Cyrus
Chan

I came across a chart published by the hedge fund Coatue that I found to be very interesting. I love the work they do and would strongly encourage any readers to check out some of their C:\Takes to challenge and widen their perspectives on our current investing landscape.

On the left: $680 billion in projected 2026 capex from the four hyperscalers, Microsoft, Google, Meta, and Amazon. On the right: $525 billion in projected free cash flow flowing to the largest semiconductor companies, NVIDIA, Samsung, SK Hynix, Micron, and TSMC.

Hyperscalers — Performance

AI Beneficiaries — Performance

One thing that stands out this cycle is that the companies spending the most haven't been the ones rewarded the most by the stock market, even though that spending is intended to position them for the next wave of AI-driven productivity and growth. Looking specifically at the four hyperscalers highlighted by Coatue, Alphabet is up roughly 21% YTD, Amazon 14%, Microsoft is down about 4%, and Meta is down nearly 8%. It is important to note that this skepticism, as reflected by the stock prices, isn't entirely unwarranted. It is difficult to imagine a world where the billions in growth capex being spent will ever generate a return that comes close to justifying the scale of investment being made. Meanwhile, the companies supplying the AI buildout have seen parabolic gains. SanDisk is up 639% YTD, SK Hynix 264%, Micron 263%, and Samsung 192%. Semiconductor and memory names have experienced a once-in-a-lifetime run.

To me, this largely sounds like a capital rotation story. Early in a new infrastructure cycle, investors tend to gravitate toward the companies seeing the most immediate benefit from the spending, which in this case, has been the semiconductor and memory companies, where the connection between AI investment and financial results is easy to see. On the other hand, the hyperscalers are making investments in areas with results that are difficult to quantify in the near-term. Until those returns become more visible, it's not surprising that investors have preferred the former over the latter.

Before discussing where I see opportunity, it's worth noting that semis and memory have historically been cyclical industries. Before the AI boom, anyone who spent meaningful time following the space understood the pattern well: demand surges, supply responds, inventories build, pricing corrects, and margins compress. It would be naive to assume those dynamics have been eliminated simply because this cycle is larger and more transformative than ever before. With that being said, I don't believe supply is remotely close to catching demand. Buildout required to support AI workloads is immense, and estimates suggest $5.2 trillion of cumulative growth capex till 2030, and forecasts tend to lean too conservative. However, history suggests that supply eventually catches up. Whether that occurs three, five, or ten years from now is debatable.

This scenario poses the question. If AI investments ultimately generate the productivity gains and economic returns, then some of the hyperscalers may be more attractively valued than the market currently appreciates. There are two names that I believe are especially well-positioned to benefit if that thesis plays out.

First is Meta. I believe that Meta is one of the most well-positioned to leverage this AI buildout because their core product, their family of apps, are direct beneficiaries of improved AI, improving recommendations, engagement, and ad targeting across Facebook, Instagram, Reels, and WhatsApp, which in turn drives stronger CPAs and ROAs for their advertising business. In addition to this, they are perhaps priced the most reasonably out of their peers, with an NTM P/E of 18.3x, EV/Sales of 5.8x, and EV/EBITDA of 10.1x. Consensus estimates place revenue growth of ~25% in FY2026, moderating to ~17% by FY2028. Assuming those estimates materialize, Meta would be trading at just ~14.5x FY2028 P/E and ~4.3x FY2028 EV/Sales at today's share price.

Second is Amazon. Anthropic runs almost entirely on AWS rather than operating its own data centers, the way OpenAI has through its Stargate partnership with Oracle. With Anthropic arguably emerging as the leading frontier AI lab, Amazon benefits from having a committed anchor tenant embedded within its infrastructure. As a result, a significant portion of the capacity Amazon is building already has identifiable demand behind it. Beyond Anthropic, AWS is also the default infrastructure layer for a large share of enterprise AI development, positioning itself to benefit from the growing adoption of AI across the broader economy. The infrastructure story isn't the only part of why I find Amazon interesting. Amazon is the highest revenue company among the hyperscalers, and the bulk of that revenue runs through a retail and logistics operation that has been the most structurally constrained part of the business. However, robotics and AI-driven automation across Amazon's fulfillment network represent a path to improving margins in a segment that has historically been dragging on the overall business. If that plays out, Amazon becomes a company that is simultaneously a builder of AI infrastructure as a direct beneficiary of external AI demand through AWS, and an internal beneficiary of AI deployment across its own enterprise. That unique combination is difficult to find anywhere else in the market.

Taken together, these two names represent what I think is the clearest expression of the rotation thesis. With Micron recently hitting a 1.0 trillion dollar market cap, it is really hard not to position heavily within the most direct beneficiaries of AI spend, which is understandable, and for the past two years, it has been the right call. But I find the more interesting question to be where the returns ultimately accrue once the infrastructure is built. The way I think about it is owning both sides of the AI value chain. On one side are the companies capturing the spending today, the semiconductor, memory, and infrastructure providers with the greatest exposure to where capital is flowing. On the other hand, the hyperscalers are most likely to benefit as the returns on that investment become increasingly visible. The goal is not to predict exactly when the market's focus shifts from builders to beneficiaries, but to maintain exposure to both. If the AI buildout continues, the infrastructure layer should continue to win. If the productivity gains ultimately materialize, the capital may eventually rotate back toward the companies best positioned to monetize them.

As of May 19th, 2026, Google announced in its Google I/O keynote, its transition away from the traditional search interface that had defined the internet for decades, signaling a broader shift from static keyword-based search toward a Gemini-powered, conversational AI-native discovery experience set to begin rolling out on May 26th. I view this as a risk with monitoring closely. Google's push toward an AI-native search experience appears strategically designed to keep users increasingly contained within its own ecosystem, reducing the need to navigate to external websites while simultaneously strengthening user engagement and reinforcing the long-term value of its advertising platform.

While the ultimate impact of this transition remains uncertain, there are several potential implications that warrant close attention. Most notably, the shift toward an AI-native search experience could reduce click-through rates to Reddit's platform. Although Reddit continues to rank among the most frequently cited sources within LLM-generated responses, increasingly comprehensive AI summaries may disincentivize less-engaged or non-core users from clicking through to the underlying discussion threads themselves.

This dynamic isn't entirely new. We already began to see early signs of this behavior emerge with the rollout of Google AI Overview, where users were increasingly able to extract summarized information directly from the search interface without visiting external websites. However, the distinction now is that Google appears to be repositioning AI-generated responses from a supplemental feature into the primary search experience itself. Previously, AI Overviews largely functioned as a complement to traditional search. However, under the new AI-native system, conversational summaries increasingly become primary, while external links and citations become secondary.

My primary concern is relatively straightforward. Reddit's business model remains heavily dependent on advertising economics, and thus any reduction in traffic flowing onto the platform could ultimately become a headwind to user growth, especially given that 60% of Reddit's traffic comes from organic search. If AI-native search experiences increasingly satisfy informational intent directly within the search interface, less-engaged users may become materially less likely to click through to Reddit discussions themselves. Over time, this could make it more difficult for Reddit to attract and retain more casual or infrequent users, particularly those who primarily access the platform through search-driven discovery rather than intentional engagement. While I expect DAUs to remain relatively sticky, the greater risk likely lies with MAUs, the more occasional visitors whose engagement patterns are less durable.

From a monetization perspective, this matters because Reddit's long-term topline growth depends on both improving monetization efficiency and continued expansion of its user base. Even if monetization per user continues to improve, slower user growth or declining traffic would still create a meaningful offset on an absolute revenue basis. Put simply, users who never arrive on the platform cannot be monetized. We see this already being realized today with the implementation of Google's AI overviews, which represent a structural threat to web-based platforms. Zero-click searches, where users obtain answers directly from Google's AI-generated summaries without visiting any external site, rose from 56% to 69% of all Google searches between May 2024 and May 2025, meaning nearly seven in ten searches now conclude before reaching any destination site. Looking ahead, the Reuters Institute's January 2026 Digital News Report found that publishers expect search-driven traffic to decline by 43% on average over the next three years.

While the picture above may seem bleak, there are redeeming factors that deserve equal consideration. The same AI Overview ecosystem driving zero-click searches across the broader internet may, in fact, represent a structural tailwind for Reddit specifically.

Unlike traditional publishers whose content isn't surfaced in AI overviews, Reddit's unique position as the most cited source across AI platforms, appearing in 49.4% of AI Overviews and cited in 40.1% of responses across ChatGPT, Perplexity, Gemini and Claude. Critically for Reddit, this citation and visibility dominance in AI-generated results translates into measurable traffic benefits. Research shows that sites cited within AI Overviews earn approximately 120% more organic clicks per impression versus uncited competitors on the same query, and the traffic that does arrive is demonstrably higher quality, converting at 14.2% versus traditional organic search's 2.8%, a near 5x premium. This is already showing up in Reddit's own traffic data, with the platform growing to 1.4 billion monthly visits by April 2025 against a broader backdrop of publisher traffic decline. Taken together, the structural shift toward AI-mediated search does not appear to be compressing Reddit's funnel, but rather it may in fact be concentrating discovery traffic toward platforms with authentic, experience-based human-generated content, of which Reddit remains unrivaled.

This perspective is not without conditionality, however. It assumes Google's retrieval algorithms continue to prioritize Reddit's user-generated content at current rates, a dynamic that cannot be taken for granted. That said, we believe Reddit's content licensing agreements with Google and OpenAI structurally underpin this favorability, effectively anchoring its position within AI retrieval architecture.

While the full materiality of this shift will likely not be apparent until the latter half of 2026, it remains a development worth monitoring closely. Regardless, my conviction remains anchored in Reddit's strong underlying fundamentals and its demonstrated ability to not only navigate but benefit from the structural changes AI-mediated search has introduced.

On the surface, Meta's launch of Forum, a standalone Reddit-like application built on top of Facebook Groups, which was quietly released on May 22nd 2026, may appear to be an unambiguously bearish development for Reddit. A competitor of Meta's scale entering the community discussion space directly compresses Reddit's addressable market and introduces a well-capitalized adversary with 3 billion+ users already on its platforms. The knee-jerk reaction from markets reflected exactly this, with it trading down 5.6% on the day.

However, a more considered reading of this development yields a more nuanced conclusion, and perhaps even a counterintuitively bullish one. When a company of Meta's size, with virtually unlimited resources and optionality across its product portfolio, chooses to build a Reddit clone, it inadvertently validates the entire premise of Reddit's business model. It is an acknowledgment that community-driven, interest-based, pseudonymous discussion is not a niche use case but a durable and valuable category of internet behavior worth competing for. That said, validation alone does not insulate Reddit from competitive pressures, and the more substantive question is whether Forum can actually replicate what makes Reddit structurally valuable, particularly its search discovery flywheel. Here, we believe the competitive threat is considerably more limited than the market reaction suggests.

Reddit's dominance in search discovery is not simply a function of having community discussion content, but rather a function of that content being entirely open, public, and crawlable by default. Every post, comment, and thread on Reddit is immediately Google-indexable without an account, without a login wall, and without friction. This is what has made Reddit the most cited source across AI platforms. It is what has driven Reddit's organic search traffic up 10x and allowed Google to formalize that relationship through a $60 million annual content licensing agreement. Thus, the open architecture of Reddit serves to be its most foundational principle.

Meta's content ecosystem is structurally the opposite. Across its entire suite of applications, Facebook, Instagram, WhatsApp, Messenger, and Threads, the default has always leaned toward private, closed, and walled off from meaningful external discovery. Private Facebook Groups, which constitute much of the group activity on the platform, have never been crawlable by Google. WhatsApp and Messenger are entirely closed systems with zero indexability. Instagram occupies a middle ground. While public professional accounts became technically indexable by Google following a July 2025 update, the experience is largely superficial. A user can find an Instagram post in search results, but clicking through hits a login wall before any content can actually be consumed. Discoverability only exists at the surface level, but the valued content itself remains gated, which is a different proposition from Reddit, where the full thread, every reply, and every answer is readable by anyone without an account.

While public Facebook pages and a small subset of public groups are technically indexable, these represent a fraction of Meta's total content universe. Forum, built on top of this same closed infrastructure, inherits these same constraints. A user cannot append "Forum" to a Google search the way millions append "Reddit" daily, because the content is not publicly available. The internal "Ask" feature Meta has built into Forum searches only within a user's own groups, explicitly not from the broader web, which further underscores how closed and self-contained the product is by design. At the end of the day, I ask myself this question. When someone has a question they want answered, where do they actually go first? I would almost certainly say not Forum. As long as Meta keeps its walls up, capturing meaningful query intent at scale will remain structurally out of reach, especially when compared to the likes of Reddit.

This distinction matters enormously for the long-term competitive dynamic. Reddit's search flywheel is self-reinforcing. Content gets indexed, gets cited in AI Overviews, drives organic traffic, brings in new users who generate more content, which gets indexed again. Forum cannot replicate this loop without fundamentally restructuring Meta's privacy architecture, which would conflict directly with years of ingrained user expectations around data privacy from their suite of applications. Thus, this appears to be a structural ceiling that limits Forum's ability to compete for the specific type of search-driven user acquisition that underpins Reddit's growth thesis.

Upon first impressions, I view this as noise more than a bearish signal. Nothing official has been announced yet on how Forum will actually come to market, but a cursory look at what is known does not give us much reason to worry. Reddit and Meta may share surface-level similarities in architecture and experience, but they are not competing in the same verticals or fishing from the same pool of core users. Reddit captures intent-driven, anonymous discovery. Forum captures intentional conversation within existing social graphs.