也足以捕获那次结构性转变的 beta。同样的逻辑在 2020 年代初适用于电动车：相关的主题问题是电动车渗透率是否会加速，而不是哪家整车厂会称霸。锂矿商、电池材料供应商和充电基础设施公司都参与了那次 beta。

今天的 AI 基础设施属于同一类别。相关的分析问题不是英伟达 2027 财年的毛利率会不会收窄，而是对 AI 算力、存储和互联的结构性需求，在未来几年内是否会持续增长。这是一个关于产业必然性的问题，而非关于公司层面竞争格局的问题。考虑到 AI 模型部署的速度、推理应用场景的扩张，以及企业已经宣布的资本支出承诺，方向性的答案已经清晰到足以行动。

这里还有一个更深层的观点，关于投资者应该把认知精力放在哪里。Alpha 需要你有一个既正确又与市场共识不同的判断，门槛很高。它需要专有信息、分析上的优势，或者一种愿意以令人不舒服、甚至威胁职业生涯的方式犯错的勇气。Beta 需要你对方向有判断，门槛更低，然后在你选定的板块内接受市场水平的回报。第二种策略对大多数人的期望值更高，不是因为它在智识上更容易，而是因为胜率更好。

投资的目标不是证明自己比市场聪明，而是让自己暴露在正确的结构性趋势下。把这个敞口选对，就足以在时间的维度上积累财富。如果大多数投资者把更多的分析精力放在第一序问题（主题方向）上，而不是第二序问题（在他们已经认为有吸引力的主题内选股）上，结果会更好。