Artificial intelligence (AI) and technology moved back into the driver’s seat in the second quarter, with the Information Technology sector up +32%—the only sector to beat the broad S&P 500 Index. But beneath the surface, the market was increasingly binary, rewarding perceived AI beneficiaries while punishing nearly everyone else. Understanding what is driving those returns is becoming just as important as understanding AI itself.
Software stocks had been the market darlings for much of the past few years until late in the third quarter of 2025, when fear mounted that AI would disrupt existing software business models. Through the first six months of 2026, the AI trade had centered on infrastructure buildout, and the market’s winners were largely the companies benefiting directly from this spending—the “check cashers,” as we refer to them. These are the companies that provide all the semiconductors and hardware compute equipment that goes into a data center and the companies building them. The term “compute shortage” became ubiquitous by May, as many of the in-demand memory and storage chipmakers saw their stocks gain over 150% in the second quarter alone. Bottlenecks in hardware led to stratospheric returns for many memory stocks this year, including SanDisk, which peaked at +883% on June 25, Western Digital (+333% on 6/18), and Seagate Technology (+297% on 6/22). The Philadelphia Semiconductor Index (SOX) experienced its best quarter ever, at +88%, and at the end of Q2, the industry group represented over 20% of the S&P 500 Index, larger than the Industrials, Materials, Utilities, Energy, and Consumer Staples sectors combined.
On the other side of the buildout, the hyperscalers, or “check writers”, whose aggregate expected capital expenditures (capex) for 2026 have eclipsed $800 billion, were under some pressure through the first half of 2026, though sentiment shifted back in their favor in July. Looking forward, capital spending by the likes of Google, Amazon, and Microsoft is expected to exceed $1 trillion in each of 2027 and 2028.
Since mid-June, semiconductor and memory stocks have re-rated dramatically lower, with some having fallen over 50% from their recent highs, and the SOX index down over 20% from its mid-June high.
Lost in the market movement, at least in our opinion, is whether enough attention is being paid to actual business fundamentals and the growing number of near-term risks. The AI theme is undoubtedly transforming the economy and markets, and inevitably there will be another stretch where the market anoints new “winners” and punishes a new set of “losers”. It’s anyone’s guess where the next group of beneficiaries will come from or when the trend will shift, but understanding what could potentially go wrong may help investors separate structural opportunities from cyclical risks.
AI Is Not One Investment Theme
While often discussed as a single investment theme, AI encompasses a range of businesses with very different economics. Often, the market indiscriminately lumps groups of stocks together, regardless of business model nuances or fundamental metrics. The reality is that there will likely be both big winners and big losers throughout the AI value chain.
Risks Worth Monitoring
Extraordinary returns often depend on extraordinary assumptions. When such explosive growth is tied to a single catalyst, even the slightest change in tenor could reverse the momentum and bring volatility.
What if the Check Writing Slows?
AI-related capex represents approximately 5% of US GDP yet was responsible for most of the incremental GDP growth year-to-date. To maintain this level of growth, companies will increasingly need to raise capital in either the debt or equity markets. Other signals that should give investors pause include:
- Hyperscaler capex approaching or exceeding operating cash flow
- Future capex estimates continue moving higher
- More debt and equity issuance
- Memory companies announcing large capacity expansion projects
Eventually, investors will stop asking “how much are you investing?” and start asking “what has your return been on this investment?” Any slowing in AI spend would likely be a net negative for the hardware companies (check cashers) and potentially net positive for hyperscalers as free cash flow improves.
What if Supply Catches Up?
Many of today’s AI hardware winners are benefiting from a great imbalance between supply and demand. Memory shortages have fueled exponential earnings growth and driven these stocks sharply higher. In fact, about 38% of all S&P 500 earnings and half of all earnings growth have come from these companies through the first half of the year.
History suggests, however, that scarcity is not permanent. As capacity expands, pricing normalizes. We’ve seen this pattern repeat itself in commodity industries. Companies with the highest operating leverage often experience the biggest swings, so if you are betting on continued outperformance of AI hardware, pay attention to balance sheets.
What if Cheaper AI is Good Enough?
Perhaps the most overlooked risk is one that emerged in June. The AI narrative that has captivated investors started showing cracks when Chinese startup Moonshot AI unveiled its Kimi K3 model, reportedly developed at a fraction of the cost of comparable U.S. models while delivering competitive, if not stronger, performance. This news sent shares of several hardware providers sharply lower as investors reassessed whether future AI progress will require increasing investment in the most advanced chips and infrastructure.
While newer, cheaper models are inevitable in any technology cycle, they also challenge an important assumption embedded in today’s AI trade: that broader adoption will require a proportional increase in infrastructure spending. If enterprises find that smaller, local, open-source, or lower-cost models meet most of their needs, demand for the most expensive hardware could fall short of today’s expectations. In that scenario, earnings estimates for AI infrastructure providers would likely come under pressure, even as AI adoption itself continues to accelerate.
The Bigger Picture
History is filled with transformative technologies that created enormous value, but not always for the obvious companies. The internet changed the world, but many of its early market leaders disappeared. Smartphones transformed computing, yet many early hand-held device companies were eventually displaced. AI may prove just as revolutionary, but that doesn’t guarantee today’s winners will remain tomorrow’s winners.
As the AI buildout evolves, investors should ask themselves whether a company’s earnings are durable or simply a product of today’s extraordinary spending cycle, and what assumptions are baked into its valuation. Markets are pricing today’s AI infrastructure darlings as though this capex cycle will continue unchanged. The more likely truth is the AI value chain will evolve as the technology matures.