The current market is being propelled by a single, dominant force: a spending cycle built around artificial intelligence infrastructure. Capital is flowing out of the biggest spenders—the hyperscalers—and into the more constrained, bottleneck segments of the AI trade. While this dynamic has produced spectacular returns, it has also created a market whose performance rests on an unusually narrow base. Understanding the implications of that narrowness, and what it means for how investors protect themselves, is one of the most important questions facing portfolios today.
A Single Return Driver
The most immediate concern with this environment is that the AI spending trade has become so dominant that buying a broad index no longer delivers what investors assume they are buying. Rather than gaining exposure to a diversified set of economic engines, an index investor today is effectively betting on one return driver. This runs directly against the most basic principle of prudent investing: you are not supposed to put all your eggs in one basket. Yet the structure of the market is quietly forcing investors to do exactly that, because AI spending now drives the majority of both earnings and returns.
This creates a genuine dilemma for active managers. Their craft is built on assembling exposure to different and uncorrelated market drivers—the very definition of diversification. In a market dominated by one theme, that discipline has become a headwind, causing diversified strategies to lag. But it would be a mistake to penalize managers for being prudent. The problem is not their approach; it is a market that has temporarily rewarded concentration over balance.
Diversification That Isn't
What makes the situation especially tricky is that the surface-level data can give a false sense of breadth. Look at year-to-date returns and you might conclude that diversification is alive and well: value stocks are outperforming, small- and mid-cap names are outperforming, and emerging markets have been the top-performing segment of all. On its face, that looks like a healthy spread of winners.
The reality dissolves under scrutiny. Take the extreme case of emerging markets. Strip out just the top three names—TSMC, Samsung, and SK Hynix—and all of that outperformance disappears. The same pattern holds for the broad U.S. market. Analysts compiling lists of S&P 500 companies with the greatest exposure to AI spending arrive at perhaps 40 to 45 names. Those companies make up roughly half of the index by weight, but they account for far more than half of its performance and its earnings growth. They are, in short, punching well above their weight. The apparent diversification across regions and styles is largely an illusion created by the same underlying AI driver showing up in different places.
None of this means the gains are unjustified. The spending is real, there is visibility into it, and companies feeding the data center supply chain are sitting on backlogs that stretch out a couple of years. The strength has a foundation. But investors need to dig into the numbers to grasp how truly narrow the returns are—and how much of their own portfolio is exposed to that narrow base, often without realizing it.
The Hourglass and the Coming Rotation
A useful way to picture the market's evolution is as an hourglass. At the top, the market first rewarded the hyperscalers. That capital is now filtering down through the narrow neck of the hourglass into the bottleneck companies—the AI infrastructure plays addressing supply constraints. The extreme moves seen off recent earnings, particularly in memory, illustrate the intensity here, with analysts issuing dramatic price-target boosts as they scramble to catch up with the trade.
The history of the internet offers an imperfect but instructive parallel. In its early phase, the market exploded in the companies building out the internet's physical capacity—those laying dark fiber and the optical companies that supplied them. Over time, though, the market handed off from that capacity play to the actual beneficiaries: the businesses that used the internet. Those users ultimately proved to be the more attractive investments, the ones with the longer legs.
The same handoff seems likely with AI. Right now, the capacity build is what works. But eventually the market should rotate from the builders to the users of AI—and that is where the more durable, longer-lasting opportunity may lie. The difficult part is timing. The shift is probably coming within the next year or two. For a trader that feels distant, but for a genuine long-term investor it is a remarkably short window. That compressed timeline is precisely why it pays to ensure you hold diversified exposure and that you truly understand how much of your portfolio is concentrated in the narrow part of the hourglass before the rotation arrives.
Internet, Fiber, and the Question of Payoff
There is a second comparison worth weighing carefully: the fiber build-out of the late 1990s. That episode did not unfold as expected. It eventually proved a net positive, but it was not the outcome that had been priced in at the peak. The echo today is unmistakable—hundreds of billions of dollars are being poured into AI infrastructure without clear visibility into the returns. The honest assessment is that, at this point, the evidence that this spending will pay off remains limited.
Yet there are meaningful structural differences between AI and the internet that complicate any direct analogy. The internet was essentially open-source; nobody owns it. AI, by contrast, is a closed system. The model builders own their models, and that ownership fundamentally changes the economics and competitive dynamics of the technology.
The supply-and-demand picture also differs sharply from the fiber era. With fiber, the industry reached a point of sufficiency and then overbuilt dramatically—supply overwhelmed demand. With AI, we are nowhere near that point today; demand still far exceeds supply. We could reach balance two years from now, but two forces cut against a repeat of the fiber glut. On one side, real-world constraints—above all, access to power—may slow the pace of capacity build. On the other, the proliferation of AI use, combined with models that keep growing more intelligent, could allow demand to continue outstripping supply for a period more extended than anything seen during the fiber build-out.
What Investors Should Take Away
The throughline of all this is not pessimism about AI but realism about concentration. The spending cycle is genuine, the backlogs are real, and the technology may well sustain demand longer than past infrastructure booms. But the returns driving the market are far narrower than they appear, the apparent diversification across styles and regions is hollow, and the eventual rotation from builders to users—while uncertain in timing—is plausibly only a year or two away. The prudent response is not to abandon the trade but to understand exactly how exposed you are to its narrow core, and to make sure that exposure is a deliberate choice rather than an accident of owning the index.